Вы находитесь на странице: 1из 111

Abstract

In this thesis, an embedded crack model based on the nite element method for the
simulation of fracture processes in quasi-brittle materials and structures is presented.
The model is formulated within the framework of the strong discontinuity approach
under the assumption of small strains.
The nonlinear (softening) process is modeled with the statically and kinematically op-
timal nonsymmetric formulation. The continuum part of the material is considered to
be linear elastic, while a specic damage-based traction-separation law is included in
the model which links the traction transmitted by the discontinuity to the displacement
jump.
The detailed numerical implementation strategy of nite elements with embedded cracks
is described in this thesis and a brief overview of implementing the user subroutine in
Abaqus is presented. In the present model, a constant crack opening is introduced in
two dimensional nite elements, including constant strain triangle element and quadri-
lateral element. To cover the physical phenomenon as much as possible, numerical
treatment of closed crack is introduced in the model; an implicit/explicit integration
scheme is used to improve the robustness of the computation of softening materials and
structures; in order to deal with the crack locking problem in the standard algorithm,
crack adaptation is adopted.
The proposed model is rst validated with different test cases and then applied to real
structures; the results are analyzed and compared with the experimental data. The nu-
merical work is mainly done in Matlab and the nite element program Abaqus making
use of user subroutines.
i
ii
Acknowledgement
iii
iv
Contents
1 Introduction 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . 1
1.2 Goal of the work . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . 5
2 Basic formulation 7
2.1 Standard nite element method . . . . . . . . . . . . . . . 7
2.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Principle of virtual work . . . . . . . . . . . . . . . 12
2.2 Strong discontinuity approach . . . . . . . . . . . . . . . . 16
2.2.1 Three-eld variational formulation . . . . . . . . . 16
2.2.2 Finite element discretization . . . . . . . . . . . . . 18
2.2.3 Basic types of enhancement . . . . . . . . . . . . . 21
2.2.4 Governing equations . . . . . . . . . . . . . . . . . 24
3 Elements with embedded displacement discontinuity 27
3.1 Localization band two dimensions . . . . . . . . . . . . . . 27
3.2 Statically optimal symmetric formulation . . . . . . . . . . 29
3.2.1 Element with a localization band . . . . . . . . . . 29
3.2.2 Element with a discontinuity line . . . . . . . . . . 30
3.2.3 Constant strain triangle element . . . . . . . . . . 31
3.3 Kinematically optimal symmetric formulation . . . . . . . 32
i
ii CONTENTS
3.3.1 Element with a localization band . . . . . . . . . . 32
3.3.2 Element with a discontinuity line . . . . . . . . . . 35
3.3.3 Constant strain triangle element . . . . . . . . . . 36
3.4 Statically and kinematically optimal nonsymmetric formu-
lation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Derivation from the principle of virtual work . . . . . . . . 38
3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4 Numerical Procedure 43
4.1 Damage-based traction-separation law . . . . . . . . . . . 43
4.2 Evaluation of internal forces and element stiffness matrix . 48
4.3 Numerical treatment of closed crack . . . . . . . . . . . . 53
4.4 An implicit/explicit integration scheme for non-linear model 54
5 Computational implementation 59
5.1 Algorithm for local iterative solution procedure . . . . . . 59
5.1.1 1D iterative solution procedure . . . . . . . . . . . 60
5.1.2 2D iterative solution procedure . . . . . . . . . . . 63
5.2 Implementation of the user element in Abaqus . . . . . . . 68
5.2.1 Pre-processing and Solver . . . . . . . . . . . . . . 68
5.2.2 Post processing . . . . . . . . . . . . . . . . . . . . 73
5.3 Crack adaptation . . . . . . . . . . . . . . . . . . . . . . . 75
6 Validation and Application 79
6.1 Single element test . . . . . . . . . . . . . . . . . . . . . . 81
6.2 Three-point bending test of a notched beam . . . . . . . . 89
6.3 L-shape panel test . . . . . . . . . . . . . . . . . . . . . . . 94
7 Conclusion and outlook 99
7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Chapter 1
Introduction
1.1 Background and Motivation
Fracture mechanics is the eld of mechanics concerned with the study
of the propagation of cracks in materials. It is a method for predicting
failure of a structure containing cracks. It is an important tool in im-
proving the mechanical performance of materials and components. It is
fact that most failure in structures began with cracks. These cracks may
be caused by material defects (dislocation, impurities), discontinuities in
assembly and/or design (sharp corners, grooves, nicks, voids), harsh en-
vironments (thermal stress, corrosion) and damages in service (impact,
fatigue, unexpected loads). Most microscopic cracks are arrested inside
the material but it takes one run-away crack to destroy the whole struc-
ture. The straightforward design consideration to structural failure is
that maximum stress local to the crack may exceed the strength of the
material and thus fracture can occur.
Figure 1.1: Overview of fracture mechanics [10].
Basically fracture mechan-
ics was introduced to ana-
lyze the relationship among
stresses, cracks and frac-
ture toughness(ability of
material containing a crack
to resist fracture). Fracture
mechanics is primarily used
to prevent fracture but the
methodology is of great use
1
2 CHAPTER 1. INTRODUCTION
to understand the forma-
tion and nature of natural fractures in the structures. In recent years,
many numerical models have become very famous to calculate the driv-
ing force on a crack and to characterize the materials resistance to frac-
ture.
Particularly in case of brittle fracture which involves little or no plas-
tic deformation, usually associated with aws or defects in the material
where bulk stresses concentrate. Crack propagation in brittle materials
is very fast and it is nearly perpendicular to the direction of the applied
stress. Crack is unstable which propagates rapidly without increase in
applied stress. With increasing tensile stress, the brittle material shows
rst a linearly ascending stress-strain curve representing the elasticity of
the material, followed by a progressively descending branch indicating
the reduction of the stiffness and strength. In the case of compression,
the material shows a more or less ductile material behavior, which is
rather different from that under tension. Brittle materials such as con-
crete, rock, glasses, mortars, polymers and necking in metals are just
few examples which show nonlinear elastic and inelastic behavior under
operating conditions that involve large loads.
Traditionally smeared crack approach or the discrete crack approach are
the classical material models aiming at a macroscopic description of frac-
ture processes characterized by strain localization in brittle materials. In
the approaches of the smeared crack models, the crack is smeared out in
a continuum fashion, it is not discrete. The presence of a crack affects
the stress and the material stiffness in each material point. The soft-
ening behavior described by an appropriate stress-strain relation within
the framework of continuum mechanics is characterized by reducing the
material stiffness. According to the [24] and other researchers, smeared-
crack models for brittle fracture suffer by stress locking, i.e., by spurious
stress transfer across a widely open crack. Stress locking is mainly due
to shear stresses generated by a rotation of the principal strain axes after
the crack initiation for xed crack models with a nonzero retention fac-
tor. In [22] the source of this phenomenon was analyzed. It was shown
that the poor kinematic representation of the discontinuous displacement
eld around a macroscopic crack is the cause of spurious stress transfer.
To avoid the stress locking by improving the kinematic representation of
1.1. BACKGROUND AND MOTIVATION 3
highly localized strains, discrete approach has been developed.
The discrete models aim at the incorporation of discontinuous displace-
ment elds to properly capture the strong discontinuity kinematics of a
discrete crack [5, 2]. The softening behavior as a consequence of the
fracture process in a material is not described using a stress-strain rela-
tion but rather with tractions, depending on displacement jumps along
the discontinuity. In these models a discontinuity interface is introduced
within the solid and its behavior is governed by a discrete tractionseparation
law. Hence, these models are also known as cohesive crack models [25].
Discrete models most often make use of adaptive procedures and exploit
the concept of interface elements that allow the numerical representation
of the displacement discontinuity [3].
During the last few years the development of elements with embedded
discontinuities increased the popularity of cohesive crack models leading
to the strong discontinuity approach (SDA). The concept of elements with
embedded discontinuities is based on the idea of enriching elements with
element level additional degrees of freedom representing the displacement
jump across the discontinuity or crack. The displacement discontinuity is
introduced within the domain of an element but is rather not constrained
to the element boundaries anymore. The enriched elements have been
designed to model the displacement eld inside solid elements as the
sum of two contributions: a regular (continuous) part, and a possibly
discontinuous one which is introduced after a suitable activation criterion
has been satised.
Hence in SDA approach, without the need for adaptive procedures the
macroscopic crack path may cross a given spatial discretization in a more
or less arbitrary way. For the efcient modeling of regions with highly lo-
calized strains, this idea of incorporating strain or displacement disconti-
nuities into standard nite element interpolations has become a powerful
techniques and showed to be a valuable approach for numerical predic-
tion in failure mechanics.
Beside the concept of elements with displacement discontinuities the
more recently emerged extended nite element method (X-FEM) also
aims at the computational resolution of the strong discontinuity kine-
matics. The X-FEM is based on the partition of unity concept with the
4 CHAPTER 1. INTRODUCTION
C
0
-continuous standard basis of interpolation enriched by a discontinu-
ous part. Since the nodes serve as the support for this enhanced nite
element interpolation the X-FEM is based on a nodal enrichment. In con-
trast to the elements with embedded discontinuities the discontinuous
part of the displacement eld is represented by additional global degrees
of freedom. It is shown in [21] that the X-FEM offers a better and more
versatile kinematic description of solids exhibiting displacement discon-
tinuities. However, since within the X-FEM displacement discontinuities
are interpolated by means of additional nodal degrees of freedom with
the increasing number of equations in the course of an analysis appears as
its major drawback for the implementation in a commercial FE-program
[3].
In this thesis we focus on the computational modelling of failure of brit-
tle materials by SDA approach which is based on element enrichment
(in contrast to an nodal enrichment). According to the work by Jirasek
and Zimmermann [19], basically there are three fundemental class of
elements with embedded displacement discontinuity. They reffered this
nite element formulations as statically optimal symmetric (SOS), kine-
matically optimal symmetric (KOS) and statically and kinematically opti-
mal non-symmetric (SKON) formulations. This classication is based on
kinematic enhancement and of the stress contunity condition.
The drawback of the rst two approaches are, SOS formulation cannot
properly reect the kinematics of a completely open crack but it gives
a natural traction continuity condition, while the KOS formulation de-
scribes the kinematic aspects satisfactorily but it leads to an awkward
relationship between the stress in the bulk of the element and the trac-
tions across the discontinuity line. The development of the nonsymmet-
ric SKON formulation, which combines the optimal static and kinematic
equations and leads to an improved numerical performance [19].
The present work is based on the implementation of SDA approach in
particularly with SKON statically and kinematically optimal nonsymmetric
formulation according to the work by [19]. Implementation is done in
the Abaqus user subroutine and Matlab softwares. Abaqus is one of the
leading commercial programs for nite element analysis(FEA). Besides
a large number of built-in nonlinear material models, it also provides
the possibility to program (with FORTRAN) ones own model through
1.2. GOAL OF THE WORK 5
user-subroutines in this case user element subroutine and Matlab is a pro-
gramming environment for numerical computation, algorithm develop-
ment, data analysis and visualization. Using MATLAB, technical com-
puting problems can be solved faster than with traditional programming
languages such as C, C++ and FORTRAN.
1.2 Goal of the work
The main purpose of this thesis is to implement and improve the em-
bedded crack model for constant strain triangle element and quadrilat-
eral element based on statically and kinematically optimal nonsymmet-
ric (SKON) formulation with reference to [18] with the constant crack-
ing opening, including crack closure. Considering robustness issue im-
plcit/explicit integeration scheme [11] is used. In order to avoid the
crack locking, the standard algorithm is modied and crack adaptation
is adopted with reference to [15]. Finally compare and validate the re-
sults with expermental results in Matlab and Abaqus user subroutine soft-
wares.
1.3 Outline of the thesis
This thesis is divided into seven chapters, the remainder are organized as
the following:
Chapter 2 concerns on the variational formulation. Initially basic
mathematical expressions of standard nite element method is pre-
sented including principle of virtual work and then the technique
of enriching standard nite element interpolations by strain or dis-
placement discontinuities is proposed and derived from simple phys-
ical considerations, i.e., from the extended principle of virtual work,
from the Hu Washizu variational principle (Three eld variational
principle), using an EAS format and a B-bar format with reference
[19].
According to the work by [19], this chapter 3 concerns on the for-
mulation of three fundemental class of elements with embedded dis-
6 CHAPTER 1. INTRODUCTION
placement discontinuity. The classication is based on kinematic en-
hancement and of the stress contunity condition. Strong and week
points of individual formulations are critically evaluated, with the
special attention on brittle fracture.
Chapter 4 focuses on the numerical aspects of implementation of
damage based tracion seperation law, numerical treatment of closed
crack, evaluation of element internal force/stiffness matrix and ro-
bustness issue(computability) are presented.
In Chapter 5, the detailed implementation strategy of the standard
model is presented. The basic idea of implementing the user sub-
routine UEL in Abaqus is explained. The problem of crack locking
with the standard algorithm is summerised and to resolve the prob-
lem, the concept of crack adaptation is exlained and adopted to the
model.
In Chapter 6, the model is rst investigated with single element ex-
amples, i.e., for CST element and quadrilateral element. Then the
model is applied into structures and response of the structure with
different meshes are simulated. The results are analysed and com-
pared with the experimental data in Abaqus and Matlab softwares.
Chapter 7 summarizes the work and proposes some future research
directions.
Chapter 2
Basic formulation
To introduce a displacement discontinuity within the domain of an nite
element, it is necessary to introduce a discontinuity into the kinematic
elds of an element. In this chapter, the effect of a displacement jump is
added to nite elements as an incompatible strain mode. The displace-
ment jump is embedded in the element formulation, with the orientation
of the discontinuity determined by the local stress and the constitutive
model. By including the effect of a displacement jump only in the strain
eld of an element as an incompatible mode, the procedure can be easily
implemented in existing nite element codes.
Signicant work has been carried out into so-called embedded disconti-
nuity elements where the effect of a displacement jump is included in the
strain eld of an element. The approach followed here is based on the
work of [18, 19]. According to the [19] the exact form of the incompat-
ible strain modes is determined by equilibrium and kinematic considera-
tions. Through careful consideration of the three eld variational state-
ments with assumed strain method, a nite element with incompatible
modes was developed based on SKON formulation.
2.1 Standard nite element method
2.1.1 Introduction
The nite element method (FEM) has been developed into a key indis-
pensable technology in the modelling and simulation of various engineer-
ing systems. The FEM was rst used to solve problems of stress analysis
7
8 CHAPTER 2. BASIC FORMULATION
after then it has been applied to many other problems like thermal analy-
sis, uid ow analysis, piezoelectric analysis and many others. Basically,
the analyst seeks to determine the distribution of some eld variable like
the displacement in stress analysis, the temperature or heat ux in ther-
mal analysis, the electrical charge in electrical analysis and so on. The
FEM is a numerical method seeking an approximated solution of the dis-
tribution of eld variables in the problemdomain that is difcult to obtain
analytically.
Figure 2.1: Illustration of basic concept of FEM
[27]
Basic idea of nite element method
is that, dividing the problem do-
main into several elements, (as
shown in Figures 2.1,2.2). Known
physical laws are then applied to
each small element, each of which
usually has a very simple geome-
try. A continuous function of an
unknown eld variable is approx-
imated using for instance piece-
wise linear functions in each sub-
domain, called an element formed
by nodes. The unknowns are then the discrete values of the eld variable
at the nodes. Next, proper principles are followed to establish equations
for the elements, after which the elements are tied to one another. This
process leads to a set of linear algebraic simultaneous equations for the
entire system that can be solved easily to yield the required eld variable.
The behaviour of a phenomenon in a system depends upon the geometry
or domain of the system, the property of the material or medium, and
the boundary, initial and loading conditions. For an engineering system,
the geometry or domain can be very complex. Further, the boundary and
initial conditions can also be complicated and important consideration in
mechanics. There are displacement and force boundary conditions for
solids and structures. It is therefore, in general, very difcult to solve the
governing differential equation via analytical means. In practice, most
of the problems are solved using numerical methods. Among these, the
methods of domain discretization championed by the FEM are the most
popular, due to its practicality and versatility.
2.1. STANDARD FINITE ELEMENT METHOD 9
Depending on the property of the material, solids can be elastic, meaning
that the deformation in the solids disappears fully if it is unloaded. There
are also solids that are considered plastic, meaning that the deformation
in the solids cannot be fully recovered when it is unloaded. Elasticity
deals with solids and structures of elastic materials, and plasticity deals
with those of plastic materials.
Basic mathematical expressions of FEM:
The motion of continuum in three-dimensional space is completely de-
ned by the position vector of a material point X = [X
1
, X
2
, X
3
]
T
and
its change of position at deformation under arbitrary internal or exter-
nal inuence. This motion of the material point from the undeformed
to the deformed state is described by means of the displacement vector
u = [u
1
, u
2
, u
3
]
T
as a function of the position of the material point [27].
A basis for this description is the consideration of a body as an ensem-
ble of material points as well as the characterization of their initial and
current position by means of the position and displacement vectors. By
considering the immediate vicinity of material points one nally gets to
the concept of strains which describe the deformation of a material body.
The equation represents the strain state of continuum which is called
Green Lagrange Strain Tensor E (for complete description see; [27]) as
E =
1
2
[u +
T
u +
T
u u] (2.1)
The displacement gradient u is decomposed into a symmetric and a
skew-symmetric part, i.e., u =
sym
u +
skw
u. Based on this decom-
position, the Green Lagrange strain tensor can be written in the following
compact form
E =
sym
u +
1
2
[
T
u u] (2.2)
where
sym
u =
1
2
[u +
T
u] which is the linear function of displace-
ment gradient and the second term is the non linear term of u which
is
1
2
[
T
u u]. The strain measure of the geometrically linear theory is
dened by the symmetric part of the displacement gradient u which
is also described as the innitesimal strain tensor, is denoted with to
represent the theory of small strains.
=
sym
u (2.3)
10 CHAPTER 2. BASIC FORMULATION
The components of the symmetric strain tensor can be described by
the denitions of the symmetric part of a second order tensor and the
gradient where
ij
=
ji
. In the context of the nite element method, the
strain state is characterized by means of the strain vector.
=
1
2
[u
i,j
+u
j,i
]e
i
e
j
(2.4)
The differential operator has to be developed as a basis for the direct cal-
culation of the strain vector from the displacement vector. The desired
kinematic relation of the strain and the displacement vectors is derived
from the denition of the strain components in equation (2.4), whereby
the components of the differential operator D

