delamination modelling and toughening …digitool.library.mcgill.ca/thesisfile104701.pdfdelamination...

136
DELAMINATION MODELLING AND TOUGHENING MECHANISMS OF A WOVEN FABRIC COMPOSITE Tadayoshi Yamanaka Department of Mechanical Engineering McGill University, Montreal February 2011 A thesis submitted to McGill University in partial fulfillment of the requirements for the degree of Doctor of Philosophy © Tadayoshi Yamanaka, 2011

Upload: doanmien

Post on 10-Apr-2018

221 views

Category:

Documents


1 download

TRANSCRIPT

DELAMINATION MODELLING AND

TOUGHENING MECHANISMS OF A WOVEN

FABRIC COMPOSITE

Tadayoshi Yamanaka

Department of Mechanical Engineering

McGill University, Montreal

February 2011

A thesis submitted to McGill University

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

© Tadayoshi Yamanaka, 2011

ii

Imagination is more important than knowledge.

Albert Einstein(1879 - 1955)

iii

Abstract

Efficient and accurate numerical simulation methods for the damage tolerance

analysis and fatigue life prediction of fibre reinforced polymers are in high

demand in industry. Problems arise in the development of such a simulation

method due to the limitations from numerical methods, i.e., delamination

modelling, and understanding of damage mechanism of woven fabric

composites.

In order to provide effective and accurate delamination modelling, a new crack

modelling method by using the finite element method is proposed in this study.

The proposed method does not require additional degrees-of-freedom in order

to model newly created crack/delamination surfaces. The accuracy of

delamination growth simulation by the proposed method and that of a

commercial FEA package are in good agreement.

The damage mechanisms of five harness satin weave fabric composite is studied

by creating a multiscale finite element model of a double cantilever beam

specimen. The weft and warp yarns, where the gaps are filled with matrix, are

individually modeled. Cohesive zone model elements are pre-located within the

matrix and interfaces of matrix-yarns and weft-yarns and warp yarns. These

meso-scale parts are bonded with homogeneous parts that are used to model

regions where no damage is expected. This constitutes a multiscale model of a

DCB specimen. The simulation results are in good agreement with the lower

bound of experimental results. The toughening mechanism contributed from the

weave structure was revealed.

This study contributes to knowledge by introducing crack modelling methods

and by providing more information in order to understand damage mechanisms

of 5HS weave fabric composite laminates during delamination growth.

iv

Résumé

Les méthodes de simulation numériques efficaces et exactes pour l'analyse de

l’endommagement et la prédiction de vie en fatigue des matériaux composites

sont essentielles pour l'industrie. Les problèmes surviennent dans le

développement d'une telle méthode de simulation en raison des restrictions des

méthodes numériques, c'est-à-dire, modélisation de la délamination et

compréhension des mécanismes de rupture de composites à base de fibres

tissées.

Pour developer un modèle de délamination efficace et précis, une nouvelle

méthode est proposée dans cette étude en utilisant la method des éléments

finis. La méthode proposée n'exige pas de degrés-de-liberté supplémentaires

pour créer de nouvelles sufaces de fissures/ délaminations. Le résultat de

simulation de délamination par la méthode proposée est comparé avec un

logiciel d'éléments finis commercial, et les résultats se comparent bien.

Les mécanismes d’endommagement d’un composite tissé typique “five-harness

satin” sont le sujet d’une étude. Ceci est fait en créant un modèle d'éléments

finis “méso-échelle” en utilisant l’exemple d’un spécimen d’essais Mode 1

(spécimen DCB). Le tissu est modélisé avec les trajectoires exactes des fibres

dans les deux directions, et les espaces entre les fibres sont remplis de la

matrice. Des éléments cohésifs sont insérés entre la matrice et les interfaces des

fibres. Les composants méso-échelles sont joints avec des parties homogènes qui

sont utilisées pour modéliser des régions où aucun endommagement n'est

prévu. La combinaison des ces parties constitue un modèle multiéchelle d'un

spécimen DCB. Les résultats de simulation d’un essai sont en accord avec les

résultats expérimentaux, du côté conservateur. Le mécanisme renforçant des

ultant du type de tissage a été démontré.

v

Cette étude contribue à la science en présentant de nouvelles méthodes pour

modéliser les fissures et pour comprendre les mécanismes d’endommagement

des composites tissés pendant la croissance des délaminations.

vi

Acknowledgement

Support for “CRIAQ Project 1.15: Optimized Design of Composite Parts” was

provided by Bell Helicopter Textron Canada, the National Research Council of

Canada (Aerospace Manufacturing Technology Centre, Institute for Aerospace

Research), Delastek Inc., McGill University, École Polytechnique de Montréal, the

Natural Sciences and Engineering Research Council of Canada (NSERC), and the

Consortium for Research and Innovation in Aerospace in Quebec (CRIAQ). The

McGill Composite Material and Structures Laboratory is a member of the Centre

for Applied Research on Polymers and Composites (CREPEC).

I would like to sincerely thank my supervisor, Prof. Larry Lessard (McGill

University), for his kind support. He gave me many invaluable opportunities

throughout the Ph.D. program.

I would like to acknowledge Victor Feret for his experimental investigation on

mode I delamination of the double cantilever beam specimen. His test results

inspired me to work on the multiscale analysis. I would like to thank Steven Roy

for his testing on a practical application of five harness satin weave carbon fibre

fabric composite and discussions on the damage behaviour. His work certainly

gave me valuable information to decide the sub topics of my work. I am grateful

to Vahid Mirjalili for discussions on the general aspects of fracture mechanics.

Finally, I would like to thank all my colleagues in the McGill Composite Materials

and Structures Laboratory for their support throughout my studies.

vii

Table of Contents

Abstract………..……………………………………...……………………………………………………………iii

Résumé………………………………………………………….…………………………………………………..iv

Acknowledgement……………………………………………………………………………………………..vi

1 Introduction ................................................................................................................ 1

2 Crack modelling method (ADD-FEM) .......................................................................... 4

2.1 Introduction ........................................................................................................ 4

2.2 Formulation ......................................................................................................... 8

2.2.1 Problem statement ..................................................................................... 8

2.2.2 Formulation details ..................................................................................... 9

2.3 Elemental level tests ......................................................................................... 14

2.3.1 Rigid body motions (zero strains) ............................................................. 15

2.3.2 Constant strains over the elements .......................................................... 17

2.3.3 Linear strains ............................................................................................. 19

2.3.4 Selection of constraint equations ............................................................. 20

2.4 Numerical examples.......................................................................................... 28

2.4.1 Mesh ......................................................................................................... 29

2.4.2 L2-norm error distribution ........................................................................ 30

2.4.3 Stress distribution ..................................................................................... 33

2.4.4 Convergence of Strain Energy Release Rate (SERR) .................................. 35

2.4.5 Delamination growth simulation .............................................................. 38

2.5 Conclusions ....................................................................................................... 43

3 Multiscale finite element analysis of a double cantilever beam specimen made of

five harness satin weave fabric composite ....................................................................... 44

3.1 Introduction ...................................................................................................... 44

3.1.1 Failure behaviour of five harness satin weave carbon fibre fabric

composite ................................................................................................................. 44

viii

3.1.2 Multiscale finite element analyses ........................................................... 46

3.1.3 Hypothesis from experimental results ...................................................... 48

3.1.4 R- curves in delamination growth simulation ........................................... 50

3.1.5 Summary ................................................................................................... 52

3.2 Meso-scale parts of a multiscale FE model ....................................................... 53

3.2.1 TexGen ...................................................................................................... 53

3.2.2 Creating geometric models and meshing ................................................. 56

3.2.3 Material properties ................................................................................... 65

3.2.4 Element length, DCB specimen size, and contact stiffness ...................... 73

3.3 Solution procedure ........................................................................................... 80

3.4 Numerical results .............................................................................................. 83

3.4.1 Comparison with experiments .................................................................. 83

3.4.2 Energy released by CZM elements ............................................................ 85

3.4.3 Toughening mechanisms .......................................................................... 89

3.5 Discussions ...................................................................................................... 107

3.6 Future work ..................................................................................................... 109

4 Concluding remarks ................................................................................................ 110

4.1 Contributions of this thesis ............................................................................. 111

4.2 Future work ..................................................................................................... 112

Reference ………………………………………………………………………………………………………113

ix

Table of Tables

Table 2.1. Crack tip displacement field for Mode I and Mode II [38]. .............................. 23

Table 3.1. DCB specimen dimensions [22]. ....................................................................... 48

Table 3.2. Ratios of experimental results to FEA results. ................................................. 52

Table 3.3. TexGen input values used to create unit cell of 5HS weave fabric. ................. 55

Table 3.4. Dimensions of multiscale 5HS weave fabric composite DCB model. ............... 61

Table 3.5. Material properties of cured epoxy resin [63]. ................................................ 66

Table 3.6. Elastic constants of carbon fibre [64]. .............................................................. 66

Table 3.7. Fibre volume fractions. .................................................................................... 66

Table 3.8. Geometries and volumes of 5HS weave fabric’s unit cell ................................ 67

Table 3.9. Material properties of 5HS weave carbon fibre fabric composite calculated at

F.V.F.=0.838. ..................................................................................................................... 68

Table 3.10. Material properties of 5HS weave fabric composite at F.V.F.=0.55. [63] ...... 70

Table 3.11. Number of iterations for various contact stiffnesses. ................................... 71

Table 3.12. Numbers of elements for the element types. ................................................ 82

x

Table of Figures

Figure 2-1. Finite element model containing a crack. ........................................................ 9

Figure 2-2. Local node numbering and the corresponding coordinates (in parentheses)

based on the natural coordinate system for the Q4 element. ......................................... 10

Figure 2-3. Elemental level test models and their global node numbering and the

corresponding coordinates in the global coordinate system. .......................................... 13

Figure 2-4. Strain and recovered deformation: (a) Translation in the x-direction; (b)

Translation in the y-direction; (c) rigid body rotation with angle . ................................ 16

Figure 2-5. Strain and recovered deformation: (a) Translation in x-direction; (b)

Translation in y-direction; (c) rigid body rotation with angle . ....................................... 17

Figure 2-6. Principal strain and deformation: (a) Uniform load applied in x-direction; (b)

Uniform load applied in y-direction; (c) pure shear applied. ........................................... 18

Figure 2-7. Principal strain and deformation: (a) Uniform load applied in x-direction; (b)

Uniform load applied in y-direction; (c) pure shear applied. ........................................... 18

Figure 2-8. Examples of strain extrapolation. ................................................................... 20

Figure 2-9. An example of FE model with distorted mesh................................................ 21

Figure 2-10. Mesh, Master nodes, Slave nodes and constrained DOFs for a bar under

uniform tension. ............................................................................................................... 22

Figure 2-11. A beam under pure bending. ........................................................................ 22

Figure 2-12. An infinite space containing through crack at the centre in which is

applied far from the region of interest. ............................................................................ 23

Figure 2-13. Definition of the polar coordinate system ahead of a crack tip. .................. 24

Figure 2-14. Difference of normalized L2-norm error under uniform tensile loading. .... 26

Figure 2-15. Difference of normalized L2-norm error under pure bending. .................... 27

Figure 2-16. Difference of normalized L2-norm error under pure Mode I. ...................... 28

Figure 2-17. (a) structured mesh; (b) distorted mesh; (c) deformation and slave nodes (in

red) for structured mesh; (d) deformation and slave nodes (in red) for distorted mesh. 29

Figure 2-18. L2-norm error in displacement for three load cases. .................................. 32

Figure 2-19. Comparison of stress distributions obtained by ADD-FEM and standard

FEM. .................................................................................................................................. 35

Figure 2-20. Convergence of strain energy release rate obtained by VCCT for structured

mesh. ................................................................................................................................. 36

Figure 2-21. Convergence of strain energy release rate obtained by VCCT for distorted

mesh. ................................................................................................................................. 37

Figure 2-22. Convergence of strain energy release rate obtained by VCCT under mixed-

mode boundary conditions. .............................................................................................. 38

Figure 2-23. Double cantilever beam specimen. .............................................................. 39

Figure 2-24. The delamination growth simulation algorithm with ADD-FEM. ................. 41

Figure 2-25. Comparison of load-opening displacement of a DCB specimen. ................. 42

Figure 3-1. An example of composite structure made of 5HS weave carbon fibre fabric

reinforced epoxy [41]........................................................................................................ 45

xi

Figure 3-2. Damage on (a) the bracket on the left and .................................................... 45

Figure 3-3. Correction factor of modified beam theory [60]. ........................................... 49

Figure 3-4. Mode I critical energy release rates of 5HS weave carbon fibre fabric

composite. ........................................................................................................................ 49

Figure 3-5. R-curve applications. ...................................................................................... 50

Figure 3-6. Load-Displacement curves of 2D plane strain models of DCB specimen. ...... 51

Figure 3-7. Load – Delamination extension curves of 2D plane strain models of DCB

specimen. .......................................................................................................................... 52

Figure 3-8. Screenshot of TexGen GUI to create 5HS weave fabric. ................................. 55

Figure 3-9. Created 5HS weave fabric by TexGen. ............................................................ 56

Figure 3-10. Surfaces of 5HS weave fabric model imported to ANSYS®. .......................... 56

Figure 3-11. Matrix and yarn volumes of a 5HS weave fabric unit cell. ........................... 58

Figure 3-12. Yarns embedded within matrix domain. ...................................................... 59

Figure 3-13. Finite element model of a multi-scale 5HS weave carbon fibre fabric

composite DCB specimen. ................................................................................................ 60

Figure 3-14. Warp yarns and weft yarns with their numbering. ...................................... 60

Figure 3-15. Enlarged meso-scale model used in the FE DCB model. .............................. 61

Figure 3-16. Side view of enlarged meso-scale model used in the FE DCB model. .......... 62

Figure 3-17. Contact elements used in the FE DCB model. .............................................. 63

Figure 3-18. Typical contact elements used with a cohesive law. .................................... 64

Figure 3-19. Side view of contact elements used with a cohesive law showing empty

space indicating no damage within yarns is assumed. ..................................................... 64

Figure 3-20. Boundary conditions for the FE DCB model. ................................................ 65

Figure 3-21. Unit cell of unidirectional fibre composite. .................................................. 67

Figure 3-22. E11 variation of warp yarn 2 due to the slope of yarn. ................................. 68

Figure 3-23. Unit cell of 5HS weave carbon fibre fabric composite for elastic constants

calculations. ...................................................................................................................... 69

Figure 3-24. Bilinear cohesive law. ................................................................................... 71

Figure 3-25. 2D plane strain long DCB model with element length 0.12mm. .................. 74

Figure 3-26. 2D plane strain short DCB model with element length 0.12mm.................. 75

Figure 3-27. Homogeneous 3D short DCB model with contact elements bonding 5 parts

together. ........................................................................................................................... 75

Figure 3-28. Mode I energy release rates obtained by homogeneous DCB models. ....... 76

Figure 3-29. Ratios of Mode I critical energy release rates. ............................................. 77

Figure 3-30. Initial linear slope of DCB models with various contact stiffnesses. ............ 78

Figure 3-31. Contact elements coloured in red, blue, and green used in homogeneous 3D

short DCB model. .............................................................................................................. 79

Figure 3-32. Load-Displacement curves up to initial stage of delamination growth of

homogeneous DCB models with various contact stiffnesses. .......................................... 80

Figure 3-33. Time-step increment history over the entire simulation of Multi2.............. 81

Figure 3-34. R-curves of the 5HS weave carbon fibre fabric composite DCB specimen and

the multiscale FE models. ................................................................................................. 84

xii

Figure 3-35. Load-displacement curves of the 5HS weave carbon fibre fabric composite

DCB specimen and the multiscale FE models. .................................................................. 85

Figure 3-36. Released energies of the multiscale FE model of 5HS weave carbon fibre

fabric composite DCB and 2D plain-strain homogeneous FE model. ............................... 87

Figure 3-37. Delaminated areas at the end of each zone of the multiscale FE model of

5HS weave carbon fibre fabric composite DCB. ............................................................... 88

Figure 3-38. Percentage of released energy by tangential debonding within the total

released energy by CZM elements.................................................................................... 89

Figure 3-39. Contour plot of on the yarns at Point Ap and delaminated elements

coloured in pink. ............................................................................................................... 90

Figure 3-40. Contour of contact pressure of the multiscale FE model and delaminated

elements (coloured in pink). ............................................................................................. 91

Figure 3-41. Contour of contact pressure of the meso-scale FE model under in-plane

tensile loading. .................................................................................................................. 91

Figure 3-42. Contact pressure distribution on at Point Ap. ........................................ 93

Figure 3-43. Contour plot of at Point Ap clipped at and delaminated elements

(coloured in pink). ............................................................................................................. 94

Figure 3-44. near delamination front at Point Ap and reversed contact stress

obtained by in-plane loading. ........................................................................................... 95

Figure 3-45. Delamination front development during the load drop from Point Ap to Ab.

