FAILURE OF COMPOSITE MATERIALS

BY KINK BAND FORMATION

Simon Peter Hald Skovsgård

Department of Engineering
Aarhus University

This dissertation is submitted for the degree of

PhD

April, 2019
 Dissertation title: Failure of composite materials by kink band formation

Dissertation submitted: April 12, 2019

for the degree of PhD

Author: MSc Simon Peter Hald Skovsgård

Department of Engineering
Aarhus University, Denmark
E-mail: [email protected]

Supervisor: Professor, dr. techn. Henrik Myhre Jensen

Department of Engineering
Aarhus University, Denmark
E-mail: [email protected]

© Copyright by Simon Peter Hald Skovsgård

Typeset in LATEX 2ε
Abstract

Fibre-reinforced composites and layered materials have a heterogeneous composition of two

or more constituents and have anisotropic material behaviour. This heterogeneous structure
enables numerous failure mechanisms. One of the dominant compressive failure mechanisms
of the latter composites is composed of so-called kink bands. These are bands of material
where the fibres inside the band have rotated relative to the fibres outside.
This dissertation mainly concerns the examination of kink band formation in fibre com-
posites and layered materials using several different methodologies. Five publications on
the latter topic have arisen based on the work done in this project. The four introductory
chapters in the dissertation are devoted to explanations and elaborations of the methods used
and developed during the project. Several finite element models are developed specifically to
examine either the peak strength of the composite or the post-buckling kink band behaviour.
Two novel constitutive formulations are developed that can be used in a general framework
to investigate the elastic-plastic behaviour of composites, including the study of kink bands.
The following results and conclusions are drawn from the project results. Compressive
kinking strength is affected by:

• Fibre-misalignments
• Insufficient fibre-to-matrix bonding
• Matrix yield strength
• Multi-axial loading

The stress at steady-state kink band broadening can be seen as the lower bound of the
compressive strength and is lowered by:

• Increasing band inclinations to a certain limit

• Decreasing matrix Poisson’s ratio
• Decreasing maximum tangent modulus of the matrix

As part of my visit to the Department of Engineering at the University of Cambridge,

contributions are made to the field of joint mechanics, particular regarding composites made
from ultra-high molecular weight polyethylene. The latter study is a secondary focus of
this dissertation. The research questions, problems, methodologies, results and conclusions
appear in an attached article.
Resumé

Fiberforstærkede kompositter og lagdelte materialer har en heterogen sammensætning af to

eller flere bestanddele og har en anisotropisk materialeopførsel. Denne heterogene struktur
muliggør adskillige fejlmekanismer. En af de dominerende fejlmekanismer i kompression for
sidstnævnte kompositter er ved såkaldte kinkbånd. Disse er bånd af materiale, hvor fibrene
inde i båndet har roteret relativt i forhold til fibrene udenfor.
Denne afhandling vedrører primært undersøgelsen af kinkbåndsformation i fiberkomposit-
ter og lagdelte materialer ved anvendelse af flere forskellige metoder. Fem publikationer om
sidstnævnte emne er opstået på baggrund af arbejdet i projektet. De fire indledende kapitler i
afhandlingen er afsat til forklaringer og uddybninger af de anvendte og udviklede metoder
under projektet. Der er udviklet adskillige finite element modeller, der er specielt designet til
at undersøge enten topstyrken eller post-buckling adfærden af kompositter. To nye konstitu-
tive formuleringer er udviklet, der kan anvendes til generelle formål ifm. undersøgelsen af
kompositters elastisk-plastiske adfærd, heriblandt undersøgelsen af kinkbånd.
Følgende resultater og konklusioner blev draget gennem projektet. Den kompressive
kinkbåndsstyrke påvirkes af:

• Vinkelimperfektioner af fibrene
• Utilstrækkelig binding mellem fiber og matrix
• Faldende matrix flydestyrke
• Multiaksial belastning

Spændingen ved kinkbåndsudvidelse kan ses som den nedre grænse af kompressionsstyrken
og sænkes ved:

• Stigende båndhældninger til en vis grænse

• Faldende matrix Poissons forhold
• Faldende maksimal tangentmodul for matrixen

Som en del af besøget ved ingeniørafdelingen ved Cambridge universitet blev der udført
forskning indenfor mekanikken for samlinger lavet af kompositter lavet af polyethylen med
ultrahøj molekylevægt. Sidstnævnte studie er et sekundært fokus ved denne afhandling.
Forskningsspørgsmålene, problemerne, metoderne, resultaterne og konklusionerne fremgår i
en vedhæftet artikel.
Preface

This dissertation is submitted in partial fulfilment of the requirements for the degree of
PhD from Aarhus University (AU). The research herein was conducted at the Department of
Engineering in the Section of Mechanical Engineering, at AU in the period of 1st May 2016 to
12th April 2019 under the supervision of Professor Henrik Myhre Jensen. The dissertation is
written as a collection of articles according to the guidelines within the Section of Mechanical
Engineering. Furthermore, the dissertation follows the official rules and regulations stated by
the Graduate School of Science and Technology (GSST) at AU. The work was funded by the
Thomas B. Thriges Fond, which is greatly appreciated.

Declaration
I hereby declare that except where specific reference is made to the work of others, the
contents of this dissertation are original and have not been submitted in whole or in part for
consideration for any other degree or qualification.

Acknowledgements
It has been a great journey and privilege to study at Aarhus University during my PhD.
During my time as a student, I have worked with and received support from several people,
to whom I am deeply grateful. First and foremost, I am grateful to my supervisor Prof Henrik
M. Jensen for considering me for the PhD project. I owe him deep gratitude for his guidance
and help through the years. His inspiring and calm method of supervision has inspired me to
strive for new discoveries and he was one of the main reasons why I looked forward to work
on early Monday mornings.
My fellow PhD students and other colleagues in the office have always been a source of
happiness and joy. I have benefited a lot from their wisdom through valuable discussions. I
acknowledge my lovely colleagues Kasper Ringgard, Joakim Vester-Petersen, Dan Thomsen,
Oliver Tierdad Filsoof, Emil Madsen and Frederik Foldager for making my time enjoyable. I
x Preface

am grateful to my colleague and friend Simon Heide-Jørgensen for several discussions about
mechanics of materials. I would like to thank Vedad Tojaga, who was a master’s student in
2017, for a great collaboration on the simulations of kink bands in open-hole composites. In
addition, I must thank Jens Lycke Wind, a previous PhD student who worked on kink bands,
I have benefited a lot from the foundation that he established.
Furthermore, I am grateful to Professor Norman Fleck for hosting me as a visiting PhD
student at the University of Cambridge. I learned a lot from him, including the need to be
cautious about my assumptions and conclusions and be precise in the dissemination of my
research. I would like to thank all the lovely people in the Oatley Lab for making my stay
enjoyable.
Last but not least, I would like to thank my family and friends for their love and support.
I am especially grateful to my girlfriend, Katrine, who has given me unconditional love and
support and had to do without me during my long stay in Cambridge. She listened patiently
to many of my research frustrations throughout my time, although they probably did not
make much sense.
To each and every one of you, thank you.

Simon Peter Hald Skovsgård

April, 2019
Table of contents

Abstract v

Resumé vii

Publications and conference presentations xiii

Nomenclature xv

1 Introduction 1
1.1 Compressive failure modes . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Kink band formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Scope of work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Finite element models 13

2.1 Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Imperfection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Semi-analytical kink band model 25

3.1 General relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Homogenisation of constituents . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Kink band bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Kink band model and numerical scheme . . . . . . . . . . . . . . . . . . . 43
3.5 Kink band broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
xii Table of contents

4 Analytical kink band model 53

4.1 Stresses and equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Strains and continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Kink band broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 Summary and outlook 59

5.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Recommendations for further work . . . . . . . . . . . . . . . . . . . . . . 61