represent rules for deriva-


tives.
_

11

22

33
2
12
2
23
2
13
_

_
=
_

X
1
0 0
0

X
2
0
0 0

X
3

X
2

X
1
0
0

X
3

X
2

X
3
0

X
1
_

_
_
_
u
1
u
2
u
3
_
_
; = D

u (2.5)
Kinetics describes the relation between external and internal forces act-
ing on a material body. According to the stress principle of Cauchy, a
tensor eld of stresses exists in a material body as a consequence of
the external forces. Together with the static and dynamic loads acting
throughout the volume, these stresses form the local balance of momen-
tum or the equilibrium of forces. The balance of momentum must be sat-
ised throughout the deformed conguration. According to the Cauchy
Lemma, the stress vector in the interior of the body as a function of the
outward directed normal is balanced with the stress vector of the inward
directed normal. The orientation of the surface is characterized by means
of its normal vector n. The theoremof Cauchy demands that a tensor eld
related to the vector t exists, which satises a linear mapping as follows
t(X, n) = (X) n. (2.6)
The so-postulated symmetric stress tensor is known as Cauchys stress
tensor where =
i,j
e
i
e
j
and =
T
.
The balance equation of the linear momentum describes the equilibrium
of the internal forces and the stresses. The local balance of momentum
2.1. STANDARD FINITE ELEMENT METHOD 11
can be dened in accordance with continuum mechanics, based on the
integral balance of momentum and under consideration of Cauchys the-
orem and some mathematical simplications [27] as
u = div + b (2.7)
where div symbolizes the divergence of the Cauchy stress tensor .
Constitutive equations in the classical sense presume the existence of a
relation between forces and deformation, respectively between stresses
and strains, which is exclusively local, i.e., at the considered material
point. A material law sets the relation between stresses , strains and
strain rates , which describe the velocity dependence of the stress tensor
and internal variables , which represent the dependence of the stresses
on the history (plastication or damage).
= (, , ) (2.8)
This generalized material law contains a number of material models for
the description of nonlinear material behaviour taking into account mi-
crostructural damage, residual plastic strains and time-dependent effects.
For the modelling of reversible, time-independent, elastic processes, the
stress state can be dened only based on the strain state with the stress
tensor turning into a null tensor in the undeformed conguration.
= () (2.9)
Furthermore, it is to be assumed that the material is homogeneous and
that the material properties are not dependent on the direction. Elastic-
ity means that the stress state only depends on the instantaneous strain
state and not on the stress path. The desired path-independence is only
guaranteed, if the stress tensor can be derived by differentiation of an
elastic potential function W() with respect to the strain tensor.
() =
W()

(2.10)
If the deformation is independent of the path, the corresponding material
laws are hyperelastic. Derivation of the stress tensor with respect to the
strain tensor yields the tangential modulus of elasticity, constitutive tensor
12 CHAPTER 2. BASIC FORMULATION
or material tensor D. The material tensor represents the linear mapping
of the strain tensor onto the stress tensor.
D =

; = D : (2.11)
As a consequence of the symmetry of the stress and strain tensors, the
constitutive tensor satises the symmetry properties. The constitutive
matrix can be described by means of the modulus of elasticity E and
the Poisson ratio . For the deformation analysis of two-dimensional
continua, the plane stress and the plane strain states are of interest. Plane
stress states are structural members of small depth. In the case of a
plane stress state it is assumed that the stress components
33
,
13
and

23
vanish,

33
=
13
=
23
= 0. (2.12)
where the corresponding
33
is different from zero but
13
and
23
be-
comes zero. By summarizing linearly dependent terms one obtains the
linear elastic material law of the plane stress state in the form
_
_

11

22

12
_
_
=
E
1
2
_
_
1 0
1 0
sym
1
2
_
_
_
_

11

22
2
12
_
_
. (2.13)
The plane strain state is mostly used in cases where the dimension in
one direction is very big with the loading in this direction remaining
unchanged. For the generation of the plane strain state it is assumed that
the strain components
33
,
13
and
23
vanish

33
=
13
=
23
= 0. (2.14)
Here stress components
13
and
23
become zero, the stress
33
on the
contrary, is different from zero. The constitutive equation of plane strain
turns into
_
_

11

22

12
_
_
=
E
(1 )(1 2)
_
_
1 0
1 0
sym
12
2
_
_
_
_

11

22
2
12
_
_
. (2.15)
2.1.2 Principle of virtual work
The principle of virtual work is the fundamental for the formulation of
the nite element method. The basis of the so-called weak formulation
2.1. STANDARD FINITE ELEMENT METHOD 13
of the initial boundary value problem of elastodynamics is characterized
by the description of the deformation of a material body by means of the
displacement eld and the corresponding strains (Kinematics), the force
equilibrium of stresses on a differential volume element (Kinetics), geo-
metric and static boundary conditions and the constitutive relationship
between stresses and strains (Material Law) [27].
The change from the strong form of the partial differential equation and
its boundary conditions to the weak form gives in the end the principle
of virtual work. In the weak form the geometric boundary conditions
are strongly satised, whereas the balance of momentum and the static
boundary conditions must only be satised in an integral form. This inte-
gral formulation hence allows the exact solution of the initial boundary
value problem to be replaced by an approximated solution, which satis-
es the integral but not the local form of the corresponding differential
equation.
All three components (in tensor notation equations (2.7,2.11,2.3)) to-
gether form the second order partial differential equation of linear elas-
todynamics with the displacement eld as the solution variable. In gen-
eral, the solution of this differential equation is not possible analyti-
cally. Therefore approximation methods, in particular the nite element
method, is used in order to nd an approximate solution. For the gen-
eration of the principle of virtual work, this differential equations are
scalarly multiplied by a vector-valued test function and integrated over
the volume, respectively over the neumann boundary condition. As test
function the virtual displacements u are chosen. This test function has
the special properties see [27].
The weak formulation of the balance of momentum(2.7) and of the static
boundary condition (2.6) results from the reformulation of fundamental
equations and multiplication by the test function u, integration over
the volume, respectively over the neumann boundary and addition of the
integral terms.

u( u b)dV +

u divdV

u ( n t

)dA = 0 (2.16)
For simplication of above equation, product rule for divergence is ap-
plied. Additionally, the interchangeability of the order of application of
14 CHAPTER 2. BASIC FORMULATION
variation with the symbol and differentiation with the symbol is uti-
lized for further simplication. Furthermore, the Gauss theorem for the
divergence of a rst order tensor is applied see [27] for detail formula-
tion. Finally, the principle of virtual work is derived in its usual form,
with the scalar product of the variation of the strain tensor and the stress
tensor.

u udV +

: dV =

u bdV +

u t

dA = 0 (2.17)
The rst termin above equation is described as virtual work of the inertial
forces W
dyn
, second term as internal virtual work W
int
and right side
terms as virtual work of the external forces or external virtual work W
ext
.
W
dyn
+ W
int
= W
ext
(2.18)
The denition of stresses and strains as vectors is used for the generation
of nite elements and additionally the kinematic equation (2.5) and the
constitutive law (2.11) are taken into account. At the end one obtains
the internal virtual work as function of the displacement vector u, the
constitutive matrix D and the differential operator D

.
W
int
=

: dV =

u D
T

D D

udV (2.19)
Let us consider plane nite elements, the nite element discretization
and analysis of plane continua consists of the partitioning of the struc-
ture or the domain into nite elements and the approximation of con-
tinuously distributed physical quantities (e.g. displacements) by discrete
nodal degrees of freedom and the assumption of their distribution over
the element area. This assumption is associated with the choice of shape
functions, which depend on the variables
1
and
2
(called natural coor-
dinates in isoprametric mapping) for the case of plane elements.
The topological element structure, formed by the subdivision into subdo-
mains
e
or nite elements e is called nite element mesh (Figure 2.2)
and the process of its generation is meshing or mesh generation.
Within the scope of isoparametric approximation of geometry and ele-
ment variables, the continuous position vector X can be approximated by
means of shape functions N
i
() in natural coordinates and with the help
2.1. STANDARD FINITE ELEMENT METHOD 15
Figure 2.2: Finite elemenet mesh (a) discretization with four-noded elements, (b) discretization with
three-noded elements [27].
of discrete positions of element node X
ei
X(
1
,
2
) = X()

X() =
NN

i=i
X
ei
N
i
() (2.20)
where NN generally stands for the number of element nodes. Equation
(2.20) describes the approximation of physical coordinates as function of
natural coordinates. Within the scope of the isoparametric element con-
cept we can approximate the continuous displacements, variation and
second time derivative of displacements, analogous with the approxima-
tion of the position vector in above equation.
u() u() = N() u
e
(2.21)
u() u() = N() u
e
(2.22)
u()

u() = N() u
e
(2.23)
With the approximation of continuous displacements according to equa-
tion (2.21), we get the approximation of the strain vector from equation
(2.5). Here strain vector components ought to be described in natural
coordinates. This is accomplished by applying the derivation rule to this
displacement strain relation to get approximation as
() () = D

u() = D

N()u
e
= B() u
e
. (2.24)
We get the approximation of the strain vector from equation by means of
linear mapping of the element displacement vector and the differential
operator denition called B-opertor B().
16 CHAPTER 2. BASIC FORMULATION
For plane elements, the volume element is replaced by the area element
of the middle surface dA and found to be as such with the kinematic
and kinetic assumptions and the pre-integration over the thickness co-
ordinate X
3
is depicted, see [27] for detailed description. The internal
virtual work is developed with equation (2.19) is substituted by the ap-
proximation of strain and by modifying the surface element dA and by
adjusting the integration boundaries to natural coordinates

[1, 1].
Approximation of internal virtual work comes from substitution of the
exact strain vector by the approximated strain vector in equation (2.24).


W
e
int
= u
e

1
1

1
B
T
() D B() |J()| h d
1
d
2
u
e
= u
e
K
e
u
e
(2.25)
Where J is the Jacobi matrix used for transformation of a physical co-
ordinates to natural coordinates and |J()| is Jacobi determinant. Here,
we have dened the element stiffness matrix of a plane element of an
arbitrary shape.
K
e
=
1

1
1

1
B
T
() D B() |J()| h d
1
d
2
(2.26)
The above expression ended with integral expressions dening the stiff-
ness matrix of an element. In order to conduct the nite element method,
it is necessary to solve the respective integral expressions. It is recom-
mended to apply numerical integration because the analytical integration
is very demanding or not at all possible for the entire shape diversity of
an element type. Here Gauss-Legendre integration is used to solve the
integral in equation (2.26), see [27] for detail description.
2.2 Strong discontinuity approach
2.2.1 Three-eld variational formulation
Embedded discontinuities with the nite element method is representa-
tion by powerful type of a HuWashizu variational principle. The Hu-
Washizu principle is used to develop mixed nite element methods. The
2.2. STRONG DISCONTINUITY APPROACH 17
Hu-Washizu variational principle [8, 28], which deals with three inde-
pendent elds yields the so-called three-eld variational statements. The
theorem may be written as
I(u, , ) =

1
2

T
D d +

1
2

T
D
0
d +

T
(
s
u )d

u
T

b d

S
t
u
T

t d

S
u
t
T
(u u)d = stationary
(2.27)
where
0
is initial strain (constant),

b are the prescribed body forces and

t are the prescribed tractions (surface forces). The variational is applied


to the displacement eld u, the strain eld , and the stress eld .
Except for the kinematic boundary conditions, these elds are dened
in a domain V whose boundary consists of two parts, S
u
and S
t
, with
prescribed displacements and tractions, respectively,
u = u on S
u
(2.28)
A variational theorem is stationary when the arguments (e.g. u, , )
satisfy the condition where the rst variation vanishes and for certain
restrictions on regularity, all other governing equations can be replaced
by the variational equality shown in (2.29) where the assumed elds are
completely arbitrary and mutually independent.

T
() dV +

T
(u ) dV =

V
u
T

b dV +

s
t
u
T

t dS
(2.29)
This above equality must hold for any admissible variations u, , .
In (2.29) the symbol denotes variation, () is the stress computed
from the assumed strain using the constitutive equations, is the kine-
matic operator transforming displacements into strains (the engineering-
notation counterpart of the symmetric gradient operator).
Variation of the second integral is applied, by applying the Green theorem
to the term containing u, and grouping the terms u, , together,
we can derive from (2.29) the strain displacement equations
u = (2.30)
constitutive equations
() = (2.31)
18 CHAPTER 2. BASIC FORMULATION
equilibrium equations

=

b (2.32)
and static boundary conditions
n =

t (2.33)
in which

is the adjoint operator of and n is the outward unit normal


to the boundary [19].
2.2.2 Finite element discretization
Kinematic description: Based on the regularity of the displacement eld
u(x), generally we can distinguish three types of kinematic descriptions.
The present work is the rst one incorporates strong discontinuities, i.e.,
jumps in displacements across a discontinuity curve (in two dimensions)
or discontinuity surface (in three dimensions). The strain eld (x), then
consists of a regular part, obtained by standard differentiation of the dis-
placement eld and a discontinuous part having the character of a multi-
ple of the Dirac delta distribution. The basic idea is that the displacement
eld is decomposed into a continuous part and a discontinuous part due
to the opening and sliding of a crack. This is schematically shown for the
one-dimensional case in (2.3 a).
Figure 2.3: Kinematic description with (a) strong discontinuity, (b) weak discontinuities, (c) no
discontinuities [20]
Another possible kinematic description represents the region of localized
deformation by a band of a small but nite thickness, separated from the
2.2. STRONG DISCONTINUITY APPROACH 19
remaining part of the body by two weak discontinuities, i.e., curves or
surfaces across which certain strain components have a jump but the dis-
placement eld remains continuous. This is illustrated in (2.3 b). Since
the displacement is continuous, the strain components in the plane tan-
gential to the discontinuity surface must remain continuous as well, and
only the out-of-plane components can have a jump. In physical terms, the
band between the weak discontinuities corresponds to a damage process
zone with an almost constant density of microdefects [20].
The most regular description uses a continuously differentiable displace-
ment eld, and the strain eld remains continuous. Strain localization
is manifested by high strains in a narrow band, with a continuous tran-
sition to much lower strains in the surrounding parts of the body. A
typical strain prole of this type is shown in (2.3 c). In physical terms,
this corresponds to a damage process zone with a continuously varying
concentration of defects [20].
Finite element implementation: The weak form of equations (2.30-
2.33) are solved by nite element method. The simulation of fracture
processes within the nite element (FE) context requires special provi-
sions for dealing with displacement discontinuities. The displacement
discontinuity is incorporated into standard nite element interpolations
by an additional degree of freedom. Discretised forms of the initial
value problem are expressed in matrix-vector form where the stress and
strain tensors are represented as column vectors = (
x
,
y
,
xy
)
T
and
= (
x
,
y
,
xy
)
T
in 2D case. When discretizing the problem variational
statement (2.29), representing the weak form of Eqs. (2.30-2.33) can be
utilized . We interpolate the unknown elds displacement, strains and
stresses as
u Nd +N
c
d
c
(2.34)
Bd +Ge (2.35)
Ss (2.36)
where vectors d, d
c
, e, and s are the degrees of freedom correspond-
ing to nodal displacements, enhanced displacement modes, enhanced
strain modes and stress parameters, respectively. N
c
and G are ma-
trices containing some enrichment terms for displacements and strains,
20 CHAPTER 2. BASIC FORMULATION
respectively, S is a stress interpolation matrix. N is the standard dis-
placement interpolation matrix (containing the usual shape functions),
B is the standard strain interpolation matrix (containing the derivatives
of the shape functions).
Techniques such as the EAS method [13] or the B-bar approach [9] are
used for the present interpolation which is fairly general . Substituting
approximations equations (2.34-2.36) into the variational identity (2.29)
and taking into account that (Nd) = Bd we obtain
d
T

V
B
T
(Bd +Ge)dV +e
T

V
G
T
[ (Bd +Ge) Ss]dV
+ s
T

V
S
T
(B
c
d
c
Ge)dV + d
c
T

V
B
c
T
S s dV = d
T
f
ext
+ d
c
T
f
c
(2.37)
where B
c
is a strain interpolation matrix that would correspond to the
displacement interpolation matrix N
c
(i.e., B
c
is dened by the identity
(N
c
d) = B
c
d),
f
ext
=

V
N
T

b dV +

S
t
N
T

t dS (2.38)
is the vector of (standard) external forces and
f
c
=

V
N
c
T

b dV +

S
t
N
c
T

t dS (2.39)
is the vector of nonstandard external forces. For simplicity we will as-
sume f
c
=0, that the loads are applied outside the region with enhanced
interpolation.
By taking into account the independence of variations, we obtain the
discretized equations as

V
B
T
(Bd +Ge)dV = f
ext
(2.40)

V
G
T
(Bd +Ge)

V
G
T
S dV s = 0 (2.41)

V
S
T
B
c
dV d
c

V
S
T
G dV e = 0 (2.42)
2.2. STRONG DISCONTINUITY APPROACH 21

V
B
c
T
S dV s = 0. (2.43)
The rate (incremental) formulation and differentiate (2.40-2.43) with
respect to time is applied in order to linearize the dependence of on
parameters d and e. Any incrementally linear stress-strain equations can
formally be written in the rate form for a given state as