.......................................................................................................................................... 96

Figure 3-46. The z-coordinates of delaminated elements showing the branching at Point

Ab. ...................................................................................................................................... 97

Figure 3-47. Length of positive from the delamination front at Point Ab................... 98

Figure 3-48. distribution history from Point Ab to Bb. .............................................. 100

Figure 3-49. Delaminated CZM elements from Point Ab to Bb. ....................................... 101

Figure 3-50. Weft yarn bridging of Multi2 observed at Point Bb. ................................... 102

Figure 3-51. of CZM element on the delamination front edge with the delamination

front z-coordinates at Point Bb. ...................................................................................... 102

Figure 3-52. Delamination area of Multi2 versus delamination length. ......................... 103

Figure 3-53. Delamination area of Multi1 versus delamination length. ......................... 105

Figure 3-54. Weft yarn bridging of Multi1 at Point Bb. ................................................... 106

xiii

List of Symbols

= crack length

=the strain-displacement matrix

=width of a DCB specimen

=material moduli tensor

= compliance

= artificial damping

= derivative operator matrix

= correction factor

= Load point displacement

= damage parameter in mixed mode

= damage parameter in normal direction

=Young’s modulus

= error

=small strain tensor

=strain field within an element

= normal strain energy release rate

= tangential strain energy release rate

= strain energy release rate in Mode I

= strain energy release rate in Mode II

= Mode I fracture toughness

= Mode II fracture toughness

=the boundary of

= height

= thickness

=moment of inertia

=global stiffness matrix

= normal contact stiffness of cohesive zone model

= tangential contact stiffness of cohesive zone model =stress intensity factor for Mode I =stress intensity factor for Mode II

=Kolosov constant

= length from delamination tip to specified magnitude of

=length

=Moment

=direction vector

=shear modulus

=shape function

=unit normal vector

=Poisson’s ratio

= the domain of a body

= the domain of an element

= percentage of normalized L2-norm error

= Load

=force vector

xiv

=field of real numbers

and =polar coordinate system

=Cauchy stress tensor

= ultimate tensile strength

= ultimate shear strength

=tensile stress

= current time

=prescribed tractions

= time interval

= released energy

=displacement vector

=prescribed displacements

=displacements at node in the direction of the global coordinate system

= critical normal separation

= critical tangential separation

=displacements at node in the direction of the global coordinate system

and =global coordinate system

= natural coordinate system

= width

=gradient operator

1

1 Introduction

Fibre Reinforced Polymers (FRP) are being used in the various industries. For

example, Glass Fibre Reinforced Polymers (GFRP) are used for making wind

turbine blades [1-4]. Carbon Fibre Reinforced Polymers (CFRP) are used for

making more weight critical components, such as suspensions of formula one

cars [5-6] and airframes [7-8]. It has been proven that FRPs have better

performance than the other materials in certain applications.

Computer Aided Engineering (CAE) is an essential step in the structure analysis,

especially in the early design phases. Finite Element Analysis (FEA) is the most

widely used method for analyzing the solid structures over other numerical

methods. The strength of composite structures is then predicted by using failure

criteria. For instance, Maximum stress, Maximum strain, Tsai-Wu, Hill, and

Hoffman failure criteria are supported by MSC.Nastran® [9]. In most cases, the

design of composite structures is finalized by using the failure criteria [10] in

order to predict initial damage (First Ply Failure).

If one wants to analyze beyond First Ply Failure, progressive damage modelling,

which reduces the material moduli upon failure of elements, is available in MSC.

Nastran®. However, delamination, which is one of the most critical types of

failure in composite laminates, is not explicitly modeled by the progressive

damage modelling technique, which is based on continuum damage mechanics.

Accordingly, it is not very easy to predict the onset and propagation of

delamination in a composite structure. One example illustrating the difficulty of

prediction is the delaminations that occurred in the stringers of wings and centre

wing box made of CFRP on a Boeing 787 [8, 11]. Due to this delamination

damage and resulting redesign of the structures, the development of Boeing 787

has been further delayed.

2

Some aspects of delamination analysis are provided by commercial software.

Virtual Crack Closure Technique (VCCT) is available in ABAQUS®[12] and

MSC.Nastran®. VCCT is used for calculating the energy release rate at a

delamination tip and requires existing delaminations. Thus, it is used for damage

tolerance analysis and propagation analysis of existing delaminations. Cohesive

zone models are available in ANSYS®[13], ABAQUS® and MSC.Nastran®. This

feature can predict the onset of delamination by using the out-of-plane ultimate

strength of composite laminates. Both approaches, however, require advance

specification of delamination growth path in order to analyze the propagation.

This determination artificially limits the flexibility of the delamination growth

path regardless of the criteria that are used. Accordingly, FE model may require a

very large number of elements in order to give maximum flexibility for

delamination growth or require experience in order to guess the location of the

potential regions [14-16]. Neither the VCCT for ABAQUS nor the cohesive zone

model is very efficient or useful in optimizing the design of a structure. In order

to overcome the difficulties caused by modelling of delaminations, a crack

modelling method is proposed in Chapter 2.

A new delamination modelling technique is developed in Chapter 2. However, in

order to use it for a practical case, it is necessary to understand the failure

mechanism and environment in which the actual application is used. For

composite laminates, delaminations may occur as a result of cyclic loadings. For

example, wind turbine blades will experience more than 108 load cycles in the

lifetime of 20 years, according to Ref. [1]. Fatigue life predictions based on FEA

are conducted by [2-3] for a wind turbine blade and [17] for a tail cone exhaust

structure. The S-N curves of the sample coupons and stresses obtained by FEA

are used to predict the fatigue life [2]. This method does not consider the stress

re-distributions due to accumulated damages by cyclic loadings. Progressive

3

fatigue failure analyses based on continuum damage mechanics are applied by

[3, 17]. As continuum damage mechanics do not explicitly consider

delaminations, this approach may not be very suitable for some structures that

are subjected to the loadings causing delamination damage, e.g., the suspension

system of a Formula 1 car[6]. To overcome this limitation within continuum

damage mechanics, the idea of using a cohesive zone model for delamination

onset and propagation due to cyclic loadings is suggested by [18-21]. The

cohesive zone model for fatigue failure is developed based on the Double

Cantilever Beam (DCB) specimen because it is used for the fatigue testing of

composite laminates. In order to use the cohesive zone model validated by DCB

specimens and to provide better correlation with experiments, it is necessary to

understand the damage mechanisms of DCB specimens. This is very important

for some types of composite laminates that have very complex failure

mechanism. The complex failure mechanisms are believed to contribute to the

increasing Resistance curve (R-curve). For example, it is reported that 5 harness

satin weave fabric composite has toughening up to certain crack extension [22-

23]. The analyses of toughening mechanisms were conducted by post

experimental observations. Analyzing the damage mechanism by only

experimental observations may not suffice because it is not very easy to visualize

the internal damage development and the stress/strain distributions within the

DCB specimen during the test. To reveal the damage and toughening

mechanisms of a DCB specimen made using five harness satin weave fabric

composite under static loading, which is essential to the development of

cohesive zone model for fatigue loadings, the delamination growth is simulated

by FEA in Chapter 3.

The technology required by industries is an efficient and accurate damage

prediction capability under static and fatigue loadings, as clearly stated by the

author of Ref. [6]. This study partially contributes to knowledge by introducing

crack modelling methods and by providing more information in order to better

4

understand the damage mechanisms of five harness satin weave fabric

composite laminates during delamination growth under static loadings.

2 Crack modelling method (ADD-FEM)

2.1 Introduction

Damage tolerance analyses and fatigue life simulations are an important topic

for researchers and engineers. At the same time, Finite Element Methods (FEM)

are the most widely used numerical method for solving structural applications

for design. Commercial FEA software packages, e.g., ANSYS®, ABAQUS®,

MSC.Nastran®, have crack modelling features which require a user to specify the

possible crack propagation path by inserting interface/contact elements. This

process can take significant amounts of time to create a FE model if the possible

crack propagation path is complex and/or the model itself is complex. This is

because the entire structure should be divided into two or more components

and interface/contact elements must be inserted at the interfaces of

components. If one wants to add maximum flexibility for the possible crack

growth path by inserting interface/contact elements, there will be the following

issues:

1. A very high number of components (unmeshed volumes) is required.

2. The modelling time for inserting interface/contact elements could be

high.

3. The Newton-Raphson method for nonlinear analysis with a cohesive law

is not guaranteed to converge, depending on pre-defined crack paths.

In addition to these issues, a Cohesive Zone Modelling (CZM) element requires a

small enough element length depending on the materials used [24]. If the initial

model ends up with convergence difficulties during the crack propagation

analysis, one will need to revise and re-do the modelling again until one achieves

5

a successful result. Due to all the above reasons, crack modelling methods that

do not require extensive modification of the geometric model are very attractive

for damage tolerance analysis and/or fatigue life simulation.

Performing fatigue life simulation at the design phase may reduce the risk of re-

designing without full-scale model experiments, which consequently reduces the

cost of development. In the past decades, the strong discontinuity approach has

been popular for solving crack propagation problems by FEM. This approach is

capable of containing a crack, i.e., strong discontinuity, within an element.

Consequently, crack propagation analysis by this approach will result in less

remeshing during the solution phase and less modifications of the geometric

model. One of these approaches is called eXtended FEM (XFEM), which can

model a crack within an element and enriches the singularity field near a crack

tip with additional degrees-of-freedom (DOFs), as found in Ref. [25-26].

Embedded FEM (EFEM) is another method that can model a crack within an

element by strain softening with a jump parameter, such as in Ref. [27-29]. The

additional jump parameter will be condensed before assembling the global

stiffness matrix. Therefore, there are no additional DOFs to model a discontinuity

for the purposes of modelling a crack. The drawback of EFEM is the lack of ability

for modelling the crack tip. The fracture problem considered by linear elastic

fracture mechanics does not have crack tip opening displacement. However, it is

not possible to prevent the crack tip opening displacement by using EFEM. This

drawback limits the crack propagation criterion that can be used with EFEM.

Boundary element methods (BEM) can also deal with crack propagation analysis

[30]. Although the BEM provides better solution accuracy compared to FEM for

same level of discretization, the displacements, stresses, and strains at internal

points by BEM require Gaussian integrations over all the boundary elements

[31]. When there is no initial crack inserted and stresses are used to find the

location of crack nucleation, BEM will certainly require time consuming Gaussian

6

integrations over the boundary elements many times. It is stated by the authors

[31] that “If, however, the solution is required throughout the domain of the

body, the FEM program, for a given level of solution accuracy, runs faster than

the BEM program.” Accordingly, the BEM is not very suitable for crack growth

analysis without any initial crack inserted, which consequently requires a

criterion based on internal stresses/strains to predict crack nucleation.

In addition to the crack modelling methods, remeshing approaches could be an

alternative solution for crack propagation problems. The remeshing approach

generates a new mesh that follows the crack propagation path. As examples,

crack propagation obtained by the remeshing approach for various problems can

be found in Ref. [32]. This remeshing approach is more frequently used in

simulating crack growth of isotropic materials than for laminated composite

materials. This is because it is more difficult to analyze the delamination within

CFRP by using a remeshing technique due to the fact that more complex material

properties, ply-orientation, and the geometry of laminate need to be considered

as variables while remeshing. An alternative remeshing-like technique used for

modelling delaminations within CFRP can be found in Ref. [33]. This technique

separates the nodes in order to create a delamination. Therefore, matrix cracks

and delamination locations are limited to inter-element interfaces.

In this chapter, a new approach to model a displacement discontinuity within

quadrilateral elements without additional DOFs is developed and presented.

There are two steps in the procedure to obtain the stiffness matrix: (1)

constructing constraint equations according to the geometries of elements, and

(2) applying constraint equations by using a transformation matrix to reduce the

size of the stiffness matrix. The first step uses the extrapolation of the

displacement gradient of adjacent element to the element containing a crack,

i.e., the target element. The extrapolation is obtained by forcing the shared

nodes of adjacent and target elements to have the same displacement gradient.

This condition enables one to find displacements of a node on the crack face as a

7

function of the nodes of an adjacent element. Accordingly, there is no need to

add extra nodes by introducing a crack within an element. The constraint

equations are obtained for a bilinear quadrilateral element and are used for

modelling delamination in beam/shell structures.

One of the disadvantages of the proposed method is relatively high error in the

region where the displacement gradient is high, i.e., near a crack tip. This error is

due to the extrapolation of the displacement gradient. In other words, this

extrapolation gives better performance where the displacement gradient is low,

i.e., far from a crack tip. The other disadvantage is that the slight stiffening effect

is observed. This effect results from the error caused by extrapolation as well. It

is observed that the stresses in the elements containing crack are higher than

those of standard FEM.

In this research, only the formulation for a bilinear quadrilateral element is

provided. The most suitable application for this type of element is delamination

growth simulations because of the number of elements used and moderately

accurate energy release rate can be obtained by VCCT. This proposed method

could be generalized to obtain the constraint equations for other types of

elements that are more suitable for other applications. However, the

generalization is not the focus of this research. This research rather focuses on

the practical application for which the proposed method can be immediately

applied.

It should be noted that a delamination is a type of crack that commonly occurs in

a laminate. In this paper, “delamination” refers to a particular type of crack,

whereas the term “crack” is used to express the more general case of a crack.

8

2.2 Formulation

2.2.1 Problem statement

Delaminations within CFRP, especially with a brittle matrix, i.e., epoxy, can be

successfully predicted by linear finite element methods with linear fracture

mechanics, i.e., VCCT [34]. While delamination problems sometimes require a

large deformation formulation, for the sake of development, the focus here is on

the linearized strain-displacement relationship defined as

(2.1)

where is the small strain tensor, is the displacement, and is the gradient

operator. The body force term in the equilibrium equations is neglected, i.e.,

in (2.2)

where is the Cauchy stress tensor, and is the domain of the body.

Stress-strain relationships are given by

(2.3)

where is material moduli tensor. The essential and natural boundary

conditions are

on , on (2.4)

where is the boundary of with unit normal vector , prescribed

displacements , and prescribed tractions .

The displacement discontinuity considered in this study is shown in Figure 2-1.

On the crack faces, traction-free conditions are applied. The bilinear

quadrilateral element (Q4) is chosen for the development since quadrilateral

elements are one of the suitable elements for delamination growth simulation of

composite laminates.

9

(a) Element with through crack line (b) Sub-element A with attached

element and sub-element B

Figure 2-1. Finite element model containing a crack.

2.2.2 Formulation details

Figure 2-1b shows the sub-elements divided by a crack line and the attached

element above sub-element A. These sub-elements have slave nodes’ degrees of

freedom that will be eliminated later by applying proper constraint equations

that is expressing the DOFs of slave nodes as functions of master nodes’ DOFs. In

order to condense the slave nodes’ DOFs, the solution of the attached element is

extrapolated to sub-element A. The extrapolation is managed by assuming the

derivative of the displacement field at nodes shared by the attached element

and sub-element A are the same. For the case considered in Figure 2-1b, the

assumption can be written as

(2.5)

10

where is the displacement vector of the attached element, is the

displacement vector of the sub-element, and are the natural coordinate

system as shown in Figure 2-2, and is the direction which satisfies the

condition derived in Appendix A. The condition obtained is that the direction of

cannot be parallel to the edge from Node 3 to Node 4 in Figure 2-1 (b) in order

to obtain the constraint equations. It is noted that the derivative of the

displacement field is not identical at the element boundary in the displacement

based FEM while Eq. (2.1) assumes them to be the same.