6 P1 63

7 P2 79

8 P3 93

9 P4 107

10 P5 119

11 C1 131

Appendix A Jaumann rate of Cauchy stresses 141

Appendix B Nominal stresses 145

Appendix C Constitutive formulation 151

Bibliography 159
Publications

[P1] Skovsgaard S.P.H., Jensen H.M., 2018. Three-dimensional constitutive model for
elastic-plastic behaviour of fibre-reinforced composites. International Journal of
Solids and Structures. Vol. 139-140, p. 150-162

[P2] Skovsgaard S.P.H., Jensen H.M., 2018. Constitutive model for imperfectly bonded
fibre-reinforced composites. Composite Structures. Vol. 192, p. 82-92

[P3] Skovsgaard S.P.H., Jensen H.M., 2018. Steady-State Kink Band Propagation in
Layered Materials. Journal of Applied Mechanics. Vol. 85 / 061005-1

[P4] Skovsgaard S.P.H., Jensen H.M., 2019. A general approach for the study of kink
band broadening in fibre composites and layered materials. European Journal of
Mechanics / A Solids. Vol. 74, p. 394-402

[P5] Skovsgaard S.P.H., Fleck, N.A., 2019. Load transfer at a bolted joint of a laminate
made from ultra high molecular-weight polyethylene fibres. Under review at the
International Journal of Solids and Structures

Contribution to other work:

[C1] Tojaga V., Skovsgaard S.P.H., Jensen H.M., 2018. Micromechanics of kink band
formation in open-hole fibre composites under compressive loading. Composites
Part B. Vol. 149, p. 66-73
xiv Publications

Oral conference presentations:

I DCAMM 16th Internal Symposium, February 2017 Middelfart, Denmark

Title: Modelling of failure in fiber composite materials

II 14th U.S. National Congress for Computational Mechanics, July 2017 Montreal,
Canada
Title: Kink band instabilities in layered media

III 2017 Society of Engineering Science Conference, July 2017 Boston, Massachusetts,
USA
Title: Instability in fiber composites by kink band formation

IV 18th U.S. National Congress for Theoretical and Applied Mechanics, June 2018
Chicago, Illinois, USA
Title: Kink band instability in fiber composites

V 21st International Conference on Composite Structures, September 2018 Bologna,

Italy
Title: Steady-state kink band broadening in fibre-reinforced composites and layered
materials

VI 2018 Society of Engineering Science Conference, October 2018 Madrid, Spain

Title: Kink band failure in fibre composites and layered materials

VII DCAMM 17th Internal Symposium, March 2019 Børkop, Denmark

Title: Compressive failure of fibre-reinforced and layered composites initiated by
the development ok kink bands
Nomenclature