= D D(B

d +G e) (2.44)
where D =

is the tangential stiffness matrix of the material. Substi-


tuting (2.44) into (2.40-2.43) we obtain a set of linear equations

V
_

_
B
T
DB B
T
DG 0 0
G
T
DB G
T
DG G
T
S 0
0 S
T
B 0 S
T
B
c
0 0 B
c
T
S 0
_

_
dV
_

d
e
s

d
c
_

_
=
_

f
ext
0
0
0
_

_
(2.45)
The interpolations of stress and strain can be discontinuous, so we can
select the interpolation functions such that each stress or strain param-
eter is associated with only one nite element. The same holds for the
enhanced displacement parameters. Parameters e, s and d
c
can therefore
be eliminated on the local (element) level, so that the global equations
contain only the standard displacement degrees of freedom d. From now
on, we consider Eqs. (2.40-2.43) and (2.45) written for one nite ele-
ment occupying a certain volume V
e
. Of course, the external force vector
f
ext
is then replaced by the contribution of the current element to the
internal forces f
e
int
[19].
2.2.3 Basic types of enhancement
The basic three types of nite element models with embedded discon-
tinuties known from the literature are present them as particular cases
or modications of the general formulation equations (2.40-2.43). The
detailed formulation and behavior of this three models are analysed in
chapter (3) with the nite element.
1. In the total formulation, equations (2.42) and (2.43) are linear al-
ready, and so it is better to inspect them rst. Let us assume that we
22 CHAPTER 2. BASIC FORMULATION
do not introduce any displacement enhancement terms, i.e., we en-
rich only the strain interpolation. Then all terms containing d
c
, N
c
,
or B
c
can be deleted from the formulation. Equation (2.43) then
does not exist (it has been derived by taking the variation of d
c
) and
equation (2.42) becomes

V
e
S
T
G dV e = 0 (2.46)
The element must be able to reproduce a constant stress eld ex-
actly in order to pass the generalized patch test and so the minimum
choice for the stress interpolation matrix is S=I=unit matrix. The
compatibility conditions (2.46) now reads as

V
e
G dV e = 0. (2.47)
The matrix G can always be constructed such that

V
e
G dV = 0. (2.48)
We could modify G by subtracting from each entry its mean value
over the element, if this condition of zero mean was not satised.
As the standard strain approximation = Bd always contains all
constant-strain modes, this modication does not affect the space of
functions described by the enhanced approximation = Bd +Ge.
We can therefore assume that the enhanced strain interpolation ma-
trix G satises condition (2.48). Then, equation (2.42) is satised
for any e. As the matrix multiplying s in (2.41) is now a zero matrix,
the stress parameters s completely disappear from the formulation
cite9. We nally end up with a set of nonlinear equations

V
e
B
T
(Bd +Ge)dV = f
e
int
(2.49)

V
e
G
T
(Bd +Ge)dV = 0 (2.50)
which, at a given state can be linearized into

V
e
_
B
T
DB B
T
DG
G
T
DB G
T
DG
_
dV
_

d
e
_
=
_

f
int
0
_
(2.51)
2.2. STRONG DISCONTINUITY APPROACH 23
Remark if D is symmetric then the linearized system of equations
(2.51) is also symmetric.
2. The other way is that, we can construct some suitable enhanced dis-
placement interpolation matrix N
c
and evaluate the corresponding
matrix B
c
for which (N
c
d) = B
c
d and set G = B
c
. Equations
(2.42) can be satised (independently of the choice of S) by set-
ting d
c
= e. Equation (2.43) now imposes some restrictions on the
stress parameters s but the important point is that by adding (2.43)
to (2.41), we eliminate the stress parameters and construct exactly
the same equations as before, i.e., (2.49) and (2.50). Thus both ap-
proaches lead to the same formof governing equations, with the only
difference that the matrix Geither has to satisfy condition (2.48), or
has to be constructed by applying the strain-displacement operator
to a suitable displacement enhancement cite9.
3. Modication of the governing equations was proposed by Simo and
Oliver [12] that does not follow from a variational principle, con-
ceptually it corresponds to a weak form with test functions different
from trial functions, similar to the Petrov-Galerkin method. This ap-
proach represents a compromise between the preceding two cases
because they use G = B
c
in the strain interpolation (2.35) while
replacing G
T
in (2.50) by a matrix G

that is not the transpose of


B
c
but satises the condition of zero mean (2.48). Later it will be
explained why this choice leads to an improvement of the element
performance. The resulting linearized equations

V
e
_
B
T
DB B
T
DG
G

DB G

DG
_
dV
_

d
e
_
=
_

f
int
0
_
(2.52)
are in general nonsymmetric even if the material stiffness matrix D
is symmetric.
In chapter 3 it will be shown that elements derived from the rst
above mentioned formulation nicely satisfy the traction continuity
condition but they cannot properly represent the kinematics of a
displacement or strain discontinuity. On the other hand, elements
derived from the second formulation nicely reect the discontinuity
but they lead to an awkward approximation of the traction continuity
24 CHAPTER 2. BASIC FORMULATION
condition. The above cases 1 and 2 will be referred to as the stati-
cally optimal symmetric (SOS) formulation and the kinematically op-
timal symmetric (KOS) formulation, respectively. The third approach
has the potential of representing both the kinematic and the static
aspects properly, so it will be called the statically and kinematically
optimal nonsymmetric (SKON) formulation [19].
2.2.4 Governing equations
Assumed-strain format: The basic idea is that, this is the mixed assumed
strain method within the context of physically non-linear elasticity. This
class of mixed methods includes, as a particular case, the classical method
of incompatible modes. For any of the three formulations introduced in
the preceding subsection, the governing equations can be written as
_
K
bb
K
bg
K
gb
K
gg
_
_

d
e
_
=
_

f
int
0
_
(2.53)
where
K
bb
=

V
e
B
T
DBdV, (2.54)
K
bg
=

V
e
B
T
DGdV, (2.55)
K
gb
=

V
e
G

DBdV, (2.56)
K
gg
=

V
e
G

DGdV. (2.57)
For the SOS formulation, G is a matrix with zero mean over the element
and G

= G
T
. For the KOS formulation, G = B
c
and G

= B
T
c
. Finally,
for the SKON formulation, G = B
c
and G

is a matrix with zero mean.


e = K
1
gg
K
gb

d (2.58)

f
int
= K
bb

d K
bg
K
1
gg
K
gb

d = K

d (2.59)
where
K = K
bb
K
bg
K
1
gg
K
gb
(2.60)
2.2. STRONG DISCONTINUITY APPROACH 25
is the condensed element stiffness matrix to be used in the assembly pro-
cess. This is the usual procedure exploited by EAS methods [14].
B-bar format: The symmetric formulations (SOS and KOS) can easily be
cast into the B-bar format. Substituting (2.58) into the rate form of the
strain approximation (2.35) we obtain
= B

d +G e = B

d GK
1
gg
K
gb

d = (B GK
1
gg
K
gb
)

d =

B

d (2.61)
where

B = B GK
1
gg
K
gb
(2.62)
is the so called B-bar matrix. A straightforward calculation shows that
the element stiffness matrix (2.60) can be alternatively be dened as
K =

V
e

B
T
D

BdA (2.63)
and that

V
e

B
T

dA =

V
e
B
T

dA (2.64)
so that the equivalent internal forces
f
e
int
=

V
e
B
T
dA (2.65)
that (after assembly) have to be in equilibrium with the external forces
can alternatively be computed as
f
e
int
=

t
0

V
e

B
T

dAd. (2.66)
Note that the B-bar matrix dened by (2.62) changes throughout the
calculation (because it depends on the current tangential stiffness of the
material) while the standard B-matrix reamins constant.
26 CHAPTER 2. BASIC FORMULATION
Chapter 3
Elements with embedded displacement
discontinuity
According to the kinematic relationships presented by Jirasek and Zim-
mermann in [19], the Dirac-delta distribution is included in the incom-
patible part of the strain eld. Through the use of the Dirac delta dis-
tribution, the amplitude of the incompatible mode can be interpreted
as a displacement jump. This chapter concerns on the formulation of
three fundemental class of elements with embedded displacement dis-
continuity. According to the work from [19], the three nite element
formulations are referred as statically optimal symmetric (SOS), kinemat-
ically optimal symmetric (KOS) and statically and kinematically optimal
non-symmetric (SKON) formulations. The effect of the discontinuities
are characterized by additional degrees of freedom on the element level.
This classication is based on kinematic enhancement and of the stress
continuity condition. Strong and week points of individual formulations
are critically evaluated and presented.
3.1 Localization band two dimensions
We specially focus on a plane problem, discretized by triangular or quadri-
lateral elements. Suppose that a certain element, occupying an area A
e
,
is crossed by a localization band of a constant width k. The element can
be divided into a region of localized strain L and its complement N, which
usually consists of two disjoint regions, N

and N
+
; see Figure (3.1 a) .
When dealing with a single element, we can work in a local coordinate
27
28 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
Figure 3.1: Element with a localization band [19]
system (x,y) aligned with the localization band to be captured. Axis x is
assumed to be normal to the band and axis y parallel to it. If A
L
is the
area of the localization band, we can dene an equivalent length of the
band l = A
L
/k ; see Figure (3.1 b).
It is also useful to dene an equivalent element width (characteristic
size) h = A
e
/l see Figure (3.1 c), so that A
L
/A
e
= kl/hl = /k/h. The
area of the nonlocalized region is A
N
= A
e
A
L
= (h k)l.
As it is already explained in chapter 2.2.2, the interpolations of stress and
strain can be discontinuous so we can select the interpolation functions
such that each stress or strain parameter is associated with only one nite
element. The same holds for the enhanced displacement parameters.
Parameters e, s and d
c
can therefore be eliminated on the local (element)
level, so that the global equations contain only the standard displacement
degrees of freedom d. From now on, we consider equations (2.40-2.43)
written for one nite element. Of course, the external force vector f
ext
is
then replaced by the contribution of the current element to the internal
forces f
e
int
.
Now we will introduce three basic techniques as particular cases or mod-
ications of the general formulation equations (2.40-2.43).
3.2. STATICALLY OPTIMAL SYMMETRIC FORMULATION 29
3.2 Statically optimal symmetric formulation
3.2.1 Element with a localization band
As presented in section (2.2.3):(1), we enrich only the strain interpo-
lation, i.e., we do not introduce any displacement enhancement terms,
then all terms containing d
c
, N
c
, or B
c
can be deleted from the formu-
lation. Generally standard displacement interpolation would lead to a
strain = Bd. Due to the presence of a discontinuity, a part of the strain
is relaxed by crack opening and sliding. The SOS formulation starts from
the standard strain and partially relaxes the strain component normal to
the discontinuity line and the shear component.
From linearized system of equations (2.51), D is symmetric then the lin-
earized system of equations (2.51) is also symmetric, hence the name
statitally optimal symmetric formulation. The approach of Belytschko et
al. [26] is recovered from the SOS formulation if we use a strain enrich-
ment that allows a constant jump in strains
x
and on the boundaries
of the localization band. This can be achieved by setting
G =
1
kA
e
(A
N

L
A
L

N
)P, (3.1)
where
P =
_
_
1 0
0 0
0 1
_
_
(3.2)
and
L
and
N
are the characteristic functions of the localization band
and of its complement, respectively. Matrix G is thus piecewise constant
and can be presented as
G = P
L

L
+P
N

N
, (3.3)
here
P
L
=
A
N
kA
e
P, P
N
=
A
L
kA
e
P (3.4)
The factors at
L
and
N
in (3.1) are chosen such that G satises the
condition of zero mean (2.48). Indeed,

A
e
(A
N

L
A
L

N
)dA = A
N
A
L
A
L
A
N
= 0 (3.5)
30 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
and so

A
e
GdA = 0. The common scaling factor 1/kA
e
is chosen such
that the components of e have the meaning of normal and shear strain
difference between the localized and nonlocalized region multiplied by
the width of the band. In the limit for a localization band width ap-
proaching zero, they represent the opening and sliding component of a
strong (displacement) discontinuity.
According to (2.49) , the internal forces are evaluated as
f
e
int
=

A
e
B
T
dA =

A
L
B
T

L
dA +

A
N
B
T

N
dA (3.6)
where
L
is the stress in the localization band, computed from the strain

L
= Bd +P
L
e = Bd +
A
N
kA
e
Pe (3.7)
and
N
is the stress outside the localization band, computed from the
strain

N
= Bd +P
N
e = Bd
A
L
kA
e
Pe (3.8)
The stresses must satisfy the internal equilibrium condition (2.50), which
for the matrix G dened by (3.1) leads to
A
N
kA
e

A
L
P
T

L
dA
A
L
kA
e

A
N
P
T

N
dA = 0 (3.9)
after simple arrengenet we obtain
1
A
L

A
L
P
T

L
dA =
1
A
N

A
N
P
T

N
dA. (3.10)
Note that the multiplication by P
T
selects the rst and third component
from
L
or
N
, i.e., the normal stress
x
and the shear stress
xy
. Equa-
tion (3.10) can therefore be interpreted as a weak stress continuity con-
dition, stating that
x
and
xy
averaged over the localization band must
be the same as
x
and
xy
averaged over the nonlocalized region.
3.2.2 Element with a discontinuity line
The band collapses into a curve S
L
, if we consider the limit case when
the thickness of the localization band tends to zero. From the section
3.2. STATICALLY OPTIMAL SYMMETRIC FORMULATION 31
(3.1) we know the deniton of A
N
and A
L
and from the denition ac-
cording to equation (3.1) of the enhanced strain interpolation matrix can
be rewritten as
G =
1
khl
[(h k)l
L
kl
N
]P =
_
h k
hk

L

1
h

N
_
P (3.11)
As k 0, the term
L
/k tends to Dirac distribution
L
and the G tends
to
G =
_

1
h
_
P. (3.12)
The contribution of the localization band to the internal forces vanishes
and (3.6) reads
f
e
int
=

A
e
B
T
dA, (3.13)
where we write instead of
N
because the nonlocalized region N now
extends over the entire element with the exclusion of the discontinuity
curve S
L
. For the same reason, we will drop the subscript at
N
.
In equation (3.10), the product P
T

L
has to be replaced has to be re-
placed by the cohesive tractions t, and the resulting internal equilibrium
condition
1
l

S
L
t dS =
1
A
e

A
e
P
T
dA (3.14)
means that the tractions averaged over the discontinuity line S
L
must be
equal to the rst and third component (
x
and ) of the stress averaged
over the bulk of the element. This is the weak form of the condition of
traction continuity across the localization line.
3.2.3 Constant strain triangle element
To understand further insight into the structure of the discretized equa-
tions, consider the simplest nite element (CST) with a nite localiza-
tion band [19]. Strains
L
inside the localization band are constant, and
strains
N
outside the band are also constant (but in general different
from
L
). Consequently, the stresses are also piecewise constant and the
internal forces can be evaluated as
f
e
int
= B
T
(A
L

L
+ A
N

N
). (3.15)
32 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
The internal equilibrium condition (3.10) for the CST element reads
P
T

L
= P
T

N
. (3.16)
Due to the fact that the stresses are piecewise constant, the traction con-
tinuity condition on the boundary between L and N is enforced in the
strong sense.
The structure of the governing equations is shown in the diagram in Fig-
ure (3.2 a). The top part describes to the equilibrium equations, the
middle part to the constitutive relations and the bottom part to the kine-
matic equations. Dashed arrows indicate that the source is added to the
target. For example, the strain
N
is given by the sum of of Bd and
P
N
e. Symbol denotes the constitutive operator, i.e.,
N
= (
N
) and

L
= (
L
).
In the limit case for k 0 we obtain a CST element with an embedded
displacement discontinuity line, described by the equations schematically
shown in Figure (3.2 a). Symbol s denotes the constitutive operator for
the interface. Solid arrows mean that the source must be equal to the
target. For example, the tractions t are not the sum of P
T
and s(e) but
they are equal to either of these expressions, i.e., t = P
T
= s(e).
3.3 Kinematically optimal symmetric formulation
3.3.1 Element with a localization band
In this technique as described in chapter (2.2.3):(2), KOS formulation is
clearly superior from the kinematic point of view. As stated earlier, due
to the presence of a discontinuity, a part of the strain is relaxed by crack
opening and sliding. The KOS formulation rst subtracts the contribution
of the displacement jump from the nodal displacements and only then ap-
plies the kinematic operator.
The difference between the SOS and KOS technique is that, the matrix
G either has to satisfy condition (2.48) of zero mean (SOS formulation),
or has to be constructed by applying the strain-displacement operator to
a suitable displacement enhancement; i.e., Kinematically optimal sym-
metric formulation. The approach of Lot and Shing [17, 7] is recovered
3.3. KINEMATICALLY OPTIMAL SYMMETRIC FORMULATION 33
Figure 3.2: Structure of the equations describing the SOS formulation (a) with a localization band
and (b) with a localization line [19].
from the KOS formulation if we use an enhanced displacement interpo-
lation matrix
N
c
=
_
N
c
0
0 N
c
_
(3.17)
with the special shape function
N
c
(x) = H
L
(x)
N

i=1
H
L
(x
i
)N
i
(x), (3.18)
where N is the number of nodes per element, N
i
,i = 1, 2, ....., N, are the
standard shape functions; x
i
, i = 1, 2, ..., N are the local coordinates vec-
tors; and H
L
is a ramp function assuming a constant value H
L
= 0 in
N

and a constant value H


L
= 1 in N
+
with a linear transition from
0 to 1 in L. The corrective term

H
L
(x
i
)N
i
(x) is a linear combina-
tion of the standard shape functions while the function H
L
supplies the
actual enrichment of the displacement interpolation. Inorder to make
that the standard degrees of freedom d keep the meaning of nodal dis-
placements , this correction makes sure that the value of the nonstan-
dard shape function N
c
at every node is zero. Taking into account that
34 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
N
i
(x
j
) =
ij
= Kronecker delta, we get
N
c
(x
j
) = H
L
(x
j
)
N

i=1
H
L
(x
i
)N
i
(x
j
) = H
L
(x
j
)
N

i=1
H
L
(x
i
)
ij
= 0.
(3.19)
We obtain the enhanced strain interpolation matrix by applying the strain-
displacement operator to the enhanced displacement interpolation
B
c
=
_
_
H
L,x
0
0 H
L,y
H
L,y
H
L,x
_
_

i=1
H
L
(x
i
)
_
_
N
i,x
0
0 N
i,y
N
i,y
N
i,x
_
_
. (3.20)
Here subscript after a comma indicates partial differentiation, e.g., H
L,y
=
H
L
/y. If the coordinate system is aligned with the localization band
gure (3.1a) , we have H
L,x
=
L
/k and H
L,y
= 0.
For isoparametric nite elements, each of the matrices
B
i
=
_
_
N
i,x
0
0 N
i,y
N
i,y
N
i,x
_
_
(3.21)
is a certain submatrix of the standard strain interpolation matrix
B =
_
B
1
B
2
... B
N
.