Figure 2-2. Local node numbering and the corresponding coordinates (in

parentheses) based on the natural coordinate system for the Q4 element.

The basic notations used in the Q4 element are reviewed before deriving the

constraint equations. The shape functions of the Q4 element are given by

(2.6)

where and are coordinates of and at local node number given in

Figure 2-2, respectively. The displacements at node in and of the global

coordinate system are given by and , respectively. Another local node

1

11

numbering is introduced for the attached element and sub-element A as shown

in Figure 2-1b in order to derive the constraint equations.

The displacement gradient along a direction of at a point within an element

can be expressed as a function of nodal displacements, i.e.,

where

(2.7)

By using Eq. (2.7), the displacement gradients along of sub-element A and the

attached element are respectively given by

(2.8)

and

12

(2.9)

where superscripts , , , and indicate sub-element, attached element,

slave node and master node, respectively. Now, the displacement gradients

along of sub-element and attached element are expressed by a linear function

of the DOFs. By using the assumption Eq. (2.5), Eq. (2.8), and (2.9) can be

equated, and isolating gives

(2.10)

where is the component of the derivative operator matrix obtained for the

location of node 1 defined by Figure 2-3. As shown in Eq. (2.10) above, the DOF

of a slave node is expressed by a linear function of master nodes. The

component of derivative operator matrix can be computed once the coupling

of attached element and sub-element are modeled. Therefore, the constraint

equation can be explicitly obtained prior to the solution procedure of the FEM.

13

The analogous procedure is applied to obtain the constraint equations for the

rest of the slave nodes for which the derivations are given in Appendix B.

(a) Structured mesh (b) Distorted mesh

Figure 2-3. Elemental level test models and their global node numbering and the

corresponding coordinates in the global coordinate system.

Once the set of constraint equations for each set of the attached elements and

sub-elements are obtained, they can be rewritten in matrix form. The

displacement vector of the entire system including slave nodes is expressed by

(2.11)

where is a transformation matrix obtained by the constraint equations and

is the displacement vector of all master nodes. The system before condensation

is given by

(2.12)

where is the global stiffness matrix and is the force vector. By using the

matrix, the condensed system is given by

where (2.13)

14

.

This method is a general method for applying linear constraint equations without

re-ordering the stiffness matrix. It should be noted that it is not necessary to

assemble the global stiffness matrix before applying the constraint equations.

To take advantage of this method, it is preferable to apply the constraint

equations while assembling the condensed stiffness matrix. Any type of methods

that utilizes the idea of transformation, e.g., Ref. [35-36], can be used to apply

the constraint equations.

The condensed global stiffness keeps the same number of DOFs while

introducing new slave nodes for modelling cracks. Therefore, by using the

proposed method, the crack growth simulation does not increase the number of

DOFs regardless of the increase in the number of slave nodes. This additional

DOF elimination procedure to model the cracked faces is named Assumed

Displacement Discontinuity Finite Element Method (ADD-FEM).

2.3 Elemental level tests

The constraint equations should not inappropriately lock the element behaviour

as has been observed in EFEM [29]. As a minimum requirement, the sub-

elements should be capable of undergoing rigid body motions and have

adequate extrapolation of the derivatives of displacement fields from the

attached element. The strain field extrapolated to the sub-element is

investigated in this section to understand the behaviour of the constraint

equations under prescribed strain on the attached element. The two models

described in Figure 2-3 are used for numerical verifications.

Figure 2-3a shows a model (Test case A) having identical element shape for the

attached-element and sub-element. Figure 2-3b shows another model (Test case

B) having different and distorted element shapes for the attached-element and

sub-element.

15

The strain field within an element is given by

(2.14)

where is the strain-displacement matrix. The strain field within the sub-

element for the test cases is given by

where

.

(2.15)

Accordingly, the strain fields of the attached element and the sub-element are

functions of . This relationship suggests that the transformation matrix

controls the strain field of the sub-element. The following sub-sections describe

the behaviour by using various numerical examples.

2.3.1 Rigid body motions (zero strains)

The constraint equations derived in the previous section should not induce

extra-constraints preventing rigid body motions of the system. When

displacements causing rigid body rotations or translations are applied to the

master nodes of an attached element, the sub-element has to be able to

undergo the rigid body motion as well. All master nodes’ displacements have to

be prescribed in order to have rigid body motion. Accordingly, the solving

process is not required for this test since the slave nodes’ displacements are

directly recovered by using Eq. (2.11). Also, material properties are not required

for this test as constraint equations and strains are independent of them.

Figure 2-4a and b show the maximum principal strain and the recovered

deformation of sub-element and attached element. The magnitude of maximum

principal strain is zero in the attached element and sub-element for both cases.

Figure 2-4c shows the shear strain, i.e., , whose value is also nearly zero. It

should be noted that strain components and are not zero under rigid body

rotation due to the infinitesimal strain assumption. These examples show that

the constraint equations do not prevent the required rigid body motions. In

16

other words, the zero-strain field of the attached element is extrapolated to the

sub-element successfully.

(a) (b) (c)

Figure 2-4. Strain and recovered deformation: (a) Translation in the x-direction;

(b) Translation in the y-direction; (c) rigid body rotation with angle .

The results of test case B are shown in Figure 2-5. Analogous to Figure 2-4, Figure

2-5a and b show the maximum principal strain and the recovered deformation of

sub-element and attached element. Figure 2-5c shows the shear strain under

rigid body rotation. The distorted geometry of elements does not influence the

zero-strain extrapolation property of the constraint equations.

17

(a) (b) (c)

Figure 2-5. Strain and recovered deformation: (a) Translation in x-direction; (b)

Translation in y-direction; (c) rigid body rotation with angle .

2.3.2 Constant strains over the elements

The next test is to check the strain extrapolation when the attached element has

constant strains applied. When the displacement gradients at shared nodes have

the same value, the constant strains should be exactly extrapolated to the sub-

element. To verify this, the displacements that cause constant strains of 0.2 on

the attached element were applied to the master nodes. In these cases, the sub-

element is expected to have exactly the same strain field. A Young’s modulus of

1.0 Pa and a Poisson’s ratio of 0.3 were used.

The strains in the sub-element and the attached element are identical as shown

in Figure 2-6 for all cases. Since the superposition principle holds within linear

finite element methods, the sub-element and the attached element have

identical strains under any combination of constant strains. It is also verified

that there is no effect of mesh distortion as shown in Figure 2-7.

18

(a) (b) (c)

Figure 2-6. Principal strain and deformation: (a) Uniform load applied in x-

direction; (b) Uniform load applied in y-direction; (c) pure shear applied.

(a) (b) (c)

Figure 2-7. Principal strain and deformation: (a) Uniform load applied in x-

direction; (b) Uniform load applied in y-direction; (c) pure shear applied.

19

2.3.3 Linear strains

Besides constant strains, the Q4 element is capable of handling bi-linearly

distributed strains within the element. As shown in Eqs. (2.14) and (2.15), the

strains of the attached element and sub-element are functions of displacements

at the nodes of the attached element. Displacements are the unknown

variables used to obtain the strains in the attached element and sub-element. By

introducing the new strain-displacement matrix for the sub-element, Eq. (2.15)

can be rewritten as

where

(2.16)

Strains of the attached element and sub-element are explicitly defined by Eq.

(2.14) and (2.16), respectively. The difference between the strains is governed by

the difference between and . If the attached element and sub-element

are rectangular, the derivative of and with respect to the local

coordinate system shown, in Figure 2-3a, yields

(2.17)

The components of matrix and are constants in this case.

This fact suggests that the gradient of strain is constant for both attached and

sub-element meaning that the gradient of strain in the attached element is

extended to the sub-element.

An example of linear strain computed at the nodal position is visually shown in

Figure 2-8. Displacements are prescribed at the attached element’s nodes as in

the previous section. A Young’s modulus of 1.0 Pa and a Poisson’s ratio of 0.3

were used. Under this boundary condition, , and show the linear

extrapolation of strain. The characteristics of linear strain extrapolation are only

observed when both the elements have a rectangular shape. It should be noted

that the components of matrix and are not constant when

either of the elements is not rectangle. However, it is not practical to have the

20

strain extrapolation characteristics for all possible shapes of the element set.

The example shown in Figure 2-8 provides easier understanding of the solution

behaviour due to the assumption made by Eq. (2.5).

Figure 2-8. Examples of strain extrapolation.

2.3.4 Selection of constraint equations

When distorted elements are used as shown in Figure 2-9, there are two possible

constraint equations for slave node 8; Case 1: Node 8’s constraint equation is

obtained by using the set of attached element 1 and sub-element 1 and Case 2:

Node 8’s constraint equation is obtained by using the set of attached element 2

and sub-element 2. Since only one constraint equation is allowed to be assigned

for a slave node, only one constraint equation can be chosen among them. In

order to assess the difference caused by the selection of constraint equations,

two possible constraint equations, i.e., Case 1 and Case 2, are compared. As the

strain extrapolation depends on the boundary conditions, three types of

boundary conditions are tested.

21

Figure 2-9. An example of FE model with distorted mesh.

The first problem is a bar under uniform tensile loading as an example of

uniform strain cases as shown in Figure 2-10. The exact displacement solution is

given by [37]

(16)

where is the Poisson’s ratio, and are the displacement in and ,

respectively. It should be noted that notation “ ” is hereafter used to prevent

the reader from misreading the notation for the Poisson’s ratio. The exact

displacements are applied at master nodes of elements extracted from the bar

under tensile loading as shown in Figure 2-10. The blue nodes are the master

nodes with prescribed displacements indicated by green triangles. The red nodes

are the slave nodes.

22

Figure 2-10. Mesh, Master nodes, Slave nodes and constrained DOFs for a bar

under uniform tension.

The second problem is a beam under pure bending as an example of a linearly

distributed strain case as shown in Figure 2-11. The exact displacement solution

is given by [37]

(17)

where is Young’s modulus and . The exact displacements are

applied at master nodes of elements extracted from the beam under pure

bending as shown in Figure 2-11.

Figure 2-11. A beam under pure bending.

23

The third problem is that of an infinite space containing a through crack at the

centre. The schematic of this pure mode I case is shown in Figure 2-12. The exact

displacement solutions of Mode I and Mode II at near crack tip are listed in Table

2.1. The exact displacements are applied at master nodes of elements extracted

from just above the crack face as shown in Figure 2-12.

Figure 2-12. An infinite space containing through crack at the centre in which

is applied far from the region of interest.

Table 2.1. Crack tip displacement field for Mode I and Mode II [38].

Mode I Mode II

24

and are the stress intensity factors for Mode I and Mode II, respectively.

is the shear modulus. and define the polar coordinate system as defined in

Figure 2-13. is the Kolosov constant defined for plane strain: , plane

stress: .

Figure 2-13. Definition of the polar coordinate system ahead of a crack tip.

2.3.4.1 Numerical results

In order to assess the differences due to the constraint equation selection, the

normalized L2-norm is defined in following way. First, L2-norm error is given by

(2.18)

where and is the domain of an element. The L2-norm error

is normalized by the L2-norm of the exact displacement, which is expressed as

(2.19)

where

. The difference of

normalized L2-norm error due to the constraint equation selection is defined by

(2.20)

where the sub-script indicates the case of constraint equation selection.

Crack

25

Figure 2-14Figure 2-16 show the differences of normalized L2-norm error under

uniform tensile loading, pure bending and pure Mode I loading, respectively.

Each figure contains the 4 types of mesh. Under uniform tensile loading as

shown in Figure 2-14, for which a constant strain field is expected, there is no

significant influence of mesh type selected. Under pure bending and pure Mode I

loading as shown in Figure 2-15, some differences caused by the selection are

observed. The distinguishing difference is the error in alternating behaviour

observed for the pure Mode I loading case as shown in Figure 2-16. The

difference is positive as shown in Figure 2-16b and goes to negative as shown in

Figure 2-16c. There is no consistent way to choose the best constraint equation

from Cases 1 and 2 for all types of mesh and BCs. Moreover, the difference is

relatively small, i.e., within 1%, for the tested mesh and BCs. It is, therefore,

concluded that the selection of constraint equations does not lead to significant

difference in the solutions, thus any of them can be arbitrarily picked if there are

more than one constraint equation that can be obtained for a slave node.

26

(a)

(b)

(c)

(d)

Figure 2-14. Difference of normalized L2-norm error under uniform tensile

loading.

27

(a)

(b)

(c)

(d)

Figure 2-15. Difference of normalized L2-norm error under pure bending.

28

(a)

(b)

(c)

(d)

Figure 2-16. Difference of normalized L2-norm error under pure Mode I.

2.4 Numerical examples

Crack problems taken from linear fracture mechanics are chosen in order to

demonstrate the capabilities of ADD-FEM. The exact boundary conditions listed

in Table 2.1 are applied to the outer boundary of the FE model as depicted in

Figure 2-17. Mixed mode boundary conditions are obtained by superimposing

the displacements of pure Mode I and Mode II. By this superposition, any mixed

mode ratio can be achieved.

29

(a)

(b)

(c)

(d)

Figure 2-17. (a) structured mesh; (b) distorted mesh; (c) deformation and slave

nodes (in red) for structured mesh; (d) deformation and slave nodes (in red) for

distorted mesh.

2.4.1 Mesh

Two types of FE model used for comparative studies are introduced here. The

first type of FE model (10 10 uniform mesh) is shown in Figure 2-17a. The

dimensions for the model are: and . The second type

of FE model (10 10 distorted mesh) is shown in Figure 2-17b. The domain size

and the crack length are the same as those of the first model, but new

30

parameters are introduced to describe distorted mesh: and

. For convergence tests, is fixed, but the divisions of ,

, and are changed. The applied exact boundary conditions at the boundary

nodes are indicated by green triangles, as shown in Figure 2-17c and d. The

nodes at the edge of crack have active DOFs in order to properly apply the

displacement boundary conditions. Each set of elements that are formulated as

ADD-FEM has the same color. The red nodes indicate the slave nodes that are

eliminated during the stiffness matrix assembly and recovered in the post-

processing. For the comparison, the same mesh was used in a standard FEM

model by simply changing the slave nodes to master nodes. The exact

displacement fields near crack tip are shown in Table 2.1. An isotropic material,

which has Young’s modulus of 206.9GPa and Poisson’s ratio of 0.29, under plane

stress conditions, is used throughout the comparative studies.

2.4.2 L2-norm error distribution

The drawback that results from modelling without adding DOFs is the accuracy

of solutions and so its error pattern should be clearly understood. The proposed

extrapolation method is similar to the Euler method in the sense of using the

first order derivative of primary solutions. However, differences in error

propagation exist. The Euler method sequentially solves an ODE with a time

increment, so the approximation error propagates forward. The FEM solution is

obtained by solving the matrix simultaneously, so the error due to the constraint

equations will propagate spatially. In order to capture the error propagation

caused by eliminating the additional DOFs, the L2-norm error distribution of

standard FEM and that of ADD-FEM are compared. The L2-norm error of

standard FEM is defined by

(2.21)

where . The L2-norm error of ADD-FEM is defined by

31

(2.22)

where . The difference of error index is then given by

(2.23)

The L2-norm error distribution under pure Mode I with is

shown in Figure 2-18a, pure Mode II with is shown in Figure

2-18b, and mixed Mode is shown in Figure 2-18c. There

is no consistent pattern for error propagation observed by comparing the

figures. The figures also clearly show that the error propagation pattern depends

on the boundary condition applied. The common behaviour throughout these

three examples is that the error seems to be large at the bottom region just

ahead of the crack tip and at the set of attached and sub-elements. Even though

it does not seem to be possible to generalize the error propagation behaviour,

the error difference between standard FEM and ADD-FEM is not unacceptably

large and it ranges from -1.5 to 1.5%.

32

(a) Pure Mode I

(b) Pure Mode II

(c) Mixed Mode

Figure 2-18. L2-norm error in displacement for three load cases.

33

2.4.3 Stress distribution

The post-processed solution is also an important result that can be obtained by

FEM simulations. Stresses and strains are obtained from post-processed

solutions, by using displacement based FEM. These stresses and strains are used

in various failure criteria and also used to extract the strain energy release rate

by the energy domain integral method [39]. The L2-norm error distribution

shows that the error ranges for the three examples are not significantly different

from each other. Therefore, only the pure Mode I boundary condition was

chosen to compare the stress distribution obtained by standard FEM and by

ADD-FEM.