Roman Symbols

A Area

a Wave amplitude

b Kink band width

c Volume fraction

Ci jkl Tensor of incremental nominal moduli

E Young’s modulus

fα Arbitrary function describing a vector field

Fi j Deformation gradient tensor

Fi Force vector

G Shear modulus

Hi j Inverse of help factor Mi j

J Jacobian determinant, volume ratio

J2 Second invariant of the stress deviator tensor

L0 Initial geometry length

Li jkl Incremental stiffness tensor

Mi j Help factor used in homogenisation formulations

Ni Surface normal in initial state

xvi Nomenclature

ni Surface normal in deformed state

Ri j Directional cosines

Eαβ Green-Lagrange strain tensor

Sαβ Second Piola-Kirchhoff stress tensor

Si Surface vector in initial state

si Surface vector in deformed state

t Time

t0 Initial geometry thickness

ti j Nominal stress tensor

Ti Traction vector in initial state

ti Traction vector in deformed state

V Volume

vi, j Velocity gradients

vi Velocity vector

WE External work per unit volume

WI Internal work per unit volume

w0 Initial geometry width

wi, j Velocity gradients, stationary coordinate system

u Displacement

ui Displacements tensor

Xi Initial coordinates of spatial field

xi Deformed coordinates of spatial field

Greek Symbols

α Fibre misalignments outside band

Nomenclature xvii

β Kink band orientation

δi j Kronecker delta

ε̇i j Strain increment tensor

λ Wavelength

κ Fibre-matrix dependent bonding factor

µ Fibre-matrix bonding factor

ν Poisson’s ratio

φ Fibre misalignment

φ0 Largest initial imperfection angle

ρ Density

σ Stress

σi j Cauchy stress tensor

τi j Kirchhoff stress tensor

ωi j Spin tensor

Superscripts

(•)−1 Inverse

(•)cr Critical value

(•)c Arbitrary constituent

(•)c Common material state

(•) f Fibre constituent

(•)i Inner material state

(•)m Matrix constituent

(•)o Outer material state

(•)ss Steady-state value

xviii Nomenclature

(•)y Yield value

Subscripts

(•)0 Quantity associated with initial state

(•)α Greek indices that take values 1 or 2

(•)i Latin indices that take values 1,2 or 3

(•),i Comma notation denotes partial derivative with respect to xi

Other Symbols

det(•) Determinant

d(•) Infinitesimal piece

˚
(•) Co-rotational Jaumann rate
˙
(•) Material time derivative

d/dt(•) Material time derivative

Acronyms / Abbreviations

2D 2 (Two) Dimensional

3D 3 (Three) Dimensional

FE Finite Element

PBC Periodic Boundary Conditions

Q4 Bilinear Rectangular element

Q8 Biquadratic Rectangular element

RVE Representative Volume Element

UHMWPE Ultra-High Molecular Weight PolyEthylene

Chapter 1

Introduction

Throughout human evolution,humans have striven to find the ideal materials for a given
application. During the prehistoric periods, the Stone Age, Bronze Age and Iron Age, humans
developed tools made from different materials to achieve desired properties such as increased
stiffness and strength.
In the modern world of structural mechanics efforts to find and develop new materials is
still ongoing and particularly materials with low density and high strength are desirable in
many applications. One way to achieve these properties is to make a composition of different
constituent materials, leading to the material term, composite materials. A composite material
is composed of two or more constituents on the macroscopic level i.e. the constituents can be
observed with the naked eye.
Besides high stiffness and low density it is possible to obtain other tailored properties that
are different from the properties of the individual constituents in their bulk form. Jones (1975)
identified some of the properties that can be improved: strength, stiffness, corrosion resistance,
wear resistance, attractiveness, weight, fatigue life, temperature-dependent behaviour, thermal
insulation, thermal conductivity and acoustical insulation.
One of the earliest uses of composite materials made by humans was wattle and daub,
which is a woven lattice of wooden strips embedded in clay (Shaffer, 2006). Wattle and
daub were initially used in the Neolithic era i.e. the late Stone Age. When talking about
composites today, concrete, glass- and fibre-reinforced polymer composites are among the
most widely known composite materials. Fibre-reinforced composites find wide application
and examples of industries where composites materials are used are shown in Fig. 1.1.
In the 1970s, fibre composites were used in aerospace and defence-related industries due
to their superior properties. Like many other technologies, the use of fibre composites in
everyday products was initially limited due to price. Today, prices are reduced and composites
2 Chapter 1 - Introduction

(a) (b)

(c) (d)

(e) (f)

Figure. 1.1 Examples of industries where composites are used. (a) TAHOE 700 Sport boat (b)
Bulletproof vest (c) F-35 Lightning II combat aircraft (d) Ferrari’s Formula one car 2018 (e) the
International Space Station (ISS) (f) Vestas 8 MW Wind turbine
 3

$
106 lb
lb

150 50

Price
Consumption

75 25

5
0
1970 1985 1990 2004 2008

Rocket & Sporting Industrial Automotive Wind Energy

Aerospace goods Fuel cell
Figure. 1.2 Consumption of composites in million of pounds versus time, and price of composites in
dollars per pounds versus time. Shown in a reconstructed edition. Source: The University of Tennessee
Space Institute (https://www.utsi.edu/)

are increasingly used in the automotive, construction and sport-industries, as indicated in

Fig. 1.2, taken from the University of Tennessee Space Institute. The price of composites
is still relatively high due to the complexity of producing these materials e.g. laminated
structures composed of fibre plies are still mainly hand-laid.