(3.22)
Therefore, the enhanced strain interpolation matrix from (3.20) can be
presented in a compact form
B
c
=
1
k

L
P BH
L
(3.23)
where P is the Boolean matrix dened by (3.2) and
H
L
=
_

_
H
L
(x
1
) 0
0 H
L
(x
1
)
. .
. .
H
L
(x
N
)
0
0
H
L
(x
N
)
_

_
. (3.24)
3.3. KINEMATICALLY OPTIMAL SYMMETRIC FORMULATION 35
Formula (3.6) for the internal forces remains valid but this time the stress

L
is computed from the strain

L
= Bd +
_
1
k
P BH
L
_
e = B(d H
L
e) +
1
k
Pe = Bd
N
+
1
k
Pe
(3.25)
and the stress
N
is computed from the strain

N
= Bd BH
L
e = B(d H
L
e) = Bd
N
, (3.26)
where
d
N
= d H
L
e. (3.27)
For the present choice G = B
c
, the internal equilibrium condition (2.50)
reads

A
L
_
1
k
P BH
L
_
T

L
dA

A
N
(BH
L
)
T

N
dA = 0 = 0, (3.28)
which can be transformed into
H
T
L
_
A
L
B
T

L
dA +

A
N
B
T

N
dA
_
=
1
k

A
L
P
T

L
dA. (3.29)
According to (3.6) the expression in parentheses on the left-hand side
is equal to the vector of internal forces. Summing the internal forces
corresponding to the nodes that are located in N
+
and adding a weighted
contribution of the nodes that happen to be in the localization band L is
equivalent to multiplication by H
T
L
. Thus, rewriting (2.56) as
H
T
L
f
e
int
=
1
A
L

A
L
P
T

L
dA. (3.30)
We obtain a condition with the following interpretation, sum of inter-
nal forces acting on the positive side of the band must be equal to the
average values of stress components
x
and
xy
in the localization band
multiplied by the length of the band.
3.3.2 Element with a discontinuity line
The internal forces are evaluated according to (3.13), in the limit for
k 0, and the weak stress continuity condition (3.30) reads
H
T
L
f
e
int
=

S
L
t dS (3.31)
36 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
This means that the cohesive tractions integrated along the discontinuity
line must be equal to the sum of internal forces acting on the positive
side of the band. In other words, if we cut the element along the disconti-
nuity line and replace the interaction between the separated parts by the
cohesive tractions, either of the separated parts of the element must sat-
isfy the conditions of force equilibrium (but not necessarily the condition
of moment equilibrium); see Figure (3.3).
Figure 3.3: Equilibrium of an element split by a discontinuity line [19].
3.3.3 Constant strain triangle element
The structure of the governing equations for a CST element based on the
KOS formulation with a nite localization band is shown in Figure (3.4 a)
and a discontinuity line is shown in the Figure (3.4 b). Symbol I denotes
the identity operator, i.e., d
N
is equal to the sum of Id d and H
L
e.
3.4 Statically and kinematically optimal nonsymmetric formulation
As it is already described in chapter (2.2.3):(3), this approach is a com-
promise between SOS and KOS formulation. A general version of the
SKON formulation for any type of parent element was proposed by Simo
and Oliver [12]. They use G = B
c
in the strain interpolation (2.35) while
replacing G
T
in (2.50) by a matrix G

that is not the transpose of B


c
but
satises the condition of zero mean (2.48). The linearized equations

V
e
_
B
T
DB B
T
DG
G

DB G

DG
_
dV
_

d
e
_
=
_

f
int
0
_
are in general nonsymmetric. In SKON formulation we use the enhanced
strain interpolation matrix B
c
given by (3.23) but in the internal equilib-
rium condition (2.50) we use the transpose of the matrix G from (3.1).
3.4. STATICALLY AND KINEMATICALLY OPTIMAL NONSYMMETRIC FORMULATION 37
Figure 3.4: Structure of the equations describing the KOS formulation (a) with a localization band
and (b) with a localization line [19].
Consequently, the stresses
L
and
N
are calculated from the strains
L
and
N
given by (3.25) and (3.26), same as for the elements due to Lot
and Shing [17, 7], but the stress continuity condition (3.10) or (3.14) is
taken from the approach due to Belytschko et al [26].
The SKON formulation of a CST with a discontinuity line leads to the
element constructed based on simple physical considerations. The nodal
displacements are decomposed into a part due to uniform strain in the
bulk of the element and a part due to crack opening. In our notation,
these parts correspond to d
N
and H
L
e, respectively. The strains = Bd
N
are related to stresses using a linear elastic law, and the crack opening
e to the cohesive tractions t using a traction seperation law.
Finally, a natural traction continuity requirement t = P
T
was imposed,
and the internal forces were calculated as f
e
int
= A
e
B
T
. The structure
of the governing equations for this element is shown in Figure (3.5 b)
and (3.5 a) presents a element with a nite localization band.
38 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
Figure 3.5: Structure of the equations describing the SKON formulation (a) with a localization band
and (b) with a localization line [19].
3.5 Derivation from the principle of virtual work
With reference [19]; The derivation of the symmetric formulations from
the PVWis as follows. When looking at a single element, we may consider
the forces f
e
int
as external ones and express the external virtual work as
W
ext
= f
e
int
T
d. (3.32)
For a model with a strong discontinuity, the internal virtual work is done
by the stresses in the continuous part on the corresponding virtual strains
and by the tractions transmitted by the discontinuity on the virtual open-
ing and sliding components of the displacement jump. Hence we can
write
W
int
=

A
e

T
dA+

S
L
t
T
e dS. (3.33)
According to the PVW, the internal virtual work must be equal to the ex-
ternal one for any virtual change of the kinematic state. By denition, the
virtual change must be kinematically admissible. If we postulate a certain
kinematic assumption with reference to kinematic equation (3.26), e.g.,
= B(d H
L
e) (3.34)
3.5. DERIVATION FROM THE PRINCIPLE OF VIRTUAL WORK 39
then the virtual change is uniquely described by d and e, which may
be regarded as independent parameters that determine the virtual strain
= B(d H
L
e) (3.35)
Substituting (3.35) into (3.33) and setting the resulting expression equal
to (3.32) we obtain

A
e

T
B(d H
L
e)dA+

S
L
t
T
e dS = f
e
int
T
d. (3.36)
This equality holds for arbitrary d and e if and only if

A
e
B
T
dA = f
e
int
(3.37)
and

A
e
H
T
L
B
T
dA +

S
L
t dS = 0. (3.38)
The rst condition is the standard formula for the evaluation of equiva-
lent nodal forces while the second one can be rewritten as

S
L
t dS = H
T
L
f
e
int
. (3.39)
This is condition (3.31) which enforces (in the weak sense) equilibrium
between the tractions and the nodal forces acting on the solitary node.
It was derived from the requirement that, if the strains in the continuous
part of the element are kept constant and the displacement discontinu-
ity is subjected to a virtual change, the virtual work done by the nodal
forces must be equal to the virtual work done by the tractions across the
discontinuity. So a derivation based on the PVW that starts from a nat-
ural kinematic assumption (3.34) automatically leads to static condition
(3.39) that is work-conjugate with the kinematic one but differs from the
natural condition of traction continuity.
In an similar manner, it is possible to proceed from a static assumption to
the work-conjugate kinematic condition derived fromthe complementary
PVW, which states that the work done by virtual nodal forces on the
actual displacements,
W

ext
= f
e
int
T
d (3.40)
40 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
must be equal to the work done by virtual stresses on the actual strains
and by virtual tractions on the actual displacement discontinuity,
W

int
=

A
e

T
dA+

S
L
t
T
e dS (3.41)
for an arbitrary virtual state that is statically admissible. The postulated
static assumptions consist of the standard relation between stresses and
nodal forces,

A
e
B
T
dA = f
e
int
(3.42)
and a condition that links the stresses and tractions across the disconti-
nuity. It is natural to require that the tractions be equal to the projected
stress components,
t = P
T
. (3.43)
Relations similar to (3.42) and (3.43) must also hold for any virtual
change of the static state. We might therefore regard the virtual stress
as independent and express the virtual nodal forces and virtual trac-
tions as f
e
int
=

A
e
B
T
dA and t = P
T
, respectively. Substituting
this into (3.40) and (3.41) and applying the complementary PVW we
obtain

A
e

T
dA+

S
L

T
P e dS =

A
e

T
B dA d. (3.44)
The virtual stress eld is not completely arbitrary but has to be self-
equilibrated inside A
e
. This condition holds in particular for constant
stress elds. In this case, we might take out of the integrals, and by
standard arguments we arrive at

A
e
dA+

S
L
P e dS =

A
e
B d dA. (3.45)
For the special case of a CST, this may be further simplied to
= Bd
l
A
e
Pe (3.46)
which is the kinematic condition work-conjugate with the natural con-
dition of traction continuity equation (3.10); with
N
and A
L
/k l.
3.6. DISCUSSION 41
3.6 Discussion
A number of techniques enriching the standard nite element interpola-
tion by additional terms corresponding to a displacement or strain dis-
continuity have been presented within a unied framework and critically
evaluated. It has been shown that there exist three major classes of these
models, called here SOS, KOS and SKON. The SOS formulation cannot
properly reect the kinematics of a completely open crack but it gives
a natural traction continuity condition, while the KOS formulation de-
scribes the kinematic aspects satisfactorily but it leads to an awkward
relationship between the stress in the bulk of the element and the trac-
tions across the discontinuity line [19].
These properties of the symmetric formulations were the driving force
behind the development of the nonsymmetric SKON formulation, which
combines the optimal static and kinematic equations and leads to an im-
proved numerical performance. This formulation deals with a very natu-
ral traction continuity condition and is capable of properly representing
complete separation at late stages of the fracturing process, without any
locking effects (spurious stress transfer). The price to pay is the loss of
symmetry of the tangential stiffness matrix. Results reported in the lit-
erature show that the nonsymmetric model can be used with success in
numerical simulations of localized cracking. It is also worth noting that
the SKON formulation does not require any specication of the length
of the localization band. This is an important advantage because such
length is in general not an objective quantity and its value depends on
the (partially ambiguous) rule for the positioning of the discontinuity
inside the element [19].
42 CHAPTER 3. ELEMENTS WITH EMBEDDED DISPLACEMENT DISCONTINUITY
Chapter 4
Numerical Procedure
In this chapter, numerical aspects of the implementation are presented.
This chapter complements the practical implementation of a specic model
with a strong (displacement) discontinuity embedded in a nite element
based on the statically and kinematically optimal non-symmetric formu-
lation. The constitutive description of a damaging interface is presented.
Algorithms for the evaluation of internal forces and stiffness matrix of
the element have been presented and also extended to special cases such
as a closed crack.
4.1 Damage-based traction-separation law
The type of the discontinuity affects the choice of an appropriate constitu-
tive model suitable for implementation in an element with an embedded
discontinuity. It is sufcient to postulate a continuum stress-strain law
for models incorporating weak (strain) discontinuities , but the models
with strong discontinuities require, a traction-separation law governing
the behavior of the discontinuity (crack or plastic slip surface) in addition
to a stress-strain law for the bulk material [19]. In the present study we
will focus on the strong discontinuity model. It is, therefore, necessary
to present a law that links the traction transmitted by the discontinuity
to the displacement jump. Based on the thermodynamic approach this
models can be conveniently derived , starting from an expression for the
density of Helmholtz free energy. For example, the free energy density
per unit volume for the isotropic continuum damage model with a single
43
44 CHAPTER 4. NUMERICAL PROCEDURE
scalar damage parameter is given by
(, ) =
1
2
(1 ) : D : (4.1)
where is the strain tensor, D is the elastic stiffness tensor, and is the
damage parameter growing from zero (virgin material) to one (complete
loss of integrity). For simplicity, we neglect the effect of temperature,
assuming that the process is isothermal. The potential (4.1) describes
only the energy stored in the elastic deformation of the bulk material
between microdefects such as cracks or voids.
We have to start from an expression for free energy per unit area, when
constructing a traction-separation law. The strain tensor is replaced by
a vector e characterizing the displacement jump(separation). This vec-
tor describes only the inelastic part of deformation (it is identically zero
before the cracking). The strain tensor corresponds to the sum of the
elastic and inelastic deformation. Before a crack is initiated, it does not
contribute to the deformation, and its initial stiffness has to be consid-
ered as innite. The surface density of free energy can be expressed as
(e, ) =
1
2
e

D e (4.2)
as a function of the separation vector, e, and a new internal variable
which is called the compliance parameter and varies from zero to inn-
ity. In the continuum model dened by equation (4.1), the compliance
parameter would correspond to /(1 ). Symbol

D in equation (4.2)
denotes a second-order tensor describing the stiffness of the discontinu-
ity (crack) at an intermediate reference state when = 1. This state sets
the scale for and its choice does not affect the response of the model
(in the sense that after proper rescaling of the compliance parameter the
same response is obtained with any choice of the reference state).
For any possible process, the model must satisfy the dissipation inequality

D t e

0 (4.3)
where

D is the dissipation rate (per unit area), and t is the traction
transmitted by the crack. In the absence of dissipative (viscous) stresses,
standard thermodynamic arguments [16, 6] lead to the state equations
t =

e
=
1

D e (4.4)
4.1. DAMAGE-BASED TRACTION-SEPARATION LAW 45
=

=
1
2
2
e

D e (4.5)
where is the dissipative thermodynamic force associated with . The
dissipation inequality (4.3) now reads

D 0 (4.6)
The thermodynamic force dened by Equation (4.5) is always non-
negative if the reference stiffness

D is assumed to be positive denite.
The rate of must not be negative according to condition (4.6), i.e., that
can only increase or remain constant but can never decrease. Symmetry
arguments lead to the condition that, in local co-ordinates for which the
rst axis is aligned with the crack normal n, the reference stiffness must
be represented by a diagonl matrix

D =
_
_
D
nn
0 0
0 D
ss
0
0 0 D
ss
_
_
(4.7)
because, e.g., normal opening of the crack should not generate shear trac-
tions on its faces, and sliding in a given tangential direction should not
produce shear tractions in the perpendicular direction (unless the mate-
rial is anisotropic). In view of Equation (4.7), the traction-separation law
(4.4) can be written as
t
n
=
D
nn

e
n
(4.8)
t
s
=
D
ss

e
s
(4.9)
where t
n
= t n is the normal traction transmitted by the crack, t
s
= tt
n
is the tangential traction, e
n
= e n is the normal component of the
seperation vector (crack opening) and e
s
= e e
n
n is the tangential
component of the seperation vector (crack sliding).
It is necessary to postulate an evolution law for the compliance param-
eter . Using the formalism of generalized standard materials [16] , we
assume the existence of a (dual) dissipation potential

(, ) such that
=

(, )

(4.10)
46 CHAPTER 4. NUMERICAL PROCEDURE
If the potential is nonnegative, equal to zero for = 0, and convex with
respect to the thermodynamic force (for any admissible value of the
internal variable ), the dissipation is guaranteed to be non-negative

D =

(4.11)
Similar to continuum damage models, let us dene a loading function
f

(, ) such that tha inequality f

< 0 characterizes the elastic domain.


If f

< 0 the deformation process is reversible (elastic), i.e., the dissipa-


tion rate must be zero. If f

= 0 damage grows which is accompanied by


energy dissipation.
For rate-independent models the damage increment is assumed to be in-
stantaneous and states for which f

> 0 can never be reached. All this is


reected by a dissipation potential dened as the indicator function [6]
of the elastic domain

F = {(, ) |f

(, ) 0}. The indicator function


is equal to zero in

F and equal to innity outside

F. Its gradient with
respect to that appears in equation (3.10) must be interpreted as a
subdifferential [6].
The resulting evolution law is described by
=

(, )

(4.12)
along with the Kuhn-Tucker conditions,

0, f

0,

f

= 0 (4.13)
from which we can derive the consistency condition,

= 0 (4.14)
Equations (4.13) and (4.14) are formally the same as for standard plas-
ticity, where f

would be the yield function and would be the plastic


multiplier. The response of the model is fully determined by specifying
the loading function, f