Figure 2-19a shows the contour lines of stress obtained by standard FEM

with dashed lines and that by ADD-FEM with solid lines. The largest difference

appears above the crack face where the extrapolation is used. Except in this

region, there is a good agreement. The contour lines of stress are shown in

Figure 2-19b. The contour lines of standard FEM and ADD-FEM agree very well

up to 200MPa. There are some differences below 100MPa. These lower values

are observed at the sub-elements in the ADD-FEM formulation. Figure 2-19c

shows the contour lines of stress . The closer agreement is observed below

the crack face. Overall, the stresses obtained by ADD-FEM capture a similar

stress distribution to that obtained with standard FEM.

34

(a) ; solid line: ADD-FEM, dashed line: FEM

(b) ; solid line: ADD-FEM, dashed line: FEM

35

(c) ; solid line: ADD-FEM, dashed line: FEM

Figure 2-19. Comparison of stress distributions obtained by ADD-FEM and

standard FEM.

2.4.4 Convergence of Strain Energy Release Rate (SERR)

Extracted SERR values from the FE model are compared with critical values to

check whether the crack will advance or not. Accordingly, the error in SERR

obtained by ADD-FEM has to be comparable to that by standard FEM in order to

use it for adequately accurate crack growth simulation. Also, the convergence

rate of the SERR is an important factor to give an idea of the similarities and

differences between standard FEM and ADD-FEM. The convergence is studied

for the relative error in the SERR is defined by

(2.24)

Figure 2-20 shows the convergence of SERR by standard FEM and ADD-FEM with

a structured mesh as shown in Figure 2-17a. Figure 2-20a shows the convergence

under pure Mode I deformation. The difference in relative errors of standard

36

FEM and ADD-FEM is very small. Also, the convergence rate is almost identical.

On the other hand, there is a difference in convergence rate under pure Mode II

as shown in Figure 2-20b. The convergence rate by ADD-FEM is not constant;

rather it rapidly approaches to the exact value. Figure 2-21 shows the

convergence by using the distorted mesh shown in Figure 2-17b. As observed by

the convergence of the structured mesh, the difference in relative errors of

standard FEM and ADD-FEM is very small under pure Mode I. The convergence

rate is almost identical as well. However, the convergence under pure Mode II by

ADD-FEM approached below-zero values while that of standard FEM stays as

positive error. By means of this convergence tests, it can be concluded that ADD-

FEM attains good agreement under pure Mode I. However, the convergence is

more rapid and converged value goes to negative under pure Mode II. The

difference in convergence behaviour is not significantly affected by the distortion

of elements.

(a) Convergence for pure

Mode I

(b) Convergence for pure

Mode II

Figure 2-20. Convergence of strain energy release rate obtained by VCCT for

structured mesh.

37

(a) Convergence for pure

Mode I

(b) Convergence for pure

Mode II

Figure 2-21. Convergence of strain energy release rate obtained by VCCT for

distorted mesh.

Convergence tests of mixed Mode I and II are shown Figure 2-22 with the

structured mesh. The mixed mode ratio is defined as follows:

(2.25)

This mixed mode boundary condition is applied by superimposing the exact

displacement listed in Table 2.1 in order to make an arbitrary combination of

mixed mode ratio. Figure 2-22a shows the convergence of the SERR with

. The relative error in Mode II SERR by ADD-FEM is relatively

larger and the relative error in Mode I SERR by ADD-FEM is slightly lower than

that by standard FEM. Figure 2-22b shows the convergence of SERR with

. The difference of relative error between Mode II SERR by

ADD-FEM and that by standard FEM is reduced. On the other hand, it is

increased for Mode I SERR. Considering the results for pure mode boundary

conditions, it can be concluded that the SERR obtained by ADD-FEM has a

38

smaller difference in the dominant mode. In other words, when Mode I is

dominant as in the pure Mode I case, the difference in the relative error

becomes very small and vice versa. The influence of this error on determination

of crack growth or crack growth direction is not analyzed in this paper as this

paper focuses on introducing the novel technique itself.

a) Convergence for

b) Convergence for

Figure 2-22. Convergence of strain energy release rate obtained by VCCT under

mixed-mode boundary conditions.

2.4.5 Delamination growth simulation

The ADD-FEM method is developed in order to simulate delamination growth in

laminated composite materials. To demonstrate its capability, we used the

double cantilever beam (DCB) Mode I fracture toughness test shown in Figure

2-23 where the following dimensions and material properties have been

assumed for the beam material [40]:

39

where , and are the length, width and height of the beam, is the initial

delamination length, , , , and are the elastic constants and

Poisson’s ratio of the unidirectional composite. The Mode I fracture toughness of

the composite, , is used coupling with the following

delamination propagation criterion:

. (2.26)

In order to evaluate from the solution of ADD-FEM, VCCT was used. It is

assumed that delamination propagation direction does not change. Also, the

DCB specimen was modeled with a 2D plane strain assumption. Accordingly, a

2D rectangle with dimensions is meshed. The initial delamination is

modeled by using ADD-FEM.

Figure 2-23. Double cantilever beam specimen.

The algorithm used in the demonstration is shown in Figure 2-24. There are

constant inputs, i.e., material properties, geometry and initial crack length ,

and a variable input, i.e., opening displacement increasing step-by-step. The

opening displacement is updated according to the step . When , the global

stiffness matrix should be assembled from the scratch. In this particular case,

constraint equations are also computed in order to consider the initial crack.

Next, boundary conditions are applied to the system of equations. Then the

equations are solved. evaluated by VCCT is then compared with the mode I

fracture toughness . If is greater than or equal to , the crack length is

40

extended by an increment of , which is the element length ahead of the crack

tip, and this loop continues until drops below . During this loop, only the

components of global stiffness matrix influenced by ADD-FEM formulation are

changed to model the delamination. After that, the boundary condition at next

step will be applied.

The load-opening displacement of a DCB specimen simulation by ADD-FEM is

compared to that of VCCT for ABAQUS® which is available for ABAQUS® version

6.8 [12]. The element type used for ABAQUS® is CPE4, which is the same as the

Q4 element used for ADD-FEM. The mesh has 600 divisions in length and 8

divisions in height, i.e., a mesh for both FE models. The opening

displacement is constantly increased by 0.025mm up to 5mm for ADD-FEM.

VCCT for ABAQUS® has the feature to adapt the increment to minimize the

unnecessary solving procedure in the linear part. Therefore, the increment is not

constant throughout the analysis.

Figure 2-25 shows the simulation results of ADD-FEM and VCCT for ABAQUS®.

There are two differences observed when examining the two cases. The first one

is the difference in the initial slope. The ADD-FEM result has 97.5 N/mm while

ABAQUS® result has 93.7 N/mm. Accordingly, ADD-FEM shows 4% higher

stiffness than ABAQUS®. This difference makes the opening displacement

required for the initial delamination to propagate 4% faster as well. Since the

same element formulation is used, the cause of this increase is mainly due to the

ADD-FEM formulation. The second difference is observed in the delamination

propagation part of the simulations.

41

Figure 2-24. The delamination growth simulation algorithm with ADD-FEM.

Modify global

stiffness matrix

Start

Material Properties,

Geometry,

Assemble global

stiffness matrix

Solve

Output

(Optional)

BC(t)

Stop

Yes

No

No

Yes

42

This difference is caused by the tolerance in fracture criterion used in VCCT for

ABAQUS®. The value is set to 0.01 which is the smallest possible value to use

[12].Therefore, VCCT for ABAQUS® considers the delamination to propagate

when the following criterion is met:

(2.27)

Even though the two differences are observed, the overall load-opening

displacement has a good agreement. Therefore, the ADD-FEM can be used for

simulating delamination growth as accurately as when standard FEM is used.

Also, ADD-FEM provides easier modelling as it does not require for the user to

change the geometric property of the FE model in order to consider a crack. This

is a very powerful feature when the initial delamination size and location is not

known in advance.

Figure 2-25. Comparison of load-opening displacement of a DCB specimen.

43

2.5 Conclusions

A new crack modelling method is developed by extrapolating the solutions of

master nodes near crack faces to slave nodes at the crack faces. The derivatives

of displacement at shared nodes by the attached elements and sub-elements are

assumed to be same in order to extrapolate. The extrapolation gives the

transformation matrix to eliminate the slave nodes’ DOF from the global system.

The new method is developed in order to eventually provide the mesh

independency for crack modelling. However, the concept is only shown by using

the inter-element cracks due to the rack of ADD-FEM’s modelling capability for

arbitrary located crack tip within an element.

Elemental tests are carried out to understand the characteristics of ADD-FEM.

The choice of two possible constraint equations for one slave node does not

make significant difference. It gives easier computer implementation. The

extrapolation behaviour is also checked for constant strain and linear strain.

Constant strain over the attached element is extrapolated to the sub-element.

Linear-strain is also extrapolated when both attached and sub-elements are

rectangular.

Two linear fracture mechanics problems are used to show the convergence

behaviour of ADD-FEM and that of standard FEM. Both results show similar

behaviour especially if the case is under pure mode I or II. Slight differences are

observed in the mixed mode case. Finally, delamination propagation simulation

under pure mode I is conducted by ADD-FEM and VCCT for ABAQUS®. The load-

opening displacement curves are in good agreement for a practical case.

Accordingly, the ADD-FEM gives adequately accurate results by modelling a crack

within elements without adding any DOFs.

44

For future work, the computational cost should be compared to see whether the

fact that the use of no extra DOFs will contribute to cost. Also, the influence of

numerical error due to the assumption should be studied considering the

determination of crack growth direction by using stresses.

3 Multiscale finite element analysis of a double cantilever

beam specimen made of five harness satin weave fabric

composite

3.1 Introduction

3.1.1 Failure behaviour of five harness satin weave carbon fibre fabric

composite

Five Harness Satin (5HS) weave carbon fibre fabric is frequently used as a

reinforcement in composite structures due to its better damage behaviour and

handling in manufacturing processes compared to unidirectional fibre layers. For

such composite laminates, Mode I delamination is one of the weakest modes of

damage. An example of damage development in a composite structure is shown

in Figure 3-1. The composite structure is made of 5HS weave carbon fibre fabric

reinforced epoxy manufactured by resin transfer moulding. The airfoil is loaded

in bending and the failed brackets show delaminations. The delaminations at the

early stage of damage development are shown in Figure 3-2.

45

Figure 3-1. An example of composite structure made of 5HS weave carbon fibre

fabric reinforced epoxy [41].

Figure 3-2. Damage on (a) the bracket on the left and

(b) the bracket on the right [41].

Since delaminations are the most critical type of damage, the comprehensive

investigation of Mode I, Mode II and Mixed-Mode I-II delamination tests of 5HS

weave fabric composites are reported by [22] and [23], independently. Research

in [23] reported X-rays images of delamination surfaces showing the sub-surface

damages. The sub-surface damage is the damage that occurred between warp

(longitudinal) yarns and weft (transverse) yarns, but not within the interlaminar

46

region where dominant delaminations grow. The sub-surface damages and the

measured interlaminar fracture toughness seem to have a good correlation. The

possible effect of the transverse yarn debonding mechanism of 5HS weave fabric

composite on the toughness is mentioned in [42]. The delaminated specimens

are analyzed to understand delamination mechanisms of composite laminates.

However, the experimental results did not provide enough information to give

deterministic conclusions on the source of toughening because the fractography

data does not provide the physical states during delamination growth.

3.1.2 Multiscale finite element analyses

Finite element analysis (FEA) can potentially provide more information than

experiments especially during delamination growth if the phenomenon is

modeled properly. The accuracy of analysis depends upon the FE model because

many simplifications are generally applied in order to create a FE model. The key

point in modelling is not to consider all physical phenomena, but wisely simplify

the problem so that it can be solved. One of the examples is the meso-scale FE

models of woven fabrics, i.e., a unit cell made of the warp and weft yarns, used

to predict the elastic constants [43-46]. The elastic constants are calculated by

averaging stress over the unit cell obtained by FEA and applied averaged strain

by periodic boundary conditions [47-50]. The elastic constants obtained are

usually in good agreement with experimentally obtained values. The major

assumptions typically used for unit cell analyses are; perfect bonding between

the yarns and matrix, the yarns are homogenous materials, the unit cells are

repeated infinitely in all directions, and no voids nor damages are considered.

Misalignment of fabric plies occur in real composite laminates. Woo and

Whitcomb [51] considered the effect of misalignment on elastic constant

prediction. Misalignment gives 10% difference in the in-plane modulus and 20 to

47

50% difference in the Poisson’s ratios. It seems that it is worthwhile to consider

the misalignment if each problem can be solved in a practical time range.

Beyond the elastic properties of woven fabric composites, the unit cell approach

is also used to predict damage behaviour. Karkkainen and Sankar [52] obtained

the failure envelope of plain weave carbon fibre fabric composites. Daggumati et

al. [53] explored the local damage in 5HS satin weave carbon fibre fabric

composites with focus on the accuracy of unit cell analysis. Further damage

development within woven fabric composites are studied by utilizing continuum

damage mechanics in which the stiffness matrix is degraded according to the

failure criterion. Plain weave composites under in-plane tensile loading were

investigated and the stiffness degradation was compared with that of

corresponding experiments [54]. Various types of weave, i.e., plain, 4-, 5-, 8-

harness satin and twill, were tested under in-plane tensile and compressive

loading conditions [55]. A comprehensive road map to multiscale FE modelling

from generating textile models to progressive damage analysis is provided by

[56]. Key et al. [57] showed the multiscale progressive failure of woven fabric.

This multiscale approach decomposes the failure into its constituents, i.e., matrix

and glass fibre. Gorbatikh et al. [58], on the other hand, showed results that

expose the inadequate use of continuum damage mechanics by providing an

example of using continuum damage mechanics with an embedded crack. The

multiscale simulation of a notched beam specimen made of braided composite

considering damage using a cohesive zone model is reported by [59].

Although the majority of multiscale FE analyses so far are based on continuum

damage mechanics, it is probably better to model cracks/delaminations explicitly

in order to understand the delamination growth behaviour of 5HS weave fabric

composites.

48

3.1.3 Hypothesis from experimental results

Modelling of a phenomenon usually starts from the observation of experiments.

Within the framework of this project, Feret [22] conducted pure mode I fracture

toughness tests by following ASTM standard D5528 [60]. The schematic of a

typical DCB specimen is shown in Figure 2-23 and the corresponding dimensions

are listed in Table 3.1. It is worth noting that the initial delamination front is

straight and perpendicular to the specimen edge.

Seven samples were tested and their Mode I critical energy release rates were

calculated by using the Modified Beam Theory (MBT) method. According to the

MBT method, the mode I critical energy release rate is expressed by

(3.1)

where load, load point displacement, specimen width,

delamination length, and correction factor. The correction factor is

introduced to overcome the overestimation of due to the rotation that

occurs at the delamination front. The correction factor is graphically expressed

by plotting a least squares plot of the cube root of the compliance, , as a

function of delamination length as in Figure 3-3.

Critical energy release rates of seven samples of 5HS weave composite are

plotted in Figure 3-4. Although there is a huge scatter within the critical energy

release rate, i.e.,

, the critical energy release rate curves,

also called R-curves, of all samples reach a plateau when the delamination

length increment reaches 7mm, which is approximately the same as the width of

three tows

Table 3.1. DCB specimen dimensions [22].

Length

[mm]

Length of Teflon

insert [mm]

Initial delamination

length [mm]

Width

[mm]

Thickness

[mm]

140 60 50 20 4.40

49

Figure 3-3. Correction factor of modified beam theory [60].

Figure 3-4. Mode I critical energy release rates of 5HS weave carbon fibre fabric

composite.

A hypothesis has been put forward that the initial delamination, which has

straight front, needs several tows to develop the delamination pattern of 5HS

weave fabric composite. In order to validate this hypothesis, a multiscale FE

model, which is shown in detail in later sections, is generated and critical energy

50

release rates obtained by FEA is compared with those of experiments. The FEA

results provide more internal damage information during the delamination

growth, which is not easily observed during the experiments nor by

fractography.