Composite materials are divided into four categories:

• Fibre composites
• Laminated composites
• Particulate composites
• Combinations of the above
Fibre composites, as the name suggests, are composed of fibres embedded in a resin material
also known as a matrix. Laminated composites are materials composed of two or more
layers of material and can be composed of layers of the same material where the individual
layers have anisotropic properties e.g. carbon fibre laminated composites. A sandwich
structure is another example of a laminated composite. Particulate composites are composed
of particles of one or more materials embedded in a matrix material e.g. concrete or rocket
4 Chapter 1 - Introduction

propellants consisting of aluminium powder in polyurethane. The current dissertation focuses

on fibre composites and laminated composites (also known as layered materials).
As mentioned earlier, fibre-reinforced composites are widely used due to their high
strength and stiffness together with their low density. Most commonly, stiff and strong fibres
are embedded in a soft and ductile matrix material. Fibres in one layer are primarily oriented
in a given direction e.g. uni-directional fibre composites. In these materials, high stiffness is
only achieved in the direction of the fibres. Moreover, the high strength is mainly in tension
in the direction of the fibres. The strength in compression is often only 60% of the strength
in tension (Fleck, 1997). Many structures are prone to bending where the structure, and
therefore the material, is exposed to both tension and compression e.g. a wind turbine blade
that acts like a cantilever beam. Therefore, understanding how fibre-composites react to
compression is of great importance. The study of fibre composites on the micro-mechanical
level is rather complex due to the heterogeneous structure leading to anisotropic material
behaviour. The heterogeneities can lead to several different compressive failure mechanisms,
presented in the subsequent section.

1.1 Compressive failure modes

Compressive failure can occur due to the initiation of different mechanisms. Rosen (1965)
was one of the first to investigate the mechanism of compressive failure in fibre composites.
Fleck (1997) provided a thorough review of the different failure modes of fibre composites
prone to compressive loading in the direction of the fibres. In his work, he distinguishes
between six failure modes as presented in Fig. 1.3 and as described below:
a) Elastic micro-buckling is a shear buckling instability initially studied by Rosen
(1965).
b) Plastic micro-buckling or kink band formation is a plastic shear instability that occurs
with large shearing of the matrix and was initially studied by Argon (1972).
c) Fibre crushing is a failure of the fibres prior to matrix failure and can occur due to
several mechanisms, e.g. fibre yielding, longitudinal splitting and microscopic fibre
kinking.
d) Splitting is a fracture failure of the matrix along the fibres similar to the tensile mode
I cracking and was studied by Sammis and Ashby (1986).
e) Buckle-delamination is an out-of-plane buckling mode that is seen when a surface
layer has insufficient bonding and was studied by Hutchinson and Suo (1991).
f) Shear band formation is yielding of the matrix that occurs in a band oriented at
approximately at 45◦ with respect to the loading axis.
1.1 Compressive failure modes 5

σ σ σ

Single
fibre

σ σ σ

(a) (b) (c)

σ σ σ

Shear
planes

σ σ σ

(d) (e) (f)

Figure. 1.3 Failure modes in fibre composites under compressive loading. Shown in a reconstructed
edition (Fleck, 1997). (a) Elastic micro-buckling (b) Plastic micro-buckling (c) Fibre crushing (d)
Splitting (e) Buckle-delamination (f) Shear band formation
6 Chapter 1 - Introduction

5 T800/924c

micro-buckling
T800/924c part cured
4
Fibre AS4/PEEK
crushing EXAS HSI/DX6002
AS4/PEEK (T = 77◦ C)
Elastic

τy 3 AS4/PEEK (T = 132◦ C)
φ0 Plastic Graphite/epoxy
[GPa]
micro-buckling Carbon/carbon
2

2 4 6 8 10
G [GPa]

Figure. 1