. Inspecting the state law (3.5) we observe that


the force is related to e

D e , which can be interpreted as a scalar
measure of the displacement jump across the discontinuity. The state law
can be rewritten as
=
D
nn
2
2
_
e
2
n
+
D
nn
D
ss
e
2
s
_
=
D
nn
e
2
2
2
(4.15)
4.1. DAMAGE-BASED TRACTION-SEPARATION LAW 47
where e
s
= e
s
is the Euclidean norm of the tangential component of
the separation vector (crack sliding), and
e =
_
e
2
n
+
D
nn
D
ss
e
2
s
=

e

D e
D
nn
(4.16)
can be called the equivalent separation (in analogy to the equivalent
strain in continuum damage mechanics).
Let us assume that the traction-separation curve for fracture under pure
Mode I has been identied from experiments and described by an explicit
relation t
n
= g(e
n
). This relation should be reproduced by the constitu-
tive law (4.8), which means that the compliance parameter under mono-
tonically increasing crack opening must satisfy the relation = F(e
n
)
where
F(e
n
) = D
nn
e
n
/g(e
n
) (4.17)
For decreasing functions g(e
n
) that characterize softening, F(e
n
) is always
an increasing function. This is closely related to the fact that the compli-
ance parameter (inversely proportional to the slope of the line connect-
ing the current point on the traction-separation curve with the origin) is
monotonically increasing as long as the crack opening e
n
keeps growing.
If e
n
drops below its maximum previously reached value, unloading takes
place and the compliance parameter is temporarily frozen.
As an example, consider the exponential traction-separation law, (pure
Mode I)
t
n
= g(e
n
) f
t
exp
_

e
n
e
u
_
(4.18)
where f
t
is the tensile strength and e
u
= G
f
/f
t
, G
f
being the Mode-I
fracture energy.
In 2D case, consider the exponential traction-separation lawwith isotropic
damage
t
e
= g( e)
_
t
n
t
s
_
=
f
t
e
exp
_

e
e
u
__
e
n
e
s
_
(4.19)
Denoting the equivalent traction by

t = f
t
exp
_

e
e
u
_
and where e equiv-
alent seperation, t
n
and t
s
are the normal and tangential components of
traction vector, e
n
and e
s
are the normal and tangential components of
seperation vector.
48 CHAPTER 4. NUMERICAL PROCEDURE
4.2 Evaluation of internal forces and element stiffness matrix
In an incremental-iterative analysis of a structure discretized by nite ele-
ments with embedded discontinuities, the nodal displacements are com-
puted iteratively from the global equilibrium equations, and the main
tasks on the level of one nite element are to evaluate the internal forces
and the tangent stiffness matrix for a given increment of nodal displace-
ments.
As already presented in the introduction, incorporating displacement dis-
continuities into the nite element interpolation can be substantially im-
proved by the kinematic representation of highly localized fracture . It
is shown there that the optimal performance is achieved if the kinematic
and static equations are constructed independently, based on their phys-
ical meaning. This so-called statically and kinematically optimal nonsym-
metric formulation (SKON) shall be used here.
For simplicity we assume that the bulk material surrounding the discon-
tinuity remains linear elastic and that the crack initiation is controled by
the Rankine criterion of maximum principal stress and for easy under-
standing we describe for CST element. For constant opening, extension
to quadrilateral element is in the similar way but we need to takecare of
gauss points . The basic idea is that the displacement eld is decomposed
into a continuous part and a discontinuous part due to the opening and
sliding of a crack; see Figure (4.1(b)). The same decomposition applies
to the nodal displacements of a nite element. we represent the discon-
tinuity by additional degrees of freedom collected in a column matrix e.
The effect of crack opening and sliding is then subtracted from the nodal
displacement vector d = (u1, v1, u2, v2, u3, v3)
T
, and only the nodal dis-
placements due to the continuous deformation serve as the input for the
evaluation of strains in the bulk material, ; see Figure (4.1(c)).
This leads to the kinematic equations from (3.26) in the form (strong
point of KOS formulation)
= B(d He) (4.20)
where B is the standard strain-displacement matrix, and H is a matrix
reecting the effect of crack opening on the nodal displacements. In
the context of nite elements we make use of the engineering notation,
4.2. EVALUATION OF INTERNAL FORCES AND ELEMENT STIFFNESS MATRIX 49
Figure 4.1: CST element with an embedded displacement discontinuity: (a) global co-ordinate sys-
tem x,y and local co-ordinate system n,s aligned with the crack; (b) normal and tangen-
tial component of the displacement jump; (c) element deformation due to strain in the
bulk material, with the contribution of crack opening and sliding to nodal displacements
subtracted; (d) equilibrium between tractions across the crack and stresses in the bulk
material [18].
i.e. in 2D case we denote column matrices = (
x
,
y
,
xy
)
T
and =
(
x
,
y
,
xy
)
T
instead of second-order tensors.
Normally, the displacement jump is approximated by a suitable function,
for example a polynomial one. It is easy to show that the approximation
need not be continuous. For triangular elements with linear displacement
interpolation, the strains and stresses in the bulk are constant in each el-
ement, and so it is natural to approximate the displacement jump also
by a piecewise constant function. In each element, the jump is described
by its normal (opening) component, e
n
, and tangential (sliding) compo-
nent, e
s
gure . These additional degrees of freedom have an internal
character and can be eliminated on the element level, which means that
the global equilibrium equations are written exclusively in terms of the
50 CHAPTER 4. NUMERICAL PROCEDURE
standard unknowns nodal displacements. From Figure (4.1(b)) it is clear
that the crack-effect matrix is given by
H =
_

_
0 0
0 0
0 0
0 0
c s
s c
_

_
(4.21)
means that the discontinuity line separates node 3 from nodes 1 and 2
(in local numbering). In Equation (4.21), c = cos and s = sin, where
is the angle between the normal to the crack and the global x-axis; see
Figure (4.1(a)).
Strains in the bulk material generate certain stresses, , which are here
computed from the equations of linear elasticity,
= D (4.22)
but the constitutive law for the bulk material could be non-linear in gen-
eral. The tractions transmitted by the crack (damaging surface), t, can be
computed from the displacement jump using the traction-separation law
from the preceding section from equation (4.4). The constitutive relation
for an opening or partially closing crack,
t =
1

D e (4.23)
The stresses in the bulk and the tractions across the crack must satisfy
certain conditions that express internal equilibrium and serve as static
equations associated with the internal degrees of freedom, e. The most
natural requirement is that the traction vector be equal to the stress ten-
sor contracted with the crack normal, similar to static boundary condi-
tions. This internal equilibrium (traction continuity) condition can be
derived from equilibrium of an elementary triangle with one side on the
discontinuity line; see Figure (4.1(d)). In the engineering notation the
traction continuity condition reads from equation (3.14) (strong point of
SOS formulation)
P
T
= t (4.24)
4.2. EVALUATION OF INTERNAL FORCES AND ELEMENT STIFFNESS MATRIX 51
where
P =
_
_
c
2
cs
s
2
cs
2cs c
2
s
2
_
_
(4.25)
is a stress rotation matrix. For linear triangles, both t and are constant
in each element, and so condition (4.24) can be satised exactly.
Finally, the nodal forces are evaluated from the standard relation
f
int
=

A
e
B
T
dA = A
e
B
T
(4.26)
where A
e
is the area of the element. Substituting into the traction con-
tinuity condition (4.24) Equations (4.22) and (4.20), we obtain a useful
expression for the traction vector in terms of the kinematic variables,
t = P
T
DB(d He) = A(d He) (4.27)
where we have denoted
A = P
T
DB (4.28)
Comparing Equation (4.27) with Equation (4.23),
A(d He)
1

D e = 0 (4.29)
we obtain after simple manipulation
(

D+ AH)e Ad = 0 (4.30)
Combined with the evolution equation for the compliance parameter ,
Equation (4.30) makes it possible to express the separation vector in
terms of the nodal displacements and eliminate it from the formulation.
The tangent stiffness of an element with embedded discontinuity can be
constructed by expressing the separation rate in terms of the displace-
ment rate and substituting into the rate form of the basic equations in-
troduced above. Differentiation of equilibrium equation (4.30) leads to
the rate equation
A

d +
_
AH

_
e +
_

De

2
_
= 0 (4.31)
52 CHAPTER 4. NUMERICAL PROCEDURE
we know that =

e
e and with simple modication, we have
A

d =
__
AH +

_
+
_

De

e
__
e (4.32)
From above equation , we express M as
M =
_
AH +

_
+
_

De

e
_
(4.33)
Here D
nn
= D
ss
= D
0
and with details
M = AH +
_
D
0

0
0
D
0

_
D
0
e
n

2
D
0
e
n

2
_

_
D
0
e
n

t e
+
D
0
e
n

te
u
D
0
e
s

t e
+
D
0
e
s

te
u
_
(4.34)
where

t is equivalent traction, and the realtion between

d abd e is ex-
pessed as
A

d = M e (4.35)
For growing damage, it is possible to express the above equation as the
seperation rate as
e = M
1
A

d (4.36)
The rate of nodal forces is now obtained from the rate form of equations
(4.20) , (4.22) and (4.26) as

f
int
= A
e
B
T
DB(

d H e) (4.37)
where K
e
= A
e
B
T
DB is the elastic element stiffness matrix and from
the rate form of e (4.36) , we have

f
int
= [K
e
K
e
HM
1
A]

d (4.38)
The present tangent element stiffness matrix is in the form
K
tan
= K
e
K
e
HM
1
A (4.39)
Remark: If damage does not grow (the crack is unloading), there is no
increase of damage parameter , i.e in (4.31) = 0, consecquently
M = AH +
_
D
0

0
0
D
0

_
(4.40)
The tangent stifness matrix becomes actually secant stiffness.
4.3. NUMERICAL TREATMENT OF CLOSED CRACK 53
4.3 Numerical treatment of closed crack
Unto now we have considered only cracks that are opening or partially
closing. The model can be improved by taking into account the unilateral
character of damage manifested as a stiffness recovery after a complete
crack closure. It is clear that the crack faces cannot overlap, and so the
normal component of the separation vector should never become neg-
ative. At the moment when the normal separation vanishes, the crack
faces establish contact and become capable of transmitting compressive
tractions without a further change of the normal displacement jump, e
n
.
The normal part of the traction-separation law (4.8) describes only the
case when e
n
> 0. Upon crack closure, it has to be replaced by conditions
e
n
= 0 and t
n
0. Both cases are simultaneously covered by conditions
t
n

D
nn

e
n
0, e
n
0,
_
t
n

D
nn

e
n
_
e
n
= 0 (4.41)
having again the KuhnTucker form. For a closed crack (e
n
= 0), it is
necessary to modify equation (4.9) governing the evolution of the sliding
components of the displacement jump vector. Sliding can take place even
if the crack is closed, provided that the shear traction is suffciently large
to overcome the residual cohesion of the crack by dry friction. Consider
a two-dimensional model with a line crack, for which the traction and
separation vectors have only one shear component, respectively denoted
as t
s
and e
s
. The residual cohesive resistance in shear is the shear traction
computed from the damage model (4.9).
To recovery stiffness after a complete crack closure, it is necessary to
check whether the converged separation vector has a positive normal
component. When a negative value of e
n
is detected, the crack is closed,
and a different algorithm must be applied. Most equations remain valid
but the constitutive relation (4.23) must be replaced by e
n
= 0.
Numerical treatment of a closed crack always starts from the assumption
that the crack separation vector e remains constant. This corresponds to
the sticking mode, in which the crack surfaces are in contact and do not
experience any relative motion. Setting e = (0, (e
s
)
n
)
T
and = ()
n
, we
can evaluate the trail tractions according to equation(4.27)
t = A(d
n+1
He) (4.42)
54 CHAPTER 4. NUMERICAL PROCEDURE
Here (n) and (n+1) are the step numbers and check the conditions from
reference [18]
t
n
0 (4.43)

t
s

D
ss

e
s

t
n
0 (4.44)
where . are the McAuley brackets(positive part) and is the coefcient
of friction between the cracked surfaces, which is neglected in this case.
If both conditions are satised the solution is admissible, and we can
accept e as the value e
n+1
at the end of the step. If the rst condition is
violated, the crack starts re-opening and the standard algorithm should
be applied. If the second condition is violated, the crack starts sliding and
the trial values should be corrected. Formula (4.27) makes it possible to
express the tractions in terms of the sliding relative displacement, e
s
. The
normal component of e is in the sliding mode equal to zero, and so we
have
t
n+1
= Ad
n+1
Ah
2
e
n+1
s
(4.45)
where h
2
is the second column of H matrix.
4.4 An implicit/explicit integration scheme for non-linear model
Let us consider a typical, displacement driven,non-linear solid mechanics
problem. The computability to those solid mechanics problems were ro-
bustness is an important issue. The properties of material failure models
equipped with strain softening, soft materials, etc., in terms of computa-
tional cost, robustness and accuracy are the important issues to be con-
sidered. Due to the softening behavior of the material, the most of the
integration schemes suffers from loss of computability when the global
tangent stiffness approaches negative-denite. To improve the robust-
ness (computability) of the computation of strain-softening structures,
an implicit/explict integration scheme for isotropic continuum damage
model and elasto-plastic model is introduced in [11].
Generally explicit integration schemes yield robust but expensive (in terms
of the computational cost) solving algorithms, whereas implicit integra-
tion schemes lead to accurate results, even for large time steps, but at the
4.4. AN IMPLICIT/EXPLICIT INTEGRATION SCHEME FOR NON-LINEAR MODEL 55
cost of a loss of robustness of the resulting numerical algorithmwhich, for
cases of practical interest, can also dramatically affect the corresponding
computational cost. Implicit integration schemes are generally uncondi-
tionally stable. Therefore, there is no intrinsic limitation on the length
of the time step, other than the control of the integration error, which
uses to be small, and the number of required time steps, is small when
compared with explicit algorithms [11].
In this context, non-linear solid mechanics problems involving crack for-
mation and propagation particularly discrete SDA approach, when mate-
rial failure is modeled via tractionseparation laws, which must be also
equipped with strength softening which involves computability difcul-
ties. To examine the reasons for this, let us consider the problem, dis-
cretized in time and in space in a nite element mesh and from the sec-
tion (4.2) the tangent stiffness matrix reads as shown in equation (4.39),
Negative values of the softening parameters, (for example: H and M in
this equation (4.39)), translate into loss of the robustness of the algorith-
mic problem through the following process
1. At initial stages of the analysis, the tangent stiffness matrix K
tan
is
elastic, is K
tan
= K
e
and, therefore, positive denite. So is the
global tangent stiffness matrix.
2. In subsequent stages, the tangent stiffness matrix for loading cases,
in equation (4.39) loose positive deniteness at those points where
material failure occurs and, therefore, exhibit negative eigenvalues.
3. Consequently, the general integration scheme in equation (2.52) losses
positive deniteness, and exhibits negative eigenvalues as well.
4. As material failure propagates through the solid, those local negative
eigenvalues deteriorate, via the assembling process, the condition
number of the global tangent stiffness matrix K
tan
, whose smallest
eigenvalues become progressively closer to zero.
5. Eventually, K
tan
, becomes singular and the convergence fails. In
general, there are no simple remedies for this, and the simulation
process cannot be continued beyond that point.
Remark: Through the preceding reasoning it appears that the lack of pos-
itive deniteness of the algorithmic tangent stiffness, as a consequence
56 CHAPTER 4. NUMERICAL PROCEDURE
of including strain softening in the model, is responsible for the observed
loss of robustness.
This work proposes a combination of implicit and explicit integration
schemes that exploits the advantages of both, while overcoming some of
their drawbacks. In essence, it is a combination of a standard implicit in-
tegration scheme of the stresses, , in the constitutive model in equation
(4.22) with an explicit extrapolation of the involved internal variables in
this case which depends on equivalent seperation e. This method im-
plicit/explicit algorithm, renders relevant benets when it is conveniently
exploited in computational mechanics. we now on shortened the name
as as IMPL-EX method. They can be summarized as follows:
The algorithmic tangent stiffness matrix becomes symmetric and semi-
positive denite even in those cases as the analytical one is not. This
leads to dramatic improvements of the robustness in problems where
implicit integrations result in singularity or the negative character of
the algorithmic tangent operators.
In many cases, the algorithmic tangent stiffness matrix becomes con-
stant. Therefore, in absence of sources of non-linearity other than
the constitutive model, the complete non-linear problem reduces to a
sequence of linear (at every time step) problems. The classical New-
tonRaphson procedure takes a unique iteration to converge and the
problem becomes step-linear. The effects on the computational costs
are also dramatic.
The good stability properties of the implicit integration algorithm are
inherited by the proposed IMPL-EX integration algorithm.
The order of accuracy of the IMPL-EX integration algorithm, with
respect to the size of the time step, is, at least, linear; the same as
many classical backward-Euler implicit algorithms. Nevertheless, the
absolute error is larger for the same time step length.
The method can be exploited to render robust, steplinear and com-
plex non-linear problems.
Usually, the numerical simulation of a structural made of strain-softening
material, under quasi-static and small deformation conditions, can be
carried out with displacement-controlled loading steps.
4.4. AN IMPLICIT/EXPLICIT INTEGRATION SCHEME FOR NON-LINEAR MODEL 57
At the beginning of the time interval [t
n
, t
n+1
], all the necessary data is
given: the internal force vector f
int
n
, the tangent stiffness matrix K
tan
n
and the nodal displacement vector U
n
. A stepwise increment of the load-
ing condition is applied: U
n+1
= U
n
+ U. Applying Newton-Raphson
iterative solution procedure, one can calculate the unknown state values
of the structure at t
n+1
.
Applying the essence of [11] to the present Embedded Crack Model, the
Implicit/explicit integration scheme is summarized in the following.
In a certain element (at the constitutive or local level), given loading
d
n+1
, considering the results of the last step e
n
and
n
, do the implicit
computation and obtain the variables e
n+1
and
n+1
.
Instead of returning implicit internal nodal force f
int,n+1
and element tan-
gent stiffness matrix K
tan
n+1
as in Section 4.2, they are calculated explicitly.
Based on the already known implicit values of internal parameter, which
is in the present model the maximum equivalent opening ( e
max
), of
the previous two steps:
n1
and
n
, the unknown internal parameter of
the current step is calculated directly by means of extrapolation [11] and
shown in Figure (4.2):
Figure 4.2: Internal variable extrapolation [11].

n+1
=
n

t
n+1
t
n
t
n
t
n1
(
n

n1
) (4.46)
Then, the explicit damage parameter is calculated by comparing the ex-
58 CHAPTER 4. NUMERICAL PROCEDURE
pressions of t
e
from the exponetial law and the damage law, i.e., from
equations (4.4) and (4.19) one obtains
=
D
0
e
max
f
t
exp( e
max
/e
u
)
(4.47)

n+1
=
D
0

n+1
f
t
exp(
n+1
/e
u
)
(4.48)
Unlike the tangent stiffness matrix for a loading step in (4.39), the ex-
plicit algorithmic stiffness is given in the form of secant stiffness, as if it
were an unloading step:

K
alg
n+1
= K
e
K
e
H

M
1
n+1
A (4.49)
with

M
n+1
= AH +
_
D
0

n+1
0
0
D
0

n+1
_
(4.50)
Finally, the explicit internal nodal force is obtained

f
int,n+1
=

K
alg
n+1
d
n+1
(4.51)
The explicit nodal force and stiffness are returned for the assembly of
global internal force vector and algorithmic stiffness matrix. Since the
stiffness is only dependent on
n+1
, which is constant during the step, the
global solution should converge within one single iteration.
Chapter 5
Computational implementation
In this chapter, the detailed implementation strategy of the standard
model is presented. The local Newton-Raphson solution tecnnique is
explained in detail, the problem of crack locking with the algorithm is
summerised and to resolve the problem, the concept of crack adaptation
is exlained and adopted to the model.
5.1 Algorithm for local iterative solution procedure
Suppose that the values of all variables at the end of a certain compu-
tational step number n are prescribed, and our task is to calculate their
values at the end of the subsequent step number n+1. The approxima-
tions of nodal displacements d
n+1
are supplied by the iterative solution
of the global equilibrium equations, and during each evaluation of the
element nodal forces they can be considered as given. Before crack ini-
tiation, it sufces to compute the stresses from the equations valid for
linear elasticity,

n+1
= DBd
n+1
(5.1)
and then check the crack initiation condition, formulated here as the sim-
ple Rankine criterion,
1
= f
t
, where
1
is the maximum principal stress
and f
t
is the tensile strength. The direction of maximum principal stress
at crack initiation provides the normal to the discontinuity, n, and also
determines the matrices P and H that depend on the crack orientation.
After crack initiation, relations (4.30) written at the end of step n+1
contain only the crack opening parameters e
n+1
and the compliance pa-
59
60 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
rameter
n+1
as basic unknowns. Every nite element which violates the
crack initiation condition has to satisfy the internal equilibrium (traction
continuity) condition with one side on the discontinuity line (4.24). The
detailed iterative solution procedure using Newton-Raphson algorithm,
in order to achieve the local equilibrium condition is described in both
1D and 2D cases.
5.1.1 1D iterative solution procedure
The 1D case (pure Mode-I fracture) can be investigated as a basic exam-
ple.
Assuming a 1D element with length 1.0 and the elastic modulus E. A
displacement condition d is applied. An embedded crack with opening
width e will appear, if d exceeds the elastic limit.
With a crack, having an unknown width e, the stress calculated from
the continuum is from equation (4.27)
t