3.1.4 R- curves in delamination growth simulation

Though the focus of this study is to understand the toughening mechanism

observed at early stages of delamination growth, understanding the

delamination front development phase also gives very important information to

extend the use of R-curves to more practical applications where there is no initial

delamination modeled or a very small initial delaminations exists. The basic

concept of using R-curves from 5HS weave fabric composite is shown in Figure

3-5.

Figure 3-5. R-curve applications.

One of the applications for which the R-curve can be used directly is the FEA of a

DCB specimen. This successful use of R-curves is shown in Figure 3-6 and Figure

3-7. The FE model of the DCB specimen is identical to that used in Chapter 2

except the dimensions. Accordingly, there is no inhomogeneity consideration

R-curve

Delamination

analysis without

initial delamination

Delamination

analysis with initial

delamination

Practical and useful for

more applications

Very limited applications

R-curve is unable to

be used directly.

51

except the R-curve which inherently has the inhomogeneous property. Two extra

cases, i.e., the minimum N/m and the value at the plateau

N/m, are also tested. As expected, the case using R-curves is closest to the

experimental result, Ex1, for both Displacement-Load and Delamination

Extension Length-Load relationships. On the other hand, the case with minimum

underestimates the peak load and overestimates delamination extension

length. The case with at plateau slightly overestimates the peak load and

delamination extension length. The ratios of experimental results to FEA results

are listed in Table 3.2.

Figure 3-6. Load-Displacement curves of 2D plane strain models of DCB

specimen.

52

Figure 3-7. Load – Delamination extension curves of 2D plane strain models of

DCB specimen.

Table 3.2. Ratios of experimental results to FEA results.

R-curve Minimum Maximum

Maximum delamination

extension length ratio 1.04 2.37 1.15

Maximum load ratio 1.08 0.640 1.22

3.1.5 Summary

Delamination failure could possibly be the initial failure that occurs for 5HS

weave fabric composite structures. The prediction of failure behaviour of

structures requires a good understanding of delamination growth mechanisms.

53

The Mode I delamination test using a DCB specimen shows an increasing R-

curve. The implementation of R-curves into a delamination growth model is

crucial to improve the accuracy. However, the experimentally obtained R-curve

is not directly used for any type of applications. In order to extend the use of the

experimentally obtained R-curves to structural applications, the toughening

mechanism must be understood. For this purpose, the multiscale FE analysis of

5HS weave carbon fibre fabric composite DCB specimen with CZM elements to

model delaminations and matrix cracks is conducted in this study.

3.2 Meso-scale parts of a multiscale FE model

3.2.1 TexGen

The 5HS weave fabric used as reinforcement of the composite material

investigated in this project is categorized as a 2D woven fabric. In order to create

a geometrical model of a 2D woven fabric that could be transferred to CAE

software, TexGen [61] and WiseTex [62] were considered for the modelling.

WiseTex is a commercial software developed by the composite materials group

at Katholieke Universiteit Leuven in Belgium. WiseTex is able to create models

that can be generated in ANSYS® Mechanical APDL. This capability is very

attractive and reduces the time required to create an FE model for many types of

analysis. However, the fabric model must be modified in order to insert cohesive

elements for delamination growth analysis. Thus, the capability is not a big

advantage for this particular case. On the other hand, TexGen is a free software

developed by Textile Composites Research at the University of Nottingham in

the United Kingdom. TexGen can create an IGES format file that can be read by

CAD software as well as FEA packages including ANSYS®. It also generates

meshed models with tetrahedral elements. Although both of them are capable

of creating 5HS weave fabric models that can be imported to ANSYS®, TexGen

was selected over WiseTex.

54

The TexGen website has extensive information on its software on the

Documentation page. The User Guide is well written and Graphical User

Interface (GUI) of TexGen is well designed, so it is not very difficult to create a

fabric model by using TexGen. The following data are required for creating a 2D

fabric model; number of warp yarns, number of weft yarns, yarn spacing, yarn

width, fabric thickness and gap size. The values are listed in Table 3.3 and the

corresponding screenshot of TexGen GUI is shown in Figure 3-8. While the yarn

spacing and the fabric thickness were measured values of the actual 5HS weave

fabric composite, the yarn width was changed to have large enough gap

between yarns. This gap is determined by trial and error, i.e., the processes from

model generation by TexGen to meshing by ANSYS®, were repeated several

times.

Although it is possible to make large area or many plies of 5HS weave fabric in

TexGen, only one ply and the minimum number of yarns to make a unit cell were

created before exporting the model to an IGES file. This helps to reduce the

processing time to modify and simplify the geometry for FEA. Even though the

minimum number of warp and weft yarns required for a unit cell is five, seven

warp yarns and seven weft yarns are needed as input to TexGen to make a unit

cell that has identical surfaces at each facing edge. Figure 3-9 shows the created

5HS weave fabric in which a yarn at each edge is deleted for exportation. An IGES

format file is created by TexGen and imported in ANSYS as shown in Figure 3-10.

The imported data is only the surface information to create the outline of warp

yarns along the x-direction and weft yarns along the y-direction.

55

Table 3.3. TexGen input values used to create unit cell of 5HS weave fabric.

Number of

warp yarns

Number of

weft yarns

Yarn

spacing

[mm]

Yarn

width

[mm]

Fabric

thickness

[mm]

Gap size

[mm]

7 7 2.4* 2.16 0.34* 0

*measured value

Figure 3-8. Screenshot of TexGen GUI to create 5HS weave fabric.

56

Figure 3-9. Created 5HS weave fabric by TexGen.

Figure 3-10. Surfaces of 5HS weave fabric model imported to ANSYS®.

3.2.2 Creating geometric models and meshing

No mesh dependency on delamination growth and its direction is desired.

However, there is no practically useful modelling tool causing no mesh

dependency available in either ANSYS® version 12.0 or ABAQUS®6.8. CZM seems

Warp yarns

Weft yarns

57

be more suitable for this application over VCCT available in ABAQUS®6.8 because

there will be multiple cracks, which do not have self-similar crack growth, within

the meso-scale region. ANSYS® version 12.0 offers two types of element to use

CZM, i.e., interface elements and contact elements. Since the delaminated

surfaces may be in contact again during the delamination growth simulation,

contact elements are used in order to model the cohesive zone. These contact

elements with cohesive laws can only be inserted between the volumes that are

components consisting of elements. Accordingly, the delamination growth

direction is constrained by the fineness of volumes used to model the meso-

scale region.

The possible delamination growth is restricted to occur within the matrix, at the

interfaces of matrix and yarns, and at the interfaces of weft yarns and warp

yarns. Therefore, no damage is assumed within weft and warp yarns. In other

words, no transverse cracks or fibre peeling is modeled.

Even though the larger number of generated volumes for the model tends to

give more freedom for a delamination to grow in favoured directions, it will also

significantly increase the database size and meshing time for contact elements.

The grid size in the x- and y-directions of 0.24mm seems to be fine enough to

model the yarns and limits the database size, i.e., 400MB for a single unit cell.

The created volumes before meshing are shown Figure 3-11. The embedded

volumes of yarns are plotted with transparent volumes of matrix as shown in

Figure 3-12. Also, very small gaps exist at the overlaps of weft yarns and warp

yarns. The gaps, smaller than 0.025mm, were removed by modifying yarns’

geometries. The warp and weft yarns are completely embedded in the matrix

domain with length 12mm, width 12mm, and thickness 0.36mm. Similar small

gaps exit between the outer boundary of the matrix domain and yarns. These

gaps were removed in the same manner as previously mentioned gaps.

58

It is noted that the gap where the warp yarn gap and the weft yarn gap meet

together is left empty because it is difficult to create the volumes that contact

elements can be inserted with cohesive laws by the program code written by the

author.

Figure 3-11. Matrix and yarn volumes of a 5HS weave fabric unit cell.

59

Figure 3-12. Yarns embedded within matrix domain.

In order to minimize the problem size and save computational time, the DCB

specimen with meso-scale 5HS weave fabric composite is modeled and meshed

as shown in Figure 3-13 and the dimensions are listed in Table 3.4.

Homogeneous parts are bonded to meso-scale parts by contact elements. This

simplification significantly reduces the problem size. The delamination extension

length required to reach the plateau is around 3 weft yarns in length. Therefore,

the length of DCB specimen model is also shortened. The effect of this

simplification on the FE result is studied in the section 3.2.4. The width is around

the half of a DCB specimen which contains four warp yarns as shown in Figure

3-14.

The enlarged meso-scale mesh is shown in Figure 3-15. The mesh length in the x-

direction is no longer than 0.12mm. In section 3.2.4, it is verified that this length

is short enough to give accurate results by using a cohesive law, using the

material properties.

60

The initial delamination tip is shown in Figure 3-16. The weft and warp yarns

have nine divisions in the yarn width direction. In order to have a better

transition from meso-scale parts to homogeneous parts, an initial delamination

was inserted up to the forth division of the weft yarns, i.e., 1.08mm of the meso-

scale parts has an initial delamination.

Figure 3-13. Finite element model of a multi-scale 5HS weave carbon fibre fabric

composite DCB specimen.

Figure 3-14. Warp yarns and weft yarns with their numbering.

Warp yarns Weft yarns

61

Table 3.4. Dimensions of multiscale 5HS weave fabric

composite DCB model.

Length

(mm)

Initial delamination

length (mm)

Width

(mm)

Thickness

(mm)

140 50 9.28 4.40

Figure 3-15. Enlarged meso-scale model used in the FE DCB model.

62

Figure 3-16. Side view of enlarged meso-scale model used in the FE DCB model.

Figure 3-17 shows the contact elements inserted within the simplified DCB

model. Two types of contact elements are used for the interface, which are

coloured in blue and green, and also for CZM elements, which are coloured in

red. The contact elements used for cohesive zone modelling are enlarged and

shown in Figure 3-18 and Figure 3-19. All delaminations occur only along the

contact elements. As verified in Figure 3-19, no contact elements are inserted

within yarn spaces and these spaces are left empty so that yarns cannot have

any damage.

63

Figure 3-17. Contact elements used in the FE DCB model.

64

Figure 3-18. Typical contact elements used with a cohesive law.

Figure 3-19. Side view of contact elements used with a cohesive law showing

empty space indicating no damage within yarns is assumed.

65

The boundary conditions applied to the multiscale DCB model are shown in

Figure 3-20. The displacements in the z-direction are applied at the end of each

beam to open up the DCB model, while the displacements in the x- and the y-

directions are applied to prevent rigid body motions. The red arrow shown on

the FE model indicates the acceleration due to the gravity, 9.8m/sec2. Although

there is no consideration of inertia term in this FEA, the force of gravity, which is

calculated by the mass of elements, is applied to the nodes. This force, however,

does not seem to have a significant influence on the result, but it slightly

improves the convergence of Newton-Raphson solutions.

Figure 3-20. Boundary conditions for the FE DCB model.

3.2.3 Material properties

In creating the multiscale DCB model, there are three regions with different

material properties. The matrix region within the meso-scale parts has

properties of epoxy matrix, which are listed in Table 3.5. It is noted that the

Mode II critical energy release rate and ultimate shear strength of cured epoxy

resin are assumed by the Mode I critical energy release rate and ultimate tensile

strength because the values were not available. The weft and warp yarns are

Acceleration

Displacements

66

assumed to have the same properties as unidirectional fibre composite. Its

mechanical properties are obtained by a unit cell analysis of unidirectional fibre

reinforced composite with hexagonal packing as shown in Figure 3-21. The

material properties of carbon fibre are listed in Table 3.6. The Fibre Volume

Fraction (F.V.F) of unidirectional fibre composite listed in Table 3.7 is calculated

from the volumes listed in Table 3.8. The volumes are obtained by the function

that calculates volumes of elements available in ANSYS®. The periodic boundary

conditions proposed by [49] were used to obtain elastic constants. The

calculated mechanical properties of unidirectional fibre composite are listed in

Table 3.9. The density, however, is calculated by using the rule of mixtures.

Table 3.5. Material properties of cured epoxy resin [63].

E

(GPa)

ν Density

(g/cm3)

GIC_matrix

(N/m)

(MPa)

GIIC_matrix*

(N/m)

*

(MPa)

3.1 0.3 1.22 200 70 300 105

*Note: and .

Table 3.6. Elastic constants of carbon fibre [64].

E11

(GPa)

E22=E33

(GPa)

G23

(GPa)

G12=G13

(GPa)

ν 23 ν 12= ν 13 Density

(g/cm3)

230 22.0 8.15 22.0 0.35 0.30 1.8

Table 3.7. Fibre volume fractions.

0.550 0.657 0.838

67

Table 3.8. Geometries and volumes of 5HS weave fabric’s

unit cell

Unit Cell

Thickness

(mm)

Unit Cell

Length

(mm)

Unit Cell

Volume

(mm3 )

Matrix

Volume

(mm3 )

Tow

Volume

(mm3 )

0.3600 12.00 51.84 17.80 34.04

Figure 3-21. Unit cell of unidirectional fibre composite.

68

Table 3.9. Material properties of 5HS weave carbon fibre fabric composite

calculated at F.V.F.=0.838.

E11

(GPa)

E22=E33

(GPa)

G23

(GPa)

G12=G13

(GPa)

ν 12= ν 13 ν 23 Density

(g/cm3)

195 13.7 5.13 9.22 0.301 0.348 1.62

Figure 3-22. E11 variation of warp yarn 2 due to the slope of yarn.

The warp and weft yarns have a wave pattern for which the elastic constants are

not uniform over the yarn length. One of the warp yarns, i.e., warp yarn 2, was

taken to investigate the effect of waviness on the value of E11. The slope angle

and E11 along a yarn are shown in Figure 3-22. The maximum difference of E11 is

-3% at . This difference is small and does not likely to cause significant

69

effect on the simulation result if it is not considered. Thus, material properties

are considered to be uniform over the weft and warp yarns in the meso-scale

model.

The averaged mechanical properties over a unit cell of 5HS weave composite

were also obtained by unit cell analysis. The slightly different meso-scale model

of 5HS weave fabric composite unit cell is used as shown in Figure 3-23. Unlike

the meso-scale model for DCB specimen, there is no need to insert contact

elements and it has more flexibility to mesh the model with smaller yarn gaps.

Elastic constants obtained by the unit cell analysis are compared with those

experimentally obtained, in Table 3.10. It seems that the calculated elastic

constants are not very far from the experiments. Accordingly, the similar

magnitude of agreement is assumed to exist for the other values that are not

available by experiments.

Figure 3-23. Unit cell of 5HS weave carbon fibre fabric composite for elastic

constants calculations.

70

Table 3.10. Material properties of 5HS weave fabric composite at F.V.F.=0.55. [63]

E11

[GPa]

E22

[GPa]

E33

[GPa]

G12

[GPa]

G13=G23

[GPa]

ν 12 ν 13= ν 23 Density

[g/cm3]

Unit Cell 60 60 8.73 4.60 2.80 0.0346 0.405 1.62

Exp. 64 60 n.a. 4.8 n.a. n.a. n.a. n.a.

The material properties required for a bilinear cohesive law for Mode I

debonding are critical strain energy release rate , which is the area of

shaded triangle, and strength , which is the peak normal stress, as shown in

Figure 3-24. The critical normal separation is given by

(3.2)

The softening is described by the degradation of stiffness by up to

after reaching the peak . Once reaches one, the contact elements

no longer have a cohesive law constitutive equation, but acts as standard

contact elements with no friction. The analogous relationship is applied to Mode

II debonding. The power law based energy criterion is used to define the

completion of debonding in ANSYS®:

(3.3)

where and are the normal and tangential strain energy release rate,

respectively. It should be noted that ANSYS® does not explicitly differentiate

Mode II and Mode III, but tangential strain energy release rate is used instead.

71

Figure 3-24. Bilinear cohesive law.

Since the multiscale model has many pairs of contacts, the solution behaviour is

strongly affected by the magnitude of . In order to evaluate the effect, the

DCB model shown in Figure 3-13 was loaded up to mm, which is still

in the elastic range, with the increment, mm for three values.

The maximum number of Preconditioned Conjugate Gradients (PCG) iterations

and the total number of iterations for Newton-Raphson solution for the test are

listed in Table 3.11. It is clearly shown that the higher the value, the larger

is the number of iteration needed. The maximum number of PCG iterations may

be reduced by decreasing the load increment . However, it inevitably

increases the time required for the solution. Accordingly, the initial slope

stiffness, , is determined as N/m.

Table 3.11. Number of iterations for various contact stiffnesses.

( N/m) 70 700 7000

Maximum number of PCG iterations 212 1323 9999

Total number of iterations for

Newton-Raphson solution

9 12 21

72

Noted that ANSYS® enforces the following relationship for the mixed mode:

(3.4)

where the subscripts indicates tangential variables. This expression is needed

as the damage parameter for mixed mode, , is used for calculating the

normal and the tangential contact stresses written as

(3.5)

and

(3.6)

Therefore, normal and tangential debondings are controlled only by By

using Eqs. (3.2), (3.5) and (3.6) with , Eq. (3.4) is rewritten as

(3.7)

This expression is used to give tangential contact stiffness, , for the given

material properties.

Artificial damping is used in ANSYS® to overcome the convergence difficulties in

the Newton-Raphson solution. The normal contact stress at CZM is expressed as:

(3.8)

where time interval, and is the artificial damping. It is

stated in ANSYS® Theory Reference, 4.13 Cohesive Zone Material Model[13],

that the damping coefficient has units of time, and it should be smaller than the

minimum time step size. sec. is selected from the preliminary

tests. Large artificial damping may stiffen the cohesive zone overestimation and

prevent load drop off during delamination growth. “Time” used in the cohesive

zone material model for static analysis is not the actual time as in time-

dependent analysis, but it is artificial time that monotonically increases with load

steps and used to apply the displacements with a function of the time, i.e.,

73

(3.9)

where is the current time and the constant has a unit of (mm/sec).

3.2.4 Element length, DCB specimen size, and contact stiffness

Three FE models are prepared in order to verify the material property selection.

Figure 3-25 shows the 2D plane strain model using the dimensions listed in Table

3.1. The model has a very fine mesh, i.e., element length 0.12mm with aspect

ratio 1. This result is considered a reference to ascertain the effects of a

shortened DCB specimen model and the contact stiffness of elements bonding

the meso-scale parts and the homogeneous parts. Figure 3-26 shows the 2D

plane strain version of the shortened DCB specimen model. As it does not have

bonding contact elements, this analysis shows only the end edge effect caused

by the shortened DCB specimen model. Figure 3-27 shows the 3D DCB specimen

model without meso-scale parts, but has contact elements to bond

homogeneous parts. This analysis shows the effect of contact stiffness as well as

the DCB specimen length.

First of all, it should be verified that the maximum element length in the x-

direction is 0.12mm, N/m and sec. are

appropriate selections of values for delamination growth simulation. If the 2D

plane strain long model gives energy release rate that is close to given as a

material property, the model and selected material properties are considered to

be within an appropriate range. Calculated is shown in Figure 3-28. The

correction factor introduced in Eq.(3.1) is used for this case only. The

averaged value is 204N/m which is just 2% larger than the given .

Accordingly, the selected element length and material properties are used for all

the other tests in this study.

The rest of the models are also tested and their energy release rates are shown

in Figure 3-28, with the 2D long DCB model as a reference. 2D and 3D DCB models

74

show the monotonically decreasing caused by edge effect of the shortened

specimen model. This tendency is similarly obtained for the case with

N/m, i.e., MPa and mm. Due to the strong edge effect,

the correction factor did not work properly and was not applied for these

cases resulting a slight overestimation of at the very beginning of

delamination growth. The decreasing trend is more clearly shown with the ratio

of to as shown in Figure 3-29. The trend is almost identical for 2D

and 3D models regardless of critical energy release rate. This result could

possibly be used to correct the energy release rate obtained by multiscale DCB

model as the tendency is not significantly affected by the magnitude of energy

release rate. The correction factor is the function of delamination length written

as:

(3.10)

where is the coefficient determined by polynomial curve fitting of the ratio of

to for 3D model. Dividing by gives the corrected , i.e.,

, expressed by

(3.11)

Figure 3-25. 2D plane strain long DCB model with element length 0.12mm.

75

Figure 3-26. 2D plane strain short DCB model with element length 0.12mm.

Figure 3-27. Homogeneous 3D short DCB model with contact elements bonding 5

parts together.

76

Figure 3-28. Mode I energy release rates obtained by homogeneous DCB models.

77

Figure 3-29. Ratios of Mode I critical energy release rates.

The contact stiffnesses at the bonding areas used for 3D models are investigated

by varying their values. Figure 3-30 shows the load – displacement curves within

elastic range by varying the values of , and defined in Figure 3-31

with fixed tangential contact stiffnesses, i.e., N/m and

. It shows that N/m gives fairly close

response to the 2D reference. From this test, three candidates, i.e., K3, K4 and

K1, are chosen for another test and the result is shown in Figure 3-32. It seems

that the difference between K3 and K4 are small. Also, it is already shown in

Table 3.11 that the smaller contact stiffness tends to give smaller number of

iterations. Thus, K1 and K4 are chosen for the delamination growth simulations

78

of multiscale DCB model and the results are presented in the next section as

Multi1 and Multi2, respectively.

Figure 3-30. Initial linear slope of DCB models with various contact stiffnesses.

79

Figure 3-31. Contact elements coloured in red, blue, and green used in

homogeneous 3D short DCB model.

80

Figure 3-32. Load-Displacement curves up to initial stage of delamination growth

of homogeneous DCB models with various contact stiffnesses.

3.3 Solution procedure

In addition to the FE mesh and material properties, the parameters that are

chosen for the solution procedure affect the results. The most influential factors

are the convergence criterion for the Newton-Raphson solution and time-step

increment that controls the load-step increment and the amount of contact

stress update described as Eq. (3.8). Optimized default values for the

convergence criterion, i.e., L2 norm of force tolerance equal to 0.5%,

recommended by ANSYS® has been used. A homogeneous DCB model uses a

large time-step increment until the initiation of delamination and then uses a

81

smaller one during delamination growth. In principle, the multiscale DCB model

follows the same rule as a homogeneous DCB model. However, crack-arrest

occurs during the solution which is in contrast to a homogeneous DCB model.

Accordingly, a very aggressive time-step increment change has been made as

shown in Figure 3-33.

Figure 3-33. Time-step increment history over the entire simulation of Multi2.

The minimum time-step increment is set to [sec] which is slightly

larger than that is [sec]. Therefore, the minimum time-step

increment remains larger than for the entire solution as suggested by

ANSYS®.

To reduce the number of iterations required for each step of Newton-Raphson

procedure, the predictor and the line search were activated. The predictor is

used to predict the displacements of current time by using last time-step. This

option reduce the number of iterations in general, but will predict too much

displacements and/or distortion resulting in termination of solution when one or

82

more elements are about to have rigid body motions. The predictor was

deactivated when the solution stopped due to occurrence of excessive

prediction. The line search is recommended to use with contact elements. The

line search updates the displacements of the next iteration by using those of the

current iteration. The factor used for update is automatically determined by

minimizing the energy of the system. Unlike the predictor, the line search must

be activated during the entire solution in order for the solution to converge.

Large rotation of elements is expected to occur during the delamination growth.

Accordingly, the option to activate the updated Lagrangian method is set to

“ON”.

More information on the convergence criterion, time-step increment, the

predictor, the line search, and the large rotation option can be found in ANSYS

Theory Reference [13].

The PCG solver is chosen over the most robust sparse direct solver due to the

advantage of computational time. Also, the multiscale DCB model has a very

large number of contact and target elements as listed in Table 3.12. Such a

model does not have good scalability of Distributed ANSYS and has better

performance using shared-memory parallel processing. As the scalability of

shared-memory parallel is limited by memory access issues, this type of analysis

requires a huge computational time.

Table 3.12. Numbers of elements for the element types.

Type of element SOLID185

(8-node brick

element)

CONTA173

(4-node contact

element)

TARGE170

(4-node target

element)

Number of

element 249,891 198,172 112,810

83

Finally, element death options are activated to the element that causes

convergence difficulties in the Newton-Raphson solutions. Local instability due

to losing constraints by cracks growth is one of the cases.

3.4 Numerical results

3.4.1 Comparison with experiments

It is essential to compare FEA results with experimental results in order to

correctly understand the toughening mechanism of the 5HS weave carbon fibre

fabric composite DCB specimen by the model. calculated by Eq.(3.1) for the

multiscale FE model and the corrected by using Eq. (3.11) are shown in

Figure 3-34 with two experimental results that are the representatives of lower

and upper bounds. The noticeable difference in the results is the zigzag pattern

in the multiscale FE result. In the experiments, there is a limitation on measuring

the delamination growth increment. The actual delamination growth of the 5HS

weave fabric composite DCB specimen is not very smooth, but rather repeats

rapid growth and arrest. The delamination growth length is measured while

delamination arrests and the load tracked right before the next delamination

growth. This procedure gives such a smooth R-curve connecting the local

maxima of . On the other hand, the multiscale FE result provides not only

local maxima of , but the entire history of . It should be noted that there is

no consideration of an inertia term in the multiscale FE model that may affect

the results.

Both models, called Multi1 and Multi2, have very good correlation with the lower bound

of experimental results, called Ex2. When the correction is applied, of Multi1 has

slightly higher value at point Cp while Multi2 gives good correlation with Ex2.

84

Further comparisons are made by using load-displacement curves as shown in

Figure 3-35. As the of multiscale FE models gives better correlation with Ex2,

the load-displacement curve also has closer agreement with that of Ex2 than

Ex6. Comparing Multi1 and Multi2, the steeper initial slope of Multi2 suggest

that Multi2 is stiffer than Multi1 due to the higher contact stiffness in the

bonding region. The initial slope of the load-displacement curve of Multi2 has

better agreement with Ex2 than Multi1. The lower contact stiffness of Multi1

requires more displacement to reach the critical load at which the delamination

starts growing.

Figure 3-34. R-curves of the 5HS weave carbon fibre fabric composite DCB

specimen and the multiscale FE models.

85

Although the raw value of Multi2 seems to be lower at point Cp, the

corrected is able to achieve slightly higher value than that of Ex2.

Considering the better correlation of Multi2 with Ex2, the result obtained by

Multi2 seems to be better representing the 5HS weave carbon fibre fabric

composite DCB specimen and the results of Multi2 is mainly analyzed.

Figure 3-35. Load-displacement curves of the 5HS weave carbon fibre fabric

composite DCB specimen and the multiscale FE models.

3.4.2 Energy released by CZM elements

Understanding the toughening mechanisms of 5HS weave carbon fibre fabric

composite DCB specimen is the main objective of this study. The energy release

rate obtained by experiments is based on the change of energy in global scale. FE

results, on the other hand, are able to provide alternative methods to quantify

the released energy. First, the released energy at delamination length is

defined by

86

(3.12)

where is the initial delamination length and is the critical energy release

rate. The released energy due to normal and tangential delamination can be

obtained by replacing by and , respectively.

The energy release rate is averaged over each element and multiplied by its area

to obtain the released energy. Figure 3-36 shows the released energies of Multi2

and 2D reference calculated by using the results of CZM elements and

calculated by MBT. The released energy of 2D reference calculated by MBT gives

almost identical value to that of calculated by CZM elements. This result is very

straight forward as only CZM elements release the energy due to delamination

growth.

On the other hand, the total released energy by CZM elements of Multi2 is

smaller than that obtained by MBT, i.e., global approach. This result indicates

that the multiscale DCB model were able to store more strain energy than 2D

homogeneous model for the same delamination length.

The total released energy by CZM elements consists of normal and tangential

components. seems to be very similar to that of 2D reference except slight

increase in the slope around delamination length 53mm. The increase seems to

be caused by the sub-surfaces that are created by lifting up weft yarn 2. For the

sake of analysis, the released energy is divided into three zones, i.e., Zone A

covering from the initial point to Point Ab, Zone B covering from Point Ab to Bb,

and Zone C covering from Point Bb to the ends. The delaminated elements at the

end of each zone are shown in Figure 3-37. The darker portions of the

delaminated area show multiple layers, i.e., sub-surfaces. The black portions

show the killed elements due to convergence difficulties. It is shown that sub-

surfaces appear in significant amounts from the end of Zone B and keep

increasing within Zone C. This result confirms the increase in the slope of . It

87

should be noted that there is sharp drop in the released energy due to the killed

elements during the solution. This drop should not be considered as any physical

phenomena.

The released energy by tangential debonding, i.e., , has as much as 14% of

total released energy as shown in Figure 3-38. As CZM elements are placed

around weft and warp yarns, there is always a portion of tangential debonding.

Figure 3-36. Released energies of the multiscale FE model of 5HS weave carbon

fibre fabric composite DCB and 2D plain-strain homogeneous FE model.

88

Figure 3-37. Delaminated areas at the end of each zone of the multiscale FE

model of 5HS weave carbon fibre fabric composite DCB.

89

Figure 3-38. Percentage of released energy by tangential debonding within the total released energy by CZM elements.

3.4.3 Toughening mechanisms

Although sub-surfaces and tangential debonding are indeed acting as toughening

compared to homogeneous model, that amount is not enough to explain the

difference in released energy calculated by MBT and CZM elements.

The toughening mechanism is discussed starting from Zone A to Zone C, which is

the same order as delamination growth. The toughening mechanism observed in

the multiscale FE model can be broken down to two mechanisms. The first

observed mechanism is the relaxation of stress concentration ahead of

delamination front caused by inter-yarns locking with the help of compressive

stresses ahead of delamination in DCB specimen. The deformation and stress

distribution of at Point Ap are shown in Figure 3-39. Unless otherwise noted,

the scaling factor for displaying the deformation is four. The compressive stress

ahead of delamination front, i.e., on weft yarn 2, is observed.

90

Figure 3-39. Contour plot of on the yarns at Point Ap and delaminated

elements coloured in pink.

The yarn interaction is rather easier to see with the contact pressure distribution

on the overlapping areas of yarns as shown in Figure 3-40. It indicates

compressive contact pressure, i.e., , against the yarns as positive value. As

compressive stresses exist in the warp and weft yarns at the ahead of

delamination front as shown in Figure 3-39, the overlapping areas experience

the compressive contact pressure . To understand the contributions from the

weave structure, the inter-yarn locking compressive contact pressure under in-

plane tensile loading at the same points are shown in Figure 3-41.

Delamination

91

Figure 3-40. Contour of contact pressure of the multiscale FE model and

delaminated elements (coloured in pink).

Figure 3-41. Contour of contact pressure of the meso-scale FE model under

in-plane tensile loading.

The meso-scale parts were taken from the multiscale DCB model with the initial

delaminations removed. Constant displacement is applied to the edge normal to

the x-direction. The magnitude of applied displacements is controlled to have

Delamination

92

the same range of contact pressure as that of multiscale DCB model. Figure 3-41

shows the upper warp and weft yarns for 2 plies only. The compressive contact

pressure indicates that inter-yarn locking exists on the overlapping areas where

the warp yarns change their paths to under and over the weft yarns. The

negative contact pressure, which is showing the stress to separate the yarns, is

also observed where the weft yarns are on the warp yarns. Figure 3-42 shows

the contact pressure at overlapping areas on line defined in Figure 3-41.

Contact pressures under two different magnitudes of in-plane tensile loading are

plotted with that of the multiscale DCB model. The contact pressure tendency of

in-plane loading is not very sensitive to the magnitude of applied load.

Accordingly, this tendency itself can be considered remain the same for the

multiscale DCB model at which some amount of exists. With this assumption,

the contact pressure of the multiscale DCB model at overlapping areas seems to

consist of relatively flat compressive pressure inherently caused by the DCB

specimen and pressure variations caused by yarn interactions. This implies that

there is inter-yarn locking effect in the multiscale DCB model ahead of

delamination front.

The effect of yarn interactions on the stress distribution near delamination front

at Point Ap is also investigated. Figure 3-43 shows the view of the upper warp

yarns, weft yarns, and matrix having positive clipped at from bottom

side. It is noted that the matrix near the overlap of warp yarn 3 and weft yarn 2

has positive indicating stress concentration due to delamination front is

slightly relaxed forward.