= E(d e) (5.2)
and that from the assumed exponential law within the crack from equa-
tion (4.18).
t
e
= f
t
exp
_
e
e
u
_
(5.3)
The inequality between them, which is called the residual
R(e) = t

t
e
= E(d e) f
t
exp
_
e
e
u
_
(5.4)
which is a nonlinear relation with respect to e, should be eliminated at the
end of the time step t
n+1
, using the Newton-Raphson iterative solution
method:
R + R = 0 (5.5)
R =
_
E +
f
t
e
u
exp
_
e
e
u
__
e (5.6)
With the expressions above, the iterative increment of the unknown, e,
can be calculated from the following formula
_
E +
f
t
e
u
exp
_
e
(i)
n+1
e
u
__
e = R
(i)
n+1
(5.7)
5.1. ALGORITHM FOR LOCAL ITERATIVE SOLUTION PROCEDURE 61
and used for the updating
e
(i+1)
n+1
= e
(i)
n+1
+ e (5.8)
If the residual is found to be zero, i.e. |R| TOL, the solution procedure
is considered as converged. Then EXIT, and set e
n+1
e
(i+1)
n+1
.
The algorithm above does not contain damage. To take also the damage
of the material within the crack into account, one can assume an internal
parameter [0, ]. Then the stress in the crack can be expressed from
the equation (4.23)
t
e
=
D
0
e

(5.9)
with D
0
a chosen constant, whose value is not important, indicating a
reference state of the stiffness. The residual takes this form:
R(, e) = E(d e)
D
0
e

(5.10)
the differentiation becomes
R =
_
E
D
0

_
e +
_
D
0
e

2
_
(5.11)
Comparing the expressions of t
e
from the exponetial law (4.18) and the
damage law (4.23), one obtains
=
D
0
e
f
t
exp(e/e
u
)
(5.12)
=
_
D
0
f
t
exp(e/e
u
)
+
D
0
e
f
t
exp(e/e
u
)e
u
_
e (5.13)
The equation (5.11) now has the detailed form:
R =
__
E
D
0

_
+
D
0
e

2
_
D
0
f
t
exp(e/e
u
)
+
D
0
e
f
t
exp(e/e
u
)e
u
__
e
(5.14)
After some simple manipulation, the same expression as (5.6) is ob-
tained. Therefore, the identical formula for the iterative solution (5.7)
can be used.
Therefore, with the already known state values e
n
and
n
, given a dis-
placement condition d
n+1
, the algorithm for the iterative solution proce-
dure of the unknown e
n+1
and
n+1
(considering damage), can be sum-
marized as the following:
62 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
Without crack ,the element has elastic behaviour, the stress in the el-
ement is calculated and checked with the Rankine criterion,i.e a crack
is detected, if the Rankine criterion is violated, in the present work we
assume the crack is xed once it is detected.
1. If the element is already cracked in the previous step:
Crack closed check:
(a) If the element is cracked but open (not closed) in the previous
step, solve the linear equations t
s
= t
e
, with
(i)
n+1
, as if unloading,
to nd e
trial
.
t

= A(d
n+1
He
tr
) =

De
trial

n
= t
e
(5.15)
If normal component of e
trial
is negative, then the open crack
closes again, make the normal component of e
trial
as zero and
update the calculated values of e
(i+1)
n+1
with out local newton iter-
ations,
Else
Go to next step to nd the e
(i+1)
n+1
based on iterative technique.
(b) If the cracked is already closed in the previous step, nd the trial
traction from the continuum i.e, equation (4.27), to nd t
trial
.
If the normal component of t
trial
is positive, then the closed crack
opens again. Go to next step to nd e
(i+1)
n+1
based on local newton
iterations,
Else
Update the previous values with out the iterative technique, i.e.,
the crack is still closed.
Remark: Crack closed condition is checked only during the rst global
iteration, if the closed crack is detected during this iteration, then it
is closed for the entire step.
2. Loading/unloading check: assuming unloading (of the existing crack),
compare e
trial
= e
trial
with e
max
; if e
trial
e
max
, it is an unloading
step, therefore e
n+1
e
trial
,
n+1

n
, EXIT; else, it is a loading
step, the nonlinear governing equation system (t
,n+1
t
e,n+1
= 0)
have to be solved iteratively as the following.
5.1. ALGORITHM FOR LOCAL ITERATIVE SOLUTION PROCEDURE 63
3. Initialization: i = 0, e
(i)
n+1
= e
n
and
(i)
n+1
=
n
.
(a) Residual:
t
(i)
,n+1
= A(d
n+1
He
(i)
n+1
), t
(i)
e,n+1
=
_

De

_
(i)
n+1
=
_
D
0
e
n

D
0
e
s

_
(i)
n+1
(5.16)
R
(i)
n+1
(, e) = t
(i)
,n+1
t
(i)
e,n+1
(5.17)
Remark: for the singular case of the initiation of the crack as
= 0, use t
(i)
e,n+1
= f
t
simply.
(b) Check convergence: if |R
(i)
n+1
| TOL, set e
n+1
e
(i)
n+1
and

n+1

(i)
n+1
; EXIT.
(c) Solution: if not converged, solve for e using (5.7).
(d) Update: e
(i+1)
n+1
= e
(i)
n+1
+ e and calculate directly:

(i+1)
n+1
=
D
0
e
(i+1)
n+1
f
t
exp(e
(i+1)
n+1
/e
u
)
(5.18)
for the next iteration.
Remark: the internal parameter can also be calculated incre-
mentally, using (5.13), then update
(i+1)
n+1
=
(i)
n+1
+ . The test
program shows the same nal results for the problem, although
with more iterations and worse intermediate results.
4. Based on the updated values, the element stiffness matrix by using
implicit-explicit integation scheme is calculated from the equations
(4.49) and the internal force vector is calculated from the equation
(4.51) as already described in previous chapter (4.4).
5.1.2 2D iterative solution procedure
In the present model, the maximum equivalent opening in the loading
history e
max
, plays the role of an internal parameter :
e
max
=
_
e
old
if e
new
e
old
e
new
if e
new
> e
old
(5.19)
64 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
The damage parameter is determined by the internal parameter, see also
(5.12):
=
D
0
e
max
f
t
exp( e
max
/e
u
)
=
D
0

f
t
exp(/e
u
)
(5.20)
Based on the known state e
n
and
n
at time t
n
, given loading condition in
the form of nodal total displacement d
n+1
, we are to calculate the state
values for t
n+1
.
Assume without crack, the stress in the element can be simply calculated
as

n+1
= DBd
n+1
(5.21)
which is checked with the Rankine criterion,
1
f
t
. The direction of
the maximum principal stress provides the normal to the crack, and the
subsequent matrices P and H, if the Rankine criterion is violated, i.e. a
crack is detected. In the present work, the crack is assumed to be xed.
For an inelastic load step, the conditions that must be satised at the end
(t
n+1
) are:
R
n+1
= t
,n+1
t
e,n+1
= A(d
n+1
He
n+1
)
_
_
f
t
exp
_

e
e
u
_
_
e
n
e
_
f
t
exp
_

e
e
u
_
_
e
s
e
_
_
_
n+1
= 0 (5.22)
And the internal parameter
D
0
e
n+1

t
n+1
=
n+1
(5.23)

n+1

n
0 (5.24)
The nonlinear equation system (5.22) is solved iteratively. The lineariza-
tion of the iterative residual reads simply:
R
(i)
n+1
+ R = 0 (5.25)
in detail:
R =
_
_
AH
_
t
n
e
n
t
n
e
s
t
s
e
n
t
s
e
s
_
(i)
n+1
_
_
e = R
(i)
n+1
(5.26)
5.1. ALGORITHM FOR LOCAL ITERATIVE SOLUTION PROCEDURE 65
is used for the iterative increment e, where
t
n
e
n
=

t
_

e
2
n
e
u
e
2
+
e
2
s
e
3
_
t
n
e
s
=

t
_

e
s
e
n
e
u
e
2

e
n
e
s
e
3
_
t
s
e
n
=

t
_

e
n
e
s
e
u
e
2

e
s
e
n
e
3
_
t
s
e
s
=

t
_

e
2
s
e
u
e
2
+
e
2
n
e
3
_
(5.27)
Implementation Procedure:
Similar to the 1D case, given the nodal displacement condition d
n+1
, the
solution procedure for the unknown separation vector e
n+1
and the dam-
age parameter
n+1
, can be summarized as the following:
1. If the element is already cracked in the previous step:
Crack closed check:
(a) If the element is cracked but open (not closed) in the previous
step, solve the linear equations t
s
= t
e
, with
(i)
n+1
, as if unloading,
to nd e
trial
.
t

= A(d
n+1
He
trial
) =

De
trial

n
= t
e
(5.28)
If normal component of e
trial
is negative, then the open crack
closes again, make the normal component of e
trial
as zero and
nd the shear component corresponding to the intermediate state
and update the calculated values of e
(i+1)
n+1
with out local newton
iterations,
Else
Go to next step to nd the e
(i+1)
n+1
based on iterative technique.
(b) If the element is already closed in the previous step, nd the trial
traction from the continuum i.e, equation (4.27), to nd t
trial
.
If the normal component of t
trial
is positive, then the closed crack
opens again. Go to next step to nd e
(i+1)
n+1
based on local newton
iterations,
66 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
Else
Update the previous values with out the iterative technique, i.e.,
the crack is still closed.
Remark: Crack closed condition is checked only during the rst global
iteration, if the closed crack is detected during this iteration, then it
is closed for the entire step.
2. Loading/unloading check: assuming unloading (of the existing crack),
compare e
trial
= e
trial
with e
max
; if e
trial
e
max
, it is an unloading
step, therefore e
n+1
e
trial
,
n+1

n
, EXIT; else, it is a loading
step, the nonlinear governing equation system (t
,n+1
t
e,n+1
= 0)
have to be solved iteratively as the following.
3. Initialization: i = 0, e
(i)
n+1
= e
n
and
(i)
n+1
=
n
.
(a) Residual:
t
(i)
,n+1
= A(d
n+1
He
(i)
n+1
), t
(i)
e,n+1
=
_

De

_
(i)
n+1
=
_
D
0
e
n

D
0
e
s

_
(i)
n+1
(5.29)
R
(i)
n+1
(, e) = t
(i)
,n+1
t
(i)
e,n+1
(5.30)
Remark: for the singular case of the initiation of the crack (must
be Mode-I opening), as = 0, use t
(i)
e,n+1
= [f
t
, 0]
T
simply.
(b) Test: if R
(i)
n+1
TOL, set e
n+1
e
(i)
n+1
and
n+1

(i)
n+1
; EXIT.
(c) Solution: solve for e using (5.26).
Remark 1: for the singular case, considering e 0, e
n
e,
e
s
0,

t f
t
, etc. the terms in (5.27) have the initial values:
t
n
e
n
=

t/e
u
,
t
n
e
s
= 0,
t
s
e
n
= 0,
t
s
e
s
= (5.31)
For the programming, one can set an arbitrary value for
t
s
e
s
, since
in the special situation that in the following linear equation sys-
tem for e
_
a b
c d
_ _
e
n
0
_
=
_
R
n
0
_
(5.32)
only the component
t
n
e
n
, related to a, plays a role.
5.1. ALGORITHM FOR LOCAL ITERATIVE SOLUTION PROCEDURE 67
Remark 2: although the residual above is expressed also with
(damage), but the detailed components of the linearized terms,
considering
=

e
e =
_
D
0
e
n

t e
+
D
0
e
n

te
u
D
0
e
s

t e
+
D
0
e
s

te
u
_
e (5.33)
and after some transformation, are the same as those in (5.27).
See also the comparison between (5.6) and (5.14) in the 1D sam-
ple case in Section (5.1.1).
(d) Update: e
(i+1)
n+1
= e
(i)
n+1
+ e and calculate directly:

(i+1)
n+1
=
D
0
e
(i+1)
n+1
f
t
exp( e
(i+1)
n+1
/e
u
)
(5.34)
for the next iteration.
Remark 1: similar to the note an the end of Section (5.1.1), the
internal parameter could also be calculated incrementally, us-
ing (5.33), then updated as
(i+1)
n+1
=
(i)
n+1
+ .
Remark 2: the updated crack opening for the rst iteration, ob-
tained from the linearized equation, should be checked again by
comparing with the maximum equivalent separation e
max
.
4. Based on the updated values, the element stiffness matrix by using
implicit-explicit integation scheme is calculated from the equations
(4.49) and the internal force vector is calculated from the equation
(4.51) as already described in previous chapter (4.4).
Important Remark: In case of quadrilateral element with 2*2 gauss inte-
gration, the stress in the element at every gauss point is calculated and
checked with the Rankine criterion, i.e., a crack is detected, if the Rank-
ine criterion is violated, in the present work we introduce the crack at
every gauss point in the element and we assume the crack is xed once it
is detected. The traction contunityv condition has to be satised at every
gauss point in the element. The iterative solution procedure presented
above 5.1.2 has to be called for all the gauss point where the stress is
maximum than the criterion.
68 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
5.2 Implementation of the user element in Abaqus
In this section we describe the main features related to the Abaqus user
element [1] implementation of the embedded crack model (SDA) through
the user subroutine UEL. Abaqus provides users with an extensive array
of user subroutines that allow them to adapt Abaqus to their particular
analysis requirements. A nonlinear nite element is implemented in user
subroutine UEL. The interface makes it possible to dene any (propri-
etary) element of arbitrary complexity. If coded properly, user elements
can be utilized with most analysis procedures in Abaqus/Standard. Mul-
tiple user elements can be implemented in a single UEL routine and can
be utilized together.
Implementing the SDA in the commercial code Abaqus does imposes cer-
tain restrictions, but it also provides access to many of the available fea-
tures of such a code. We will focus on the subroutines directly related to
Abaqus. We will give a general overview of the pre- and post-processing
steps used in our implementation.
5.2.1 Pre-processing and Solver
With the reference to [1], generally the user subroutines in Abaqus are
written in Fortran programming language. User subroutine UEL will be
called for each element that is of a general user-dened element type,
each time, element calculations are required and must perform all of
the calculations for the element,appropriate to the current activity in the
analysis.
In this model, the global calculation are performend by Abaqus and the
local calculations i.e, stiffness matrix, internal force vector and user de-
ned variables(for example: seperation vecor e) of every element has to
be dened and calculated in the UEL.
According to the present embedded crack model (calculation of stiffness
matrix, internal force vector) has to be written as UEL subroutine. The
detailed element level iterative procedure is already presented in section
(5.1) where we need the data for e.g, maximum equivalent seperation
e, direction of the crack, gamma , e.t.c., of the every element from the
5.2. IMPLEMENTATION OF THE USER ELEMENT IN ABAQUS 69
previous load step for the present load step calculations (because this
variables are dened by the user not the Abaqus standards), so for this
purpose we have a possibility to use Abaqus dened array called solution-
dependent state variables (SVARS) which is explained in detailed below.
A user element is dened with the USER ELEMENT option. This option
must appear in the input le before the user element is invoked with the
ELEMENT option [1].
The syntax for interfacing to UEL is as follows:
USER ELEMENT, TYPE=Un, NODES=, COORDINATES=,
PROPERTIES=, I PROPERTIES=, VARIABLES=, UNSYMM
Data line(s)
ELEMENT,TYPE=Un, ELSET=UEL
Data line(s)
UEL PROPERTY, ELSET=UEL
Data line(s)
USER SUBROUTINES, (INPUT=le name)
Parameter denition:
TYPE: (User-dened) element type of the form Un, where n is a number.
NODES: Number of nodes on the element.
COORDINATES: Maximum number of coordinates at any node.
PROPERTIES: Number of oating point properties.
I PROPERTIES: Number of integer properties.
VARIABLES: Number of SDVs.
UNSYMM: Flag to indicate that the Jacobian is unsymmetric.
A data line of the form dof
1
, dof
2
where dof
1
is the rst degree of freedom
active at the node and dof
2
is the second degree of freedom active at the
node, etc., follows the USER ELEMENT option. If all nodes of the user
element have the same active degrees of freedom, no further data are
needed.
The interface to user subroutine UEL is:
70 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
SUBROUTINE UEL(RHS,AMATRX,SVARS,ENERGY,NDOFEL,NRHS,NSVARS,
PROPS,NPROPS,COORDS,MCRD,NNODE,U,DU,V,A,JTYPE,TIME,DTIME,
KSTEP,KINC,JELEM,PARAMS,NDLOAD,JDLTYP,ADLMAG,PREDEF,NPREDF,
LFLAGS,MLVARX,DDLMAG,MDLOAD,PNEWDT,JPROPS,NJPROP,PERIOD)
INCLUDE ABA-PARAM.INC
DIMENSION RHS(MLVARX,*), AMATRX(NDOFEL,NDOFEL), PROPS(*),
SVARS(NSVARS), ENERGY(8), COORDS(MCRD,NNODE), U(NDOFEL),
DU(MLVARX,*), V(NDOFEL), A(NDOFEL), TIME(2), PARAMS(*),
JDLTYP(MDLOAD,*), ADLMAG(MDLOAD,*), DDLMAG(MDLOAD,*),
PREDEF(2,NPREDF,NNODE), LFLAGS(*), JPROPS(*)
user coding to dene RHS(nodal forces), AMATRX(stiffness matrix) and SVARS
according to this model, standard nite element procedure for calculating element stiffness
matrix, nodal forces and embedded crack model presented in the above section (5.1.2)
RETURN
END (end of user subroutine)
UEL Variables: The following quantities are available in UEL:
Coordinates; displacements; incremental displacements; and, for dy-
namics, velocities and accelerations
SDVs at the start of the increment
Total and incremental values of time, temperature, and user-dened
eld variables
User element properties
Load types as well as total and incremental load magnitudes
Element type and user-dened element number
Procedure type ag and, for dynamics, integration operator values
5.2. IMPLEMENTATION OF THE USER ELEMENT IN ABAQUS 71
Current step and increment numbers
The following quantities must be dened:
Right-hand-side vector (residual nodal uxes or forces)(RHS)
Jacobian (stiffness) matrix(AMATRX)
Solution-dependent state variables(SVARS)
Usually by default user subroutine is called for three times. In the rst
iteration of an increment this user subroutines is called twice.During the
rst call the initial stiffness matrix is being formed using the congura-
tion of the model at the start of the increment.During the second call a
new stiffness, based on the updated conguration of the model, is cre-
ated. In subsequent iterations the subroutines is called only once during
which the corrections to the models conguration are calculated using
the stiffness from the end of the previous iteration.
All the available quantities and quantities that are to be dened has the
direct meaning according to the standard nite element technique except
the quantity called solution-dependent state variables in short formSDVs.
Solution-dependent state variables (SVARS): An array containing the
values of the solution-dependent state variables associated with this ele-
ment. The number of such variables has to be dened by the user. User
has to dene the meaning of these variables. This values can be used to
evolve with the solution of an analysis.Abaqus just stores the variables for
the user subroutine. For general nonlinear steps this array is passed into
UEL containing the values of these variables at the start of the current
increment. They should be updated to be the values at the end of the
increment, unless the procedure during which UEL is being called does
not require such an update.
The total number of SDVs per element is set with the VARIABLES param-
eter in the input le.
If the element is integrated numerically, VARIABLES should be set
equal to the number of integration points times the number of SDVs
per point.
72 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
Solution-dependent state variables can be output with the identiers
SDV1, SDV2, etc. SDVs for any element can be printed only to the
data (.dat), results (.l), or output database (.odb) les and plotted
as XY plots in Abaqus/Viewer.
It is clear that, user is responsible for every input variable, all the calcu-
lations, output variables and variables that has to be stored for next step
calculations related the element dened by user.
This model is imlemented and generalized for two cases:
1. Constant strain triangle element.
2. Quadrilateral element (2*2 Gauss Integration).
In the present work, variable updated in the subroutine, i.e., number
of SVARS are 1*10 for three noded element and 4*10 for four noded
element (variables for every integration point). Solution-dependent state
variables used in the present model are as follows:
SVARS(1):Crack situation:denes whether the element is cracked (1) or
not cracked (0).
SVARS(2):Direction: cos of the angle alpha from global x-axis to crack
normal, [/2, /2].
SVARS(3):Direction: sin of the angle alpha from global x-axis to crack
normal, [/2, /2].
SVARS(4):Closure:ag of crack closure (of existing crack).
SVARS(5):Crack opening vector: normal direction (e
n
).
SVARS(6):Crack opening vector: tangential direction (e
s
).
SVARS(7):Maximun equivalent crack opening in history, at (n) step ( e
n
).
SVARS(8):Gamma:damage state ().
SVARS(9):Maximun equivalent crack opening in history, at (n-1) step
( e
n1
).
SVARS(10):Number of times uel is called. It is used to know the local
iteration number.
The above dened variables are dened for one integration point, so the
respective variables has to be stored for the respective integration points.
In this work, we need the local iteration number while applying crack
closure algorithm but the iteration number is not available in UEL. The
5.2. IMPLEMENTATION OF THE USER ELEMENT IN ABAQUS 73
only subroutine in Abaqus in which the iteration number is available is
UINTER. But this routine is called after UEL. The dummy contact has to
be dened using UINTER in the model and to save KIT (iteration number)
in a common block. In this way we make KIT (iteration number) available
for UEL.
There are various ways to dene ags and counters, with which user can
calculate the iteration number. User can dene his own techniques to get
the iteration number using ags and counters saved in a common block.
The difference between number of call of UEL and number of iteration
has to be taken care.
5.2.2 Post processing
For post processing currently, Abaqus does not have capabilities for user-
element plotting because the code does not post-process the information
generated by user elements.To plot the deformed shape after an SDA
analysis, we have to use standard elements in Abaqus with very small
(negligible) stiffness (called overlay elements). The overlay elements
should have same topology as that of user elements, should share the
same nodes for retaining the same connectivity. As the overlay elements
share the same DOFs, the deformed shape can be visualized. Only the
nodal information can be overlayed by using this overlay element, but
the element information for example stresses and strains at integration
points cannot be visualized.
Abaqus internal procedures for post processing through the user element
are not applicable to any model, since the information generated by user
elements can not be processed by Abaqus. But, the post-processing of the
element output of SDA model by using another user subroutine UVARM
in Abaqus is possible.
User subroutine to generate element output (UVARM): In order to
generate the element output in Abaqus, multiple user subroutines are
needed in the analysis. The user subroutine UVARM has to be combined
with the UEL subroutine in a single le to generate the element output.
In the UEL subroutine, compute the values of the desired eld output
quantities and store in the common block array. Create a UVARM sub-
74 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
routine in which the values of common block are copied to the UVAR
array for the standard Abaqus elements.
User subroutine UVARM:
will be called at all material calculation points of elements for which
the material denition includes the specication of user-dened out-
put variables.
may be called multiple times for each material point in an increment.
will be called for each increment in a step.
allows you to dene output quantities that are functions of any of
the available integration point quantities.
We must specify the number of user-dened output variables (NUVARM),
for a given material to allocate space at each material calculation point
for each variable. The user-dened output variables are available for both
printed and results le output and are written to the output database and
restart les for contouring, printing and XY plotting in Abaqus/CAE. Any
number of user-dened output variables can be used.
Input File Usage:
USER OUTPUT VARIABLES
NUVARM
The user subroutine interface is as follows:
5.3. CRACK ADAPTATION 75
SUBROUTINE UVARM(UVAR,DIRECT,T,TIME,DTIME,CMNAME,ORNAME,
NUVARM,NOEL,NPT,LAYER,KSPT,KSTEP,KINC,NDI,NSHR,COORD,
JMAC,JMATYP,MATLAYO,LACCFLA)
INCLUDE ABA-PARAM.INC
CHARACTER*80 CMNAME,ORNAME
CHARACTER*3 FLGRAY(15)
DIMENSION UVAR(NUVARM),DIRECT(3,3),T(3,3),TIME(2)
DIMENSION ARRAY(15),JARRAY(15),JMAC(*),JMATYP(*),COORD(*)
user coding to dene UVAR
RETURN
END (end of user subroutine)
Parameter denition: The only variable we need for post processing is
UVAR array.
UVAR(NUVARM): An array containing the user-dened output variables.
These are passed in as the values at the beginning of the increment and
must be returned as the values at the end of the increment.
Limitations:
It requires to manually update the nelem variable in both UEL and
UVARM subroutines.
The element numbers for the overlaying Abaqus elements must have
a constant offset relative to their user dened counterparts.
5.3 Crack adaptation
With the foregoing equations, the model is complete and can be inserted
in a nite element code to make computations. However, in most of
the two- and three-dimensional simulations performed with the model,
76 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
the crack initially propagates correctly but after a certain crack growth,
the cracking spreads over various elements simultaneously avoiding the
correct localization as shown in Figure (5.1). Such locking seems to be
due to a bad prediction of the cracking direction in the element ahead of
the pre-existing crack [15].
In order to solve crack locking problem, for example [10] used a smeared
rotating crack model combined with a embedded crack. In that model
the initial stage of cracking is modelled by a smeared crack model, and
when the strain in the element reaches a critical value a displacement
discontinuity is introduced. Other authors [23] used a rotating crack
model to avoid locking.
To overcome this problem without introducing global algorithms (crack
tracking and exclusion zones), we just introduce a certain amount of
crack adaptability within each element [15].
Figure 5.1: Sketch of crack locking: the prediction of cracking direction in the shaded element is
wrong [15].
The rationale behind the method is that the estimation of the principal
directions in a constant strain element is specially bad at crack initia-
tion due to the high stress gradients in the crack tip zone where the new
cracked element is usually located; after the crack grows further, the es-
timation of the principal stress directions usually improves substantially.
5.3. CRACK ADAPTATION 77
Therefore, we allow the crack to adapt itself to the later variations in
principal stress direction while its opening is small.
This crack adaptation is implemented very easily by stating that while the
equivalent crack opening e is small, the crack direction is recomputed at
each step as if the crack were freshly created. After e reaches a thresh-
old value e
th
no further adaptation is allowed and the crack direction
becomes xed. Threshold values must be related to the softening proper-
ties of the material, and values of the order of 0.10.2G
f
/f
t
are usually
satisfactory. Here G
f
is the fracture energy and f
t
is the tensile strength.
This simple expedient has proved to be extremely effective to avoid crack
locking.In our approach the limited crack adaptation is introduced to cir-
cumvent the numerical deciency in predicting accurately the principal
stress directions, and has not to be taken as a material property.
78 CHAPTER 5. COMPUTATIONAL IMPLEMENTATION
Chapter 6
Validation and Application
The numerical examples in this chapter are intended to illustrate the per-
formance of the proposed model in view of the simulation of discrete
crack propagation by strong discontinuity approach. The analysis of the
single element test is done for the following cases: loading case (also
validating Matlab and Abaqus results in loading case), loading/unloading
test, comparison between CST element and quadrilateral element in both
cases and crack closure test.
After the analysis and validation by single element tests, different exam-
ples of structural benchmarks, namely the tests of single edge notched
beam and L-shape panel made of quasi-brittle materials, are used for the
validation and application of the present model.
Initially the results of numerical examples, i.e., the crack pattern and
load-displacement curve without the crack adaptation is applied for sin-
gle edge notched beam which is discretized by triangular elements where
crack locking phenomenon is clearly seen; when the crack adaptation is
introduced the results drastically change, with better traced crack path
and load-displacement curve. Finally this model is applied to the L-shape
panel which is discretized by triangular elements and quadrilateral ele-
ments and the results are compared with expermental data reported in
[4].
The computations are under control of the displacement (displacement
control method). Implementation is done in Abaqus and Matlab soft-
wares. The crack pattern and load-displacements curve for both elements
are simulated and compared.
In the following examples, crack is assumed to be xed with constant
79
80 CHAPTER 6. VALIDATION AND APPLICATION
opening. Only one crack is allowed during calculation for CST element
while for quadrilateral element one crack at every gauss point is allowed;
no crack branching or intersecting is permitted.
6.1. SINGLE ELEMENT TEST 81
6.1 Single element test
The element behavior is illustrated by means of a simple but meaning-
ful single element test. The proposed model is illustrated by means of a
simple example. An element with length l = 5mm, height h = 1mm and
thinckness t = 1mm with a prescribed crack location at centroid of trian-
gular element and at gauss point for quadrilateral element is analysed,
which is discretized with two similar CST elements and one quadrilateral
element as shown in the Figures (6.1 a-b ) which also shows the bound-
ary conditions. The position of the crack for both considered examples
are shown in Figures (6.1 c-d). The material parameter are adopted
as: Youngs modulus E = 10000 N/mm
2
, Poissons ratio = 0.2, tensile
strength f
t
= 1.4 N/mm
2
and fracture energy G
f
= 0.014 N/mm.
c
c
a)
1 2
4 3
b)
c) d)
Figure 6.1: a)CST element, b)Quadrilateral element, c)-d) crack direction and position. (c:centroid
of the element, 1-2-3-4:gauss points)
The element is subjected to tension by imposing the right top and right
bottom horizontal displacements. In the displacement-controlled com-
putation the resulting reaction forces are compared. For the two dif-
ferent cases individually, i.e., (CST element and quadrilateral element),
initially the results from Matlab and Abaqus are compared for loading
case and then the element behaviour of loading and unloading case are
tested seperately. At the last stage of this section both the elements re-
sults for different loading cases are compared and at the end of the sec-
tion the phenomenon of closed crack is tested. As shown in the Fig-
ures (6.2/6.3/6.5/6.6) the element behaves elastically until it reaches
the maximum stress. As already presented in chapter (5), the displace-
ment discontinuity is embedded, if the element violates the crack initia-
82 CHAPTER 6. VALIDATION AND APPLICATION
tion condition. Once the strong distontinuity is introduced, softening be-
haviour in the element begins which is seen in Figures (6.2/6.3/6.5/6.6)
that the load displacement curve starts declining.
In Figures (6.2/6.3/6.5/6.6), the loaddisplacement curves obtained with
the embedded model for both the cases by using Matlab and Abaqus soft-
wares are presented. When compared the curves which are obtained
from two softwares i.e, from Figures (6.2) and (6.3) which refer to CST
element case and Figures (6.5) and (6.6) for four noded element case
individually, which show complete same in both the softwares. Here we
validated the Matlab implementation with the commercial FE software
Abaqus. The perfect match between the loaddisplacement curves ob-
tained with the two softwares are presented in Figures (6.4) and (6.7).
CST element:
Quadrilateral element:
6.1. SINGLE ELEMENT TEST 83
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Matlab
CST element
Figure 6.2: Loading case: computed load-displacement curve, CST element, Matlab result .
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Abaqus
CST element
Figure 6.3: Loading case: computed load-displacement curve, CST element, Abaqus result.
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Matlab
Abaqus
CST element
Figure 6.4: Loading case:computed load-displacement curve, CST element, comparison between
Matlab and Abaqus results.
Loading and Unloading test:
In the previous chapter under the section (5.1), the loading and un-
84 CHAPTER 6. VALIDATION AND APPLICATION
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Matlab
Quadrilateral element
Figure 6.5: Loading case: computed load-displacement curve, quadrilateral element, Matlab result.
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Abaqus
Quadrilateral element
Figure 6.6: Loading case: computed load-displacement curve, quadrilateral element, Abaqus result.
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


Matlab
Abaqus
Quadrilateral element
Figure 6.7: Loading case: computed load-displacement curve, quadrilateral element, comparison
between Matlab and Abaqus results.
loading situation is clearly explained. Figures (6.8) and (6.9) show the
loading and unloading test for CST element and quadrilateral element,
6.1. SINGLE ELEMENT TEST 85
from the gures it can be observed that, the unloading step is at the pre-
scribed horizontal displacement 0.0045 and 0.007. The element is under
unloading step until the right top and right bottom horizontal reaction
forces reaches to minimum possible positive value or zero, after then the
element starts reloading. According to the present model, during un-
loading and reloading process, (i.e., form the Figures (6.8) and (6.9)
at the displacement 0.0045 and 0.007), there is no further increase in
the damamge parameter and seperation vector untill the system starts
loading, it means no more softening behaviour in the element (This can
be observed in the gures that, only after the displacement greater than
the 0.0045 and 0.007 the system starts softening again).
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


loading & unloading
CST element
Figure 6.8: Loading/Unloading case: computed load-displacement curve, CST element.
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


loading & unloading
Quadrilateral element
Figure 6.9: Loading/Unloading case: computed load-displacement curve, quadrilateral element.
86 CHAPTER 6. VALIDATION AND APPLICATION
Comparison between CST element and Quadrilateral element:
In this model, according to assumption that in case of CST element only
one crack passes through the element which give more accurate results,
so even in case of quadrilateral element initially we started with an as-
sumption that only one crack passes through the element where there is
a maximum stress. The crack is initiated at one of the gauss point which
is weakest among the four. With this assumption, when compared the
quadrilateral element results with CST element results, numercially it is
no way closer and not even acceptable result. To know the detailed be-
haviour of the model for quadilateral element case, next we allowed two
crack to pass through one element and then three cracks where gauss
points which are weaker among the four. But according to this three as-
sumption when compared to three noded element, the results are very
far and which are not even comparable. Finally we end up with the as-
sumption to introduce the crack at all the gausspoint where the he stress
in the element at every gauss point is calculated and checked with the
Rankine criterion,i.e a crack is detected, if the Rankine criterion is vio-
lated. With this assumption when compared with CST element results,
the perfect and exact match between the loaddisplacement curves is
obtained which is shown in the gure 6.10 for loading and even for un-
loading test shown in gure 6.11.
6.1. SINGLE ELEMENT TEST 87
0 0.005 0.01 0.015
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


CST element
Quadrilateral element
Figure 6.10: Loading case: computed load-displacement curve, comparison between CST element
and quadrilateral element.
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
0
0.05
0.1
0.15
0.2
Displacement [mm]
F
o
r
c
e

[
k
N
]


CST element
Quadrilateral element
loading & unloading
Figure 6.11: Loading/Unloading case: computed load-displacement curve, comparison between CST
element and quadrilateral element.
Crack closure:
The physical behaviour of the any body or any structure with micro cracks
is, as the crack in the body starts propagating many small or tiny cracks
beside the main crack pattern (lets say) gets open and after some time
they get closed. With some restrictions the same physical phenomenon
is introduced in the present model which is physically reasonable for a
boundary value problem. The crack closure concept and implementation
technique is presented in the corresponding section (4.3) and (5.1).
Figure (6.12) describes the phenomenon of closed crack from the com-
puted load-displacement curve, in which the element is in elastic range
between the points 12 (elastic stiffness), 23 is the softening behaviour
(softened stiffness), unloading of crack is between the points 31 and the
curve between 1 4 with the (arrow titled closed crack) describes the
crack closure. According to this model, if the normal componenent of
88 CHAPTER 6. VALIDATION AND APPLICATION
crack opening vector is negative, we say that the crack is closed, numeri-
cally we make normal component as zero and instead (softened stiffness)
we use (elastic stiffness) which is shown clearly in the gure 6.12. If we
observe the gure, the curve between the points 1 2 and 1 4 have the
same slope which is elastic.
1 0 1 2 3 4 5
x 10
3
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
Displacement [mm]
F
o
r
c
e

[
k
N
]


1
2
3
4
crack closed
crack closure
Figure 6.12: Closed crack: computed load-displacement curve.
6.2. THREE-POINT BENDING TEST OF A NOTCHED BEAM 89
6.2 Three-point bending test of a notched beam
The three-point-bending beamtest was analyzed using unstructured mesh.
The important aspects of interest is to check the ability of the proposed
model to trace a straight vertical crack without crack tracking. The beam
dimensions were as follows: length = 2000 mm, thickness = 100 mm,
and depth = 500 mm. A single notch 200 mm in depth and 5 mm in width
is introduced as shown in Figure (6.13). The material parameters were
taken to be the following: tensile strength f
t
= 2.5 MPa, Young modulus
E = 20 GPa, Poissons ratio = 0.15, fracture energy G
f
= 0.1 N/mm.