The stress distributions on lines defined in Figure 3-43 and contact pressures of

contact elements at under in-plane loading are plotted in Figure 3-44. The

minimum value of is observed on warp yarn 3 and its maximum value is

observed on warp yarn 1. The contact pressure, whose sign is reversed for sake

of better understanding, shows inter-yarn locking that may reduce the . The

93

overlap of weft yarn 1 and warp yarn 3 seems to have the highest reduction on

. It is noted, however, that the exact amount of reduction on by yarn

interaction cannot be obtained.

Figure 3-42. Contact pressure distribution on at Point Ap.

94

Figure 3-43. Contour plot of at Point Ap clipped at and delaminated

elements (coloured in pink).

By analyzing the results, it seems that inter-yarn locking caused by in-plane

deformation in the x-direction at the overlap of weft yarn 1 and warp yarn 3 and

the overlap of weft yarn 1 and warp yarn 4 has a toughening effect up to Point

Ap in addition to the toughening effect by inter-yarn locking of weft yarn 2 and

warp yarn 3. The contributions from the two toughening mechanisms cannot be

separated. However, the former toughening effect is smaller than latter because

the former toughening effect vanishes with delamination growth towards Point

Bp where the load reaches a higher value than that at Point Ap. The load drop

toward Point Ab seems be triggered by the delamination front under warp yarn 1

and weft yarn 1 because it has the highest stress compared to the others.

However, the delamination front seems to grow uniformly as shown Figure 3-45.

Delamination

Weft yarn 2

Warp yarn 3

95

Figure 3-44. near delamination front at Point Ap and reversed contact stress

obtained by in-plane loading.

Delamination

96

Figure 3-45. Delamination front development during the load drop from Point Ap

to Ab.

The toughening effect can be clearly explained by the stress distribution near

delamination tip at Point Ap. This approach, however, does not work well with

the case at Point Ab, which is the starting point of toughening toward Point Bp,

Ap

Ab

97

because the delamination has already been branched from the initial

delamination front. This makes the analysis difficult as shown in Figure 3-46.

Figure 3-46. The z-coordinates of delaminated elements showing the branching

at Point Ab.

On the other hand, the positive ahead of delamination front seems be an

alternative measurement of the stress relaxation at the delamination front. The

distance from the maximum delamination front at a given location in the y-

coordinate to the specified magnitude of ahead of the delamination front is

defined as as shown in Figure 3-47. is based on stress, which has been

divided into 9 color increments as shown in Figure 3-47, and represents the

difference in distance between zero stress (color red) and stress of 8MPa (color

blue). Figure 3-47 is at Point Ab where the delamination growth has just arrested.

The longest is obtained on warp yarn 3 up to 7MPa. This warp yarn 3 seems

to prevent delamination by inter-yarn locking with weft yarn 2. for =

1MPa shows the representative trend of over the width of the multiscale

98

DCB model as shown in Figure 3-48 for various points between Point Ab and Bb.

From Point Ab to Bp during which the load is increasing, has its highest values

on warp yarn 3. on warp yarn 3 is shortened during the transition from Point

Bp to Bb. This suggests that the inter-yarn locking resisted during the transition

from Point Ab to Bp, and is released as it goes toward Point Bb. This result seems

to support the toughening caused by the inter-yarn locking of warp yarn 3 and

weft yarn 2.

Figure 3-47. Length of positive from the delamination front at Point Ab.

99

The delaminated CZM elements from Point Ab to Bb are shown in Figure 3-49

with the warp and weft yarns of the upper ply as viewed from the bottom side.

Small amounts of delamination growth mainly on warp yarn 1 are observed

while the load increasing, i.e., Point Ab-Bp. This result suggests that warp yarn 1 is

not acting as the main source of toughening toward Point Bp because there is no

load drop due to the growth. By observing Figure 3-49, the delamination growth

on warp yarn 1 seems be a trigger for the entire delamination growth because

the delamination grows on warp yarn 1 first followed by the rest of yarns. Also,

on warp yarn 1 at Point Bp is relatively smaller than the rest of them as

shown in Figure 3-48. This implies the correlation between the likelihood of

delamination growth and .

At Point Bp- Bb, it is shown that delamination on warp yarn 1 and 2 goes intra-

ply, which is indicated by slightly darker areas because of transparent blue of

weft yarn 2, while the delamination on warp yarn 3 keeps going inter-ply. This

delamination growth pattern can be clearly seen at Point Bb. The exaggeratedly

scaled deformation of yarns at Point Bb is shown in Figure 3-50. The upper and

lower weft yarn 2, coloured in green and yellow, respectively, are bridging the

upper and lower plies. This bridging creates extra delamination surfaces which

are observed by X-rays [23] and are called sub-surfaces.

The inter-yarn locking ahead of delamination front played a significant role on

the toughening up to Point Bp. However, it does not seem to continue toward

Point Cp as weft yarn 3 has no crimping warp yarn. Accordingly, there is no

significant amount of inter-yarn locking effect expected ahead of delamination,

but the effect of weft yarn bridging gradually appeared from Point Bp.

100

Figure 3-48. distribution history from Point Ab to Bb.

The bridging effect reduces the stress concentration ahead of delamination front

if the weft yarns carry enough loads to resist. of the CZM element on the

delamination front edge at Point Bb is shown in Figure 3-51 with the z-coordinate

at the delamination front edge. The corresponding delaminated area is shown in

Figure 3-49, Bb. The values of on the warp yarns are in the range of 60 to

70MPa. This distribution does not seem to show significant amount of stress

relaxation at the delamination front unlike that of Point Ap as shown in Figure

3-44. This confliction indicates that there is no evidence of stress relaxing due to

the weft yarn bridging.

101

Figure 3-49. Delaminated CZM elements from Point Ab to Bb.

Ab Ab- Bp

Bp Bp- Bb Bb

1

2

3

4 2

Numbering

102

Figure 3-50. Weft yarn bridging of Multi2 observed at Point Bb.

Figure 3-51. of CZM element on the delamination front edge with the

delamination front z-coordinates at Point Bb.

The other possible source of toughening is simply due to the creation of sub-

surfaces that increases the delaminated area. To investigate this effect, the

delamination area calculated by the mean delamination length multiplied by the

103

width, i.e., A1, and the total delaminated area of CZM elements, i.e., A2, are

obtained and shown in Figure 3-52 together with their linear curve fittings. The

figure shows that the actual delamination area is around twice that of the

dominant delamination area, i.e., A1. The global energy release rate contributed

by normal debonding of CZM elements can be expressed by

(3.13)

where and is the width of the multiscale DCB

model. The equation states that the global energy release rate due to normal

debonding can be obtained by multiplying the critical energy release rate of

epoxy matrix by the ratio of delamination area increase rate, i.e., the slope, with

respect to the width of the multiscale DCB model.

Figure 3-52. Delamination area of Multi2 versus delamination length.

Accordingly, the energy release rate by normal debonding of all CZM elements

could be 430N/m. The additional energy release rate due to tangential

104

debonding should be added. It has been already shown that the contribution of

tangential debonding to the released energy has up to 14% as discussed in

subsection 3.4.2. If so, the tangential debonding should contribute 70N/m. The

total energy release rate by all CZM elements should be 500N/m. On the other

hand, value of Multi2 calculated by MBT at Point Cp is 486N/m with the

correction. The total energy release rate by CZM elements is only 2.88% higher

than that of Multi2 obtained by MBT.

So far, Multi2 results have been analyzed, but Multi1 results are also available.

value of Multi1 has a similar trend to that of Multi2 up to Point Bb as shown in

Figure 3-34. However, Multi1 seems to have superior toughening towards Point

Cp. The analogous analysis approach is used to obtain delamination area increase

rate of Multi1. The corrected value of energy release rate at Point Cp of Multi1

obtained by MBT is 528N/m. The ratio of delamination area by CZM elements,

which is shown in Figure 3-53, with respect to the width of the multiscale DCB

model is 2.39. Assuming 14% of total energy release rate is contributed by local

tangential debonding, the increased amount of delamination area would raise

the energy release rate to 556N/m, which is 5.30% higher than that obtained by

MBT. Although there is a speculative correction to the end-edge effects, this

good correlation suggests that the main source of toughening seems be the

creation of sub-surfaces .

105

Figure 3-53. Delamination area of Multi1 versus delamination length.

The main difference between Multi1 and Multi2 is the contact stiffness that

connects the meso-scale parts to the homogenized parts. The weft yarn bridging

of Multi1 is shown in Figure 3-54. If the contact stiffness is the cause of this

difference, it implies that the transition from homogeneous parts to meso-scale

parts needs to be improved. Using a finer mesh in homogeneous model and

longer initial delamination within meso-scale parts would eliminate the effect of

the transition on the delamination growth.

106

Figure 3-54. Weft yarn bridging of Multi1 at Point Bb.

In summary, the toughening of 5HS weave fabric composite DCB model consists

of two different mechanisms; 1. Inter-yarn locking ahead of delamination front

causing stress relaxation, 2. Sub-surfaces created via the weft yarn bridging

providing additional resistance. One of the literatures [42] on the experimental

investigations on 5HS weave carbon fabric composite suggested conflicting

concluding remarks, i.e., the cause of toughening is the weft yarn, which is acting

as periodic obstacles for delamination growth. By analyzing the current results,

there is no evidence of weft yarns acting as obstacles. Moreover, the FE model

would likely cause delamination arrest by the weft yarn if they actually acted as

obstacles because delamination growth direction is restricted to the CZM

elements that are inserted. The toughening explanation in [42] is obtained by the

experiments results, which are intermittent data on the side edge of specimen

and delamination surfaces. This limited amount of information may blind other

source of toughening mechanisms.

On the other hand, the experiments indeed showed more variable information,

such as fibre breakage that suggests fibre bridging [22-23] and transverse cracks

within the weft yarn [23]. These types of damage are omitted in the FE model in

107

order to clarify the effect of meso-scale structure of 5HS weave fabric composite.

As the R-curves obtained by the multiscale FE mode are in good agreement with

lower bound of experiment results, adding the above mentioned damage

mechanism may provide a result reaching upper bound of experiment results.

3.5 Discussions

The multiscale 5HS weave carbon fibre fabric composite DCB model was

developed in order to understand the toughening mechanism and to provide

better way of using the R-curves obtained by experiments. The only damage

considered in the model is delamination within pre-defined locations of matrix,

the interface of yarns and matrix, and the interfaces of warp yarns and weft

yarns. The cohesive zone modelling elements are used with a bilinear cohesive

law with material properties of epoxy matrix. Accordingly, there is no

consideration of fibre-matrix interaction, transverse matrix cracks within yarns

or fibre breakages in the model.

R-curves obtained by the multiscale FE models are within the range of

experimental results, but they are closer to the lower bound. This result makes

sense because the multiscale FE model does not consider all possible energy

dissipation mechanisms. Although the multiscale FE models gave a good

agreement with the lower bound, it is not possible to conclude that the FE

models completely capture the delamination growth behaviour of the lower

bound. However, it indeed captures significant amount of toughening caused by

the fabric structure.

The numerical results showed two types of toughening mechanisms. One of

them is inter-yarn locking causing stress relaxation at delamination front. This

toughening mechanism explains why the energy release rate at the initial

delamination length, where no significant amount of yarn and/or fibre bridging

108

exists, is higher than that of epoxy matrix itself. In terms of the use of R-curves

for delamination growth analysis of composite structures, the direct use of R-

curves for the part where out-of-plane tensile stress is applied to propagate

delamination may overestimate the damage tolerance because the toughening is

assisted by the compressive stress that exists ahead of delamination front of DCB

specimen.

The other toughening mechanism is the creation of sub-surfaces via weft yarn

bridging which is observed after the delamination grew over weft yarn 2 entirely.

Since it requires that local delamination must be developed, applying R-curves

value at the plateau region to the model with initial delaminations may severely

overestimate the damage tolerance. One of the examples for this case is to

simulate the effect of imperfections, e.g., leftover film or other contaminations,

unintentionally caused during the manufacturing.

Also, the size of delamination width to which R-curves are applied should be at

least the same as the unit cell size because the size is the minimum requirement

for the complete toughening mechanism to exist.

Lastly, a suggestion could be given to improve the testing method to provide

better input to the FE model. As discussed, inter-yarn locking is enhanced by the

compressive stresses ahead of the delamination tip. This compressive stress is

inherent to the DCB specimen test. Accordingly, this compressive stress ahead of

the delamination tip, having the same degree of toughening as DCB specimen is

not guaranteed to occur in a practical applications. For example, a practical

application made of 5HS weave fabric composite shown in Figure 3-2 b)

experiences out-of-plane stress due to forces transferred by the bracket. Unlike

the DCB specimen, the structure should have more tensile out-of-plane stresses

than compressive stresses. An experimental method for Mode I fracture

toughness without compressive stresses ahead of delamination tip is needed.

109

3.6 Future work

Understanding of toughening mechanism of 5HS weave fabric composites is the

objective of this study in order to have a better idea for R-curve usage for the

delamination growth simulation of composite structures. The results of this

study show two types of toughening mechanisms, but the mixture of the two

toughening mechanisms was not clearly observed due to the limitation on the

problem size. The multiscale 5HS weave fabric composite DCB model with

minimum width of a unit-cell, and larger meso-scale parts should be solved by an

analogous way. Also, it would be interesting to see the effect of various factors,

e.g., voids, weave fabric ply alignment, fibre volume fraction, type of weave

fabric, etc. Even though better understanding of the effect various factors would

bring more information to judge the proper usage of R-curves, all possible

combinations of simulations are not necessary because the experimentally

obtained R-curves have the properties of the weave fabric ply alignment, voids,

fibre volume fraction and type of weave fabric.

From this study, a minimum of unit cell size is recommended for the use of R-

curves. However, this size may be too large to capture local delamination

behaviour for some composite structures. In order to overcome this difficulty,

the R-curves need to be decomposed into a scale that is the same as the weft

and warp yarn width.

As inter-yarn locking is partially assisted by the compressive stress ahead of

delamination front, the effect of magnitude of out-of-plane compressive stress

should also be investigated. By knowing this effect, the R-curves would be

applied to the region where out-of-plane tensile stress is dominant ahead of

delamination front.

It is also worthwhile to focus on the mechanical toughening due to the woven

structure. The analysis shows that the woven structure doubled the fracture

toughness of matrix. In order to reveal the mechanisms of toughening due to the

110

woven structure, a similar procedure to what is provided in this thesis may be

used.

Most importantly, better numerical techniques for modelling delamination

growth are needed to be developed in order to save on computational time and

increase the accuracy of the simulation results.

4 Concluding remarks The industry’s demand for an efficient and accurate numerical simulation

method for analyzing damage behaviour of composite structures is high. This

study contributes to the demand by introducing a new crack modelling method

and by providing more information to understand damage mechanisms of 5HS

weave fabric composite laminates during delamination growth.

In Chapter 2, the proposed crack modelling method, which is called ADD-FEM,

successfully showed the delamination growth simulation capability and accuracy

by a 2D plane strain DCB specimen model. As ADD-FEM does not require the pre-

location of delamination path by inserting interface/contact elements, it is then

very useful. Also, no additional degrees of freedom are necessary to model

newly created delamination surfaces. This is a unique feature of ADD-FEM

compared to other methods that are capable of delamination growth simulation

with comparable accuracy.

In Chapter 3, the damage mechanism of 5HS weave composite was investigated

by applying multi-scale modelling techniques. The DCB model used to investigate

the damage mechanism has meso-scale parts, which consist of individually

modeled weft and warp yarns where the gaps are filled with epoxy matrix. These

meso-scale parts are bonded with homogeneous parts, which are used to model

the region where no damage is expected. To validate the model, R-curves and

Load-displacement curves of experiments and simulations are compared. The

simulation result showed better agreement with lower bound of experimental

111

results than upper bound. The previously undiscovered source of the toughening

mechanism was found by the meso-scale analysis. Although a meso-scale

analysis is not very efficient to model at the structural application level, the

extensive information obtained from the meso-scale analysis at the coupon scale

certainly benefits development of damage analysis at the structure scale.

Although future work is necessary in order to satisfy practical industrial needs,

the study puts forward the development of efficient and accurate damage

prediction capabilities under static loading.