Figure 6.13: Three-point-bending test.
The computations were run under control of the displacement at the up-
per midpoint. The coarse unstructured which consist of 1499 linear tri-
angular elements (CST-elements) were used as shown in Figure (6.14).

Figure 6.14: Three-point-bending test: FE mesh.
A closer examination of the ability to trace the crack path is given in
Figure (6.15), which compares the crack path computed with and with-
out crack adaptation. Figure (6.15(a)) shows the crack path when crack
adaptation is prevented. The conguration corresponds to the crack lock-
ing in Figure (6.16) without crack adaptation. Initially, the crack prop-
agates correctly, but along the last half part of the crack, approximately,
90 CHAPTER 6. VALIDATION AND APPLICATION
cracked elements can be seen that lie outside the main crack part and at
the locked crack tip cracking seems to have delocalized. By contrast,
when crack adaptation is enabled (6.15(b)) the crack path is nearly con-
tinuous and generally straight. The threshold value e
th
used here is 9.0 e
4
mm.
Figure (6.16) without crack adaptation shows the load-displacement curves
computed without crack adaptation together with a reference curve com-
puted using the smeared-tip superposition method [15]. Obviously, the
response of the nite element computations is too stiff and the computa-
tion stops prematurely because of crack locking and lack of convergence.
When crack adaption is introduced (Figure (6.17) with crack adaptation)
the results drastically change: crack locking disappears and the response
of the notch beam is better captured in the tail of the curve. The notched
beam geometric properties, material properties and the reference curve
are taken from [15]. From Figure 6.17, even though the response of the
notch beam is better traced in the tail of the curve, we can see some parts
of the curve exhibit snapback and jumps, because of crack adaptation
(linear behaviour), i.e., the crack were freshly created until equivalent
crack opening reaches a threshold value.
Parametric study on the threshold value e
th
:
According to crack adaptation concept, we allow the crack to adapt itself
and is implemented by stating that while the equivalent crack opening
e is small, the crack direction is recomputed at each step as if the crack
were freshly created. It means we do not use any previous load step data.
After e reaches a threshold value e
th
no further adaptation is allowed and
the crack direction becomes xed, then we start using the previous load
step data.
6.2. THREE-POINT BENDING TEST OF A NOTCHED BEAM 91
a) b)
Figure 6.15: Three-point-bending test: Computed crack path: a) without crack adaptation and b)
with crack adaptation.
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
without crack adaptation
lock
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
with crack adaptation
Figure 6.16: Three-point-bending test: computed load-displacement curve, without crack adapta-
tion. Result compared with smeared-tip [15]
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
without crack adaptation
lock
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
with crack adaptation
Figure 6.17: Three-point-bending test: computed load-displacement curve, with crack adaptation.
Result compared with smeared-tip [15]
Generally threshold value is taken of the order of 0.10.2G
f
/f
t
(G
f
is the
fracture energy and f
t
is the tensile strength). If the high threshold value
92 CHAPTER 6. VALIDATION AND APPLICATION
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
12
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
a)
b)
Figure 6.18: Parameter study: high threshold value, three-point-bending test: a) crack path, b)
load-displacement curve.
a)
b)
0 0.2 0.4 0.6 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
smearedtip
still lock
Figure 6.19: Parameter study: small threshold value, three-point-bending test: a) crack path, b)
load-displacement curve.
for example 2.0 e
3
mm is used, large number of load steps are required
to reach the given threshold value, which results in reasonable crack path
Figure (6.18(a)), but the load-displacement curve in Figure (6.18(b))
exhibit snapback and big jumps, this corresponds to those almost verti-
cal segments of the curve which is not realistic and acceptable. If the
small threshold value for example 2.0 e
4
mm is used, it is almost similar
as standard algorithm, i.e., without crack adaptation, a little change in
crack path see Figures (6.19(a)) and (6.15(a) but the computation stops
prematurely because of crack locking and lack of convergence see Figures
(6.19(b)) and (6.16). Here we conclude that, it is very important to use
6.2. THREE-POINT BENDING TEST OF A NOTCHED BEAM 93
the appropriate threshold value.
94 CHAPTER 6. VALIDATION AND APPLICATION
6.3 L-shape panel test
The L-shaped panel has become a popular benchmark test for the vali-
dation of computational models for the numerical simulation of cracking
of plain concrete. The experimental data is from the refernece [4]. The
test setup with the geometric properties and the boundary conditions is
shown in gure (6.20(a)).
The long and the short edges of the L-shaped panel are given as 500 and
250 mm,respectively; its thickness is 100 mm. The lower horizontal edge
of the vertical leg is xed. A vertical load F,v acting uniformly across the
thickness to the direction of gravity, is applied at the lower horizontal
surface of the horizontal leg at a distance of 30 mm from the vertical
end face. Shortly before reaching the maximum load the experiment is
switched from load-control to displacement control. The material param-
eters are: Youngs modulus, Poissons ratio, the uniaxial tensile strength
and the specic fracture energy are given as E = 25850 N/mm
2
, = 0.18,
f
t
= 2.70 N/mm
2
and G
f
= 0.09 N/mm.
500
500
250 250
250
250
30
F,v
all lengths in [mm]
a)
b)
Figure 6.20: Illustration of L-shaped panel test: a) Test setup, b) scatter of observed crack paths.
The grey shaded area in Figure (6.20(b)) shows the crack path of scatter
of the experimental results from three tests on identical specimens and
in the corresponding Figure (6.22) or (6.24) depicts the relationship be-
tween the applied load F,v and the vertical displacement v at the point
of load application which is marked as red curves. The mesh data of the
6.3. L-SHAPE PANEL TEST 95
employed FE-meshes are given in Figure (6.21) consist of 1450 linear tri-
angular elements (CST-elements), whereas the FE-mesh (6.23) contains
620 bilinear quadrilateral elements (CPS4-elements). This model, which
is based on the strong discontinuity approach, is implemented for CST-
elements and quadrilateral elements, numerical results are presented for
the meshes with CST-elements and quadrilateral elements.
Apart from the FE-meshes, Figure (6.21) and (6.23) also shows the com-
puted crack paths by marking those elements with magnitude of crack
opening, which are crossed by the crack, by color shading. The com-
puted crack paths can be compared with the respective crack paths ob-
served in the tests, which is shown in Figure (6.20(b)). The area between
red color curves in gures 6.22 or 6.24 shows the scatter of the experi-
mental results, whereas the load-displacement curves in the same gures
(blue in color) refer to the numerically predicted behaviour employing
the present crack model for CST element and quadrilateral element.
According to the present model, From the section (6.1), we observed that
the, when we compare the results between CST element and quadrilat-
eral element, the perfect and exact match between the loaddisplacement
curves is obtained which is shown in the gure 6.10 for loading and for
unloading test shown in gure 6.11. But when we apply for real struc-
tures, crack pattern and load-displacement curve are quite different as
shown in Figures (6.21, 6.22) and (6.23, 6.24).
From Figure (6.21), in case of CST element case, we can see that crack
path as well as load-displacement curve shown in Figure (6.22) is very
close to realistic compared to expermental results. If we observe from
Figures (6.23, 6.24), in case of quardilateral element case, we can see
that crack path is not that realistic compared to CST element case or
even expermental results. The crack path seems to be diverting at the
last stages. We have some kind of stress locking problem. It can also be
observed that some unnecessary elements beside the main crack path get
cracked, the reason can be that we started with an assumption of constant
crack opening in the element, according to the authors knowledge.
Remark:
Regarding quadrilaterals with constant approximation of the displace-
ment jump, locking can be explained as the following: imagine two
96 CHAPTER 6. VALIDATION AND APPLICATION
Magnitude of crack opening


0
0.1
0.2
0.3
Figure 6.21: L-shaped panel test: FE-mesh (CST element) and computed crack pattern (plot of
magnitude of crack opening).
0 0.2 0.4 0.6 0.8 1
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
expermental spectrum
expermental spectrum
CST element
Figure 6.22: L-shaped panel test: computed load-displacement curve (CST element).
blocks separated by a fully formed crack and moving like rigid bodies,
(e.g., a beam under bending after full failure of the middle cross section);
if the rigid bodies rotate in opposite directions, the elements crossed by
the crack are subjected to a bending-type deformation but this cannot be
relaxed by the constant displacement jump. Compressive strains appear
in the top part of the element and tensile strains in the bottom part. Since
the element itself is elastic, the corresponding stresses are generated and
6.3. L-SHAPE PANEL TEST 97
Magnitude of crack opening


0
0.1
0.2
0.3
Figure 6.23: L-shaped panel test: FE-mesh (quadrilateral element) and computed crack pattern
(plot of magnitude of crack opening).
0 0.2 0.4 0.6 0.7 0.8
0
2
4
6
8
10
Displacement [mm]
F
o
r
c
e

[
k
N
]


unstructured mesh
expermental spectrum
expermental spectrum
Quadrilateral element
Figure 6.24: L-shaped panel test: computed load-displacement curve (quadrilateral element).
they lead to nodal forces that are equivalent to a bending moment. This
moment grows proportionally to the rotation, which is typically observed
at very late stages of structural response.
98 CHAPTER 6. VALIDATION AND APPLICATION
Chapter 7
Conclusion and outlook
7.1 Conclusion
In this thesis, a numerical model based on the nite element method for
the simulation of fracture processes in quasi-brittle structures was pre-
sented. Particularly embedded crack model was formulated within the
framework of the strong discontinuity approach for the nonlinear (soften-
ing) behavior of the quasi-brittle material was presented.
The softening behavior of quasi-brittle material was modeled with the
statically and kinematically optimal nonsymmetric (SKON) formulation
which improved numerical performance by optimal static and kinematic
equations. A damage-based traction-separation law was included in the
model, which links the traction transmitted by the discontinuity to the
displacement jump. The model was designed under the assumption of
small displacements and small strains.
In order to model physical phenomenon, numerical treatment of closed
crack was introduced in the present model which was physically reason-
able for a boundary value problem. As this is a non-linear solid mechanics
problem, to improve the robustness (computability) of the computation
of softening structures, an implicit/explict integration scheme was used.
In order to solve crack locking problem in the standard algorithm, crack
adaptation was adopted which means allowing the embedded crack in
the nite element to adapt itself to the stress eld while the crack open-
ing does not exceed a small threshold value.
In the present model, Rankine type criterion for tension was applied,
where the model was implemented for a constant crack opening with
99
100 CHAPTER 7. CONCLUSION AND OUTLOOK
one crack was allowed to pass through the constant strain triangle ele-
ment and one cracks (at every gauss point) was allowed in the quadri-
lateral element. The detailed numerical implementation strategy of -
nite elements with a localization line (embedded crack) was presented in
this thesis and a brief overview of implementing the user subroutine in
Abaqus was presented.
The proposed crack model was rst tested with one-element examples
for different test cases (loading, loading/unloading, closed crack tests) for
constant strain triangle element and quadrilateral element which are im-
plemented in standard codes (Matlab and Abaqus softwares). The com-
parison of the results indicates that both elements behave consistently.
Next the standard model was applied for a structure without and with
crack adaptation. It has been shown that the local crack adaptability is
preferable to prevent crack locking. The model was then applied into real
structures; the results are analyzed and compared with the experimental
data. The simulation shows good agreement to the reality.
In this way, the embedded crack approach turns out to be yet an effec-
tive and robust alternative to other more sophisticated methods for the
simulation of quasi-brittle damage and fracture.
7.2 Outlook
In order to improve the current model, considering the shortcomings in
the work, the following aspects are proposed to be meaningful research
in the future:
Improvement of the statically and kinematically optimal nonsymmet-
ric formulation to linear crack opening:
The present formulation is exible only for constant crack opening
which is acceptable for CST elements but it is not quite acceptable
for any other elements. If we observe from chapter (6) under the
section (6.3), in case of quardilateral element case, we can see that
crack path is not realistic compared to CST element case or even ex-
permental results. The crack path seems to be diverting at the last
stages. we have some kind of stress locking problem. It is observed
7.2. OUTLOOK 101
that some unnecessary elements beside the main crack pattern gets
cracked, the reason is that because we started with an assumption
of constant crack opening in the element. The present model has to
be developed in more physical form for the evolution of the crack
opening.
Improvement on the implementation:
The implementation can be improved by introducing the crack track-
ing algorithm for continuous crack pattern and it can be extended
from 2D case to 3D case. Much more exible numerical treatment of
closed crack concept can be adopted.
Optimization of the programs:
It is meaningful to optimize the code both the Matlab and Abaqus
user subroutine , making the code work more efciently with less
time-consuming.
Further investigation on the implementation of quadrilateral ele-
ment:
During implementation, it has been clearly mentioned in chapters
(5) and (6) that, we allowed four cracks (at every gauss point) to
pass through the quadrilateral element (it means the element is soft-
ening at every gauss points), which can be tested (improved) by al-
lowing only one crack per element but softening of stiffness at this
gauss point should be multiplied by the number of gauss point which
violated the crack initiation condition based on the current load step.
The improvement is based on the success of the above menctioned
point.
102 CHAPTER 7. CONCLUSION AND OUTLOOK
Bibliography
[1] Abaqus. User Subroutines Reference Manual. Dassault Systmes, 6.10.
[2] G. Barenblatt. The mathematical theory of equilibrium cracks in brittle fracture. Adv. Appl.
Mech, 1962.
[3] G.Hofstetter C.Feist. An embedded strong distontinuity model for cracking of plain concrete.
Comput. Meths. Appl. Mech. Eng, 2005.
[4] Hermann Lehar Christian Feist, Walter Kerber and G.Hofstetter. A COMPARATIVE STUDY
OF NUMERICAL MODELS FOR CONCRETE CRACKING. European Congress on Computa-
tional Methods in Applied Sciences and Engineering, 2004.
[5] D. Dugdale. Yielding of steel sheets containing slits. J. Mech. Phys. Solids, 1960.
[6] Maugin GA. The Thermomechanics of Plasticity and Fracture. Cambridge University Press:
Cambridge, UK, 1992.
[7] P.B. Shing H.R. Lot. Embedded representation of fracture in concrete with mixed nite
elements. Int. J. Num. Meth. Eng, 1995.
[8] H.C. Hu. On some variational principles in the theory of elasticity and plasticity. Scintia
Sinica 4, 1955.
[9] T.J.R. Hughes. Generalization of selective integration procedures to anisotropic and nonlinear
media. Int. J. Num. Meth. Eng, 1980.
[10] Jon Cartwright is a reporter for physicsworld.com and Credit: Hofmann et al. Glassy metals
are tougher than steel. http://physicsworld.com/, Feb 28, 2008.
[11] J.C. Cante J. Oliver, A.E. Huespe. An implicit/explicit integration scheme to increase com-
putability of non-linear material and contact/friction problems. Comput. Methods Appl.
Mech. Engg, 2007.
[12] J. Oliver J.C. Simo. A new approach to the analysis and simulation of strain softening in
solids. Fracture and Damage in Quasibrittle Structures, 1994.
[13] M.S. Rifai J.C. Simo. A class of mixed assumed strain methods and the method of incompatible
modes. Int. J. Num. Meth. Eng, 1990.
[14] M.S. Rifai J.C. Simo. A class of mixed assumed strain methods and the method of incompatible
modes. Int. J. Num. Meth. Eng, 1990.
103
104 BIBLIOGRAPHY
[15] Adel M.Fathy David A.Cendon J.M.Sancho, J.Planas. Three-dimensional simulation of con-
crete fracture using embedded crack elements without enforcing crack path continuity. INTER-
NATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHAN-
ICS, 2007.
[16] Chaboche J-L. Lemaitre J. Mechanics of Solid Materials. Cambridge University Press: Cam-
bridge, UK, 1990.
[17] H.R. Lot. Finite element analysis of fracture in concrete and masonry structures. Ph.D.
thesis, University of Colorado, 1992.
[18] Th. Zimmermann Milan.J. Embedded crack model: I. Basic formulation. International Jour-
nal for Numerical Methods in Engineering, 2001.
[19] M.Jirasek. Comparative study on nite elements with embedded discontinuities. Comput.
Methods Appl. Mech. Engrg., 2000.
[20] M.Jirasek. Objective modeling of strain localization. Laboratory of Structural and Contin-
uum Mechanics, 2002.
[21] Ted. Belytschko M.Jirasek. Computational resolution of strong discontinuities. World
Congress on Computational Mechanics, 2002.
[22] Th. Zimmermann M.Jirasek. Analysis of rotating crack model. J. Eng. Mech. ASCE, 1998.
[23] Gnther Meschke Mosler J. 3D modelling of strong discontinuities in elastoplastic solids:
xed and rotating localization formulations. International Journal for Numerical Methods
in Engineering, 2003.
[24] J.G. Rots. Computational Modeling of Concrete Fractur. PhD Thesis, 1988.
[25] E. Samaniego. Contributions to the continuum modelling of strong discontinuities in two-
dimensional solids. Ph.D. thesis, 2003.
[26] B.E. Engelmann T. Belytschko, J. Fish. A nite element with embedded localization zones.
Comput. Meths. Appl. Mech. Eng, 1988.
[27] Univ. Prof. Dr. techn. G. Meschke. Finite Element Methods in Linear Structural Mechanics.
Lecture Notes,Institute for Structural Mechanics,Ruhr University Bochum, 2010.
[28] K. Washizu. On the variational principles of elasticity and plasticity, Technical Report 25-18.
Aeroelastic and Structures Research Laboratory, MIT Press, Cambridge, 1955.

Вам также может понравиться