4.1 Contributions of this thesis

The contributions of this thesis are listed as follows.

1) The crack modelling method called ADD-FEM provides an alternative method

for modelling strong discontinuity within an element without additional

degrees-of-freedom.

2) The multiscale analyses of a 5-harness satin weave carbon fibre composite DCB

specimen proved the usefulness of cohesive zone modelling for analyzing

complex crack growth within woven composite materials.

3) A previously unknown toughening mechanism in mode I delamination of 5-

harness satin weave carbon fibre composite was revealed by the multiscale

analyses. The analyses also confirmed that experimentally observed sub-

surfaces are the main source of toughening in mode I delamination.

4) The multiscale analysis results provided insight on the use of R-curves obtained

by mode I DCB tests.

112

4.2 Future work

In order to develop an efficient and accurate damage prediction capability under

static and fatigue loadings, there is still a lot of work to be done. Here is a list of

ideas that could be done as a direct extension of the work presented in this

thesis.

1) 3D formulation of ADD-FEM for a 8-node brick element could be derived. The

3D formulation would be necessary for crack modelling in 3D applications.

2) Formulations of ADD-FEM for higher order elements could be useful, but the

constraint equations for the mid-nodes must be newly developed. These

formulations would extend the usage of ADD-FEM.

3) Formulations of ADD-FEM should be verified with various shapes of sub-

elements, e.g., triangle, hexagonal etc. These formulations would add more

flexibility of the crack growth path.

4) Other types of woven fabric composite and other reinforcing fibres in the woven

fabric composite could be analyzed by a similar multiscale finite element model.

Further analyses on various types of woven composite would confirm whether

or not the toughening mechanism revealed by this study is unique to this

particular composite.

5) The multiscale model used in this study was halved in width and shortened the

length extensively due to the limitations of available computational resources.

The full scale model of the DCB specimen would provide better correlation with

the compatible experimental results.

6) The meso-scale parts of the multiscale model could possibly be used to analyze

mode II and mixed-mode delamination DCB tests.

7) The damage within fibre bundle was ignored in this study due to simplification.

Adding a cohesive zone model within fibre bundle will increase the required

computational resources, but it would also give better representation of actual

woven fibre composite. With the extra cohesive zone model, the model could

be used to analyze other types of tests, e.g., ultimate tensile strength test,

ultimate compressive strength test, etc.

113

References

1. Kensche, C.W., Fatigue of composites for wind turbines. International

Journal of Fatigue, 2006. 28(10): p. 1363-1374.

2. Kong, C., T. Kim, D. Han, and Y. Sugiyama, Investigation of fatigue life for

a medium scale composite wind turbine blade. International Journal of

Fatigue, 2006. 28(10): p. 1382-1388.

3. Shokrieh, M.M. and R. Rafiee, Simulation of fatigue failure in a full

composite wind turbine blade. Composite Structures, 2006. 74(3): p. 332-

342.

4. Marín, J.C., A. Barroso, F. París, and J. Cañas, Study of fatigue damage in

wind turbine blades. Engineering Failure Analysis, 2009. 16(2): p. 656-668.

5. Savage, G., Failure prevention in bonded joints on primary load bearing

structures. Engineering Failure Analysis, 2007. 14(2): p. 321-348.

6. Savage, G., Sub-critical crack growth in highly stressed formula 1 race car

composite suspension components. Engineering Failure Analysis, 2009.

16(2): p. 608-617.

7. Marsh, G., Airframers exploit composites in battle for supremacy.

Reinforced Plastics, 2005. 49(3): p. 26-32.

8. Marsh, G., Boeing's 787: Trials, tribulations, and restoring the dream.

Reinforced Plastics. 53(8): p. 16-21.

9. Msc nastran 2007 r1 implicit nonlinear (sol 600) user's guide: 2009.

10. Cavatorta, M.P., D.S. Paolino, L. Peroni, and M. Rodino, A finite element

simulation and experimental validation of a composite bolted joint loaded

in bending and torsion. Composites Part A: Applied Science and

Manufacturing, 2007. 38(4): p. 1251-1261.

11. Times, T.S. Boeing 787 wing flaw extends inside plane. 2009.

12. Mirzadeh, F. and K.L. Reifsnider, Micro-deformations in c3000/pmr15

woven composite. Journal of Composite Materials, 1992. 26(2): p. 185-

205.

114

13. ANSYS, I., Ansys help. 2009.

14. Wimmer, G., W. Kitzmüller, G. Pinter, T. Wettemann, and H.E.

Pettermann, Computational and experimental investigation of

delamination in l-shaped laminated composite components. Engineering

Fracture Mechanics, 2009. 76(18): p. 2810-2820.

15. Riccio, A., M. Giordano, and M. Zarrelli, A linear numerical approach to

simulate the delamination growth initiation in stiffened composite panels.

Journal of Composite Materials, 2010. 44(15): p. 1841-1866.

16. Pietropaoli, E. and A. Riccio, On the robustness of finite element

procedures based on virtual crack closure technique and fail release

approach for delamination growth phenomena. Definition and

assessment of a novel methodology. Composites Science and Technology,

2010. 70(8): p. 1288-1300.

17. Fish, J. and Q. Yu, Computational mechanics of fatigue and life predictions

for composite materials and structures. Computer Methods in Applied

Mechanics and Engineering, 2002. 191(43): p. 4827-4849.

18. Muñoz, J.J., U. Galvanetto, and P. Robinson, On the numerical simulation

of fatigue driven delamination with interface elements. International

Journal of Fatigue, 2006. 28(10): p. 1136-1146.

19. Turon, A., J. Costa, P.P. Camanho, and C.G. Dávila, Simulation of

delamination in composites under high-cycle fatigue. Composites Part A:

Applied Science and Manufacturing, 2007. 38(11): p. 2270-2282.

20. Harper, P.W. and S.R. Hallett, A fatigue degradation law for cohesive

interface elements - development and application to composite materials.

International Journal of Fatigue, 2010. 32(11): p. 1774-1787.

21. May, M. and S.R. Hallett, A combined model for initiation and

propagation of damage under fatigue loading for cohesive interface

elements. Composites Part A: Applied Science and Manufacturing, 2010.

41(12): p. 1787-1796

115

22. Feret, V., Development of a mixed-mode fracture criterion for a fabric

composite manufactured by rtm, master of engineering thesis, in

Department of Mechanical Engineering. 2009, McGill University:

Montreal.

23. Gill, A.F., P. Robinson, and S. Pinho, Effect of variation in fibre volume

fraction on modes i and ii delamination behaviour of 5hs woven

composites manufactured by rtm. Composites Science and Technology,

2009. 69(14): p. 2368-2375.

24. Turon, A., C.G. Dávila, P.P. Camanho, and J. Costa, An engineering

solution for mesh size effects in the simulation of delamination using

cohesive zone models. Engineering Fracture Mechanics, 2007. 74(10): p.

1665-1682.

25. Belytschko, T. and T. Black, Elastic crack growth in finite elements with

minimal remeshing. International Journal for Numerical Methods in

Engineering, 1999. 45(5): p. 601-620.

26. Moës, N., J. Dolbow, and T. Belytschko, A finite element method for crack

growth without remeshing. International Journal for Numerical Methods

in Engineering, 1999. 46(1): p. 131-150.

27. Oliver, J., Modelling strong discontinuities in solid mechanics via strain

softening constitutive equations. Part 1: Fundamentals. International

Journal for Numerical Methods in Engineering, 1996. 39(21): p. 3575-

3600.

28. Oliver, J., Modelling strong discontinuities in solid mechanics via strain

softening constitutive equations. Part 2: Numerical simulation.

International Journal for Numerical Methods in Engineering, 1996.

39(21): p. 3601-3623.

29. Linder, C. and F. Armero, Finite elements with embedded strong

discontinuities for the modeling of failure in solids. International Journal

for Numerical Methods in Engineering, 2007. 72(12): p. 1391-1433.

116

30. Liu, H., X.-L. Zhao, and R. Al-Mahaidi, Boundary element analysis of cfrp

reinforced steel plates. Composite Structures, 2009. 91(1): p. 74-83.

31. Mukherjee, S. and M. Morjaria, On the efficiency and accuracy of the

boundary element method and the finite element method. International

Journal for Numerical Methods in Engineering, 1984. 20(3): p. 515-522.

32. Bouchard, P.O., F. Bay, Y. Chastel, and I. Tovena, Crack propagation

modelling using an advanced remeshing technique. Computer Methods in

Applied Mechanics and Engineering, 2000. 189(3): p. 723-742.

33. Sheng Liu, Z. Kutlu, and F.-K. Chang, Matrix cracking and delamination in

laminated composite beams subjected to a transverse concentrated line

load. Journal of Composite Materials, 1993. 27(5): p. 436-470.

34. Rybicki, E.F. and M.F. Kanninen, A finite element calculation of stress

intensity factors by a modified crack closure integral. Engineering

Fracture Mechanics, 1977. 9(4): p. 931-938.

35. R. D. Cook, D.S.M., M. E. Plesha, Concepts and applications of finite

element analysis, third edition. 1989, New York: John Wiley & Sons.

36. Shephard, M.S., Linear multipoint constraints applied via transformation

as part of a direct stiffness assembly process. International Journal for

Numerical Methods in Engineering, 1984. 20(11): p. 2107-2112.

37. Sadd, M.H., Elasticity: Theory, applications, and numerics. 2005,

Burlington: Elesvier Butterworth-Heinemann.

38. Anderson, T.L., Fracture mechanics: Fundamentals and applications,

second edition. 1994, Boca Raton: CRC Press LLC.

39. Raju, I.S. and K.N. Shivakumar, An equivalent domain integral method in

the two-dimensional analysis of mixed mode crack problems. Engineering

Fracture Mechanics, 1990. 37(4): p. 707-725.

40. Chen, J.C., M. Kinloch, A.J. Busso, E.P. Matthews, F.L. Qiu, Y., Predicting

progressive delamination of composite material specimens via interface

117

elements. Mechanics of Composite Materials and Structures, 1999. 6: p.

301-317.

41. Roy, S., Mechanical modeeling and teting of a composite helicopter

structure made by resin transfer moulding, master of engineering thesis,

in Department of Mechanical Engineering. 2008, McGill University:

Montreal.

42. Alif, N., L.A. Carlsson, and L. Boogh, The effect of weave pattern and crack

propagation direction on mode i delamination resistance of woven glass

and carbon composites. Composites Part B: Engineering, 1998. 29(5): p.

603-611.

43. Yurgartis, S.W. and J.P. Maurer, Modelling weave and stacking

configuration effects on interlaminar shear stresses in fabric laminates.

Composites, 1993. 24(8): p. 651-658.

44. Bigaud, D. and P. Hamelin, Stiffness and failure modelling of 2d and 3d

textile-reinforced composites by means of imbricate-type elements

approaches. Computers & Structures, 2002. 80(27-30): p. 2253-2264.

45. Lee, C.S., S.W. Chung, H. Shin, and S.J. Kim, Virtual material

characterization of 3d orthogonal woven composite materials by large-

scale computing. Journal of Composite Materials, 2005. 39(10): p. 851-

863.

46. Iarve, E.V., D.H. Mollenhauer, E.G. Zhou, T. Breitzman, and T.J. Whitney,

Independent mesh method-based prediction of local and volume average

fields in textile composites. Composites Part A: Applied Science and

Manufacturing. 2009. 40(12): p.1880-1890.

47. Whitcomb, J.D., C.D. Chapman, and X. Tang, Derivation of boundary

conditions for micromechanics analyses of plain and satin weave

composites. Journal of Composite Materials, 2000. 34(9): p. 724-747.

118

48. Tang, X. and J.D. Whitcomb, General techniques for exploiting periodicity

and symmetries in micromechanics analysis of textile composites. Journal

of Composite Materials, 2003. 37(13): p. 1167-1189.

49. Xia, Z., Y. Zhang, and F. Ellyin, A unified periodical boundary conditions for

representative volume elements of composites and applications.

International Journal of Solids and Structures, 2003. 40(8): p. 1907-1921.

50. Xia, Z., C. Zhou, Q. Yong, and X. Wang, On selection of repeated unit cell

model and application of unified periodic boundary conditions in micro-

mechanical analysis of composites. International Journal of Solids and

Structures, 2006. 43(2): p. 266-278.

51. Woo, K. and J.D. Whitcomb, Effects of fiber tow misalignment on the

engineering properties of plain weave textile composites. Composite

Structures. 37(3-4): p. 343-355.

52. Karkkainen, R.L. and B.V. Sankar, A direct micromechanics method for

analysis of failure initiation of plain weave textile composites. Composites

Science and Technology, 2006. 66(1): p. 137-150.

53. Daggumati, S., W. Van Paepegem, J. Degrieck, J. Xu, S.V. Lomov, and I.

Verpoest, Local damage in a 5-harness satin weave composite under

static tension: Part ii - meso-fe modelling. Composites Science and

Technology, 2010. 70(13): p. 1934-1941.

54. Zako, M., Y. Uetsuji, and T. Kurashiki, Finite element analysis of damaged

woven fabric composite materials. Composites Science and Technology.

63(3-4): p. 507-516.

55. Tang, X. and J.D. Whitcomb, Progressive failure behaviors of 2d woven

composites. Journal of Composite Materials, 2003. 37(14): p. 1239-1259.

56. Lomov, S.V., D.S. Ivanov, I. Verpoest, M. Zako, T. Kurashiki, H. Nakai, and

S. Hirosawa, Meso-fe modelling of textile composites: Road map, data

flow and algorithms. Composites Science and Technology, 2007. 67(9): p.

1870-1891.

119

57. Key, C.T., S.C. Schumacher, and A.C. Hansen, Progressive failure modeling

of woven fabric composite materials using multicontinuum theory.

Composites Part B: Engineering, 2007. 38(2): p. 247-257.

58. Gorbatikh, L., D. Ivanov, S. Lomov, and I. Verpoest, On modelling of

damage evolution in textile composites on meso-level via property

degradation approach. Composites Part A: Applied Science and

Manufacturing, 2007. 38(12): p. 2433-2442.

59. Smilauer, V.t., C.G. Hoover, Z. P. Bazant, F.C. Caner, A. M. Waas, and K.W.

Shahwan, Multiscale simulation of fracture of braided composites via

repetitive unit cells. Engineering Fracture Mechanics. 2011. 78(6): p. 901-

918.

60. ASTM, D5528 standard test method for mode i interlaminar fracture

toughness of unidirectional fiber-reinforced polymer matrix composites.

2002.

61. Textile Composites Research, U.o.N.

Http://texgen.Sourceforge.Net/index.Php/main_page. 2010.

62. Composite Materials Group, K.U.L.

Http://sirius.Mtm.Kuleuven.Be/research/c2/poly/software.Html. 2010.

63. CYTEC, Cycom 890 rtm epoxy system. 2002.

64. Kaw, A.K., Mechanics of composite materials. 1997: CRC Press.

120

Appendix A

The selection of direction used in Eq. (2.5) requires a certain condition to

be met in order to obtain the constraint equations. , which is defined in

Eq. (2.7), is the derivative of shape functions with respect to which is

defined in the global coordinate system. Recall

(1.A)

According to the equation above, it is required that has a non-zero

value. Since the component of is the derivative of shape functions with

respect to , the first step is to obtain the derivative of shape functions with

respect to and , i.e.,

where

(2.A)

The derivative of the shape function with respect to is then given by

(3.A)

where is the unit vector in direction.

Then, is given by

121

(4.A)

where and are the components of unit vector in and ,

respectively. It should be noted that and are substituted into

only the derivatives of shape function in the right hand side of Eq. (4.A). The

condition to have can be obtained by

(5.A)

Therefore, the direction has to be chosen in such a way that

(6.A)

The identical condition can be obtained for

.

122

Appendix B

By using the assumption expressed by Eq. (2.5),

i.e.,

, and isolating gives

.

(1.B)

By using the assumption expressed by Eq. (2.5),

i.e.,

, and isolating gives

(2.B)

where is the component of the derivative operator matrix obtained for

the location of node 2 defined by Figure 2-3.

By using the assumption expressed by Eq. Figure 2-5,

i.e.,

, and isolating gives

(3.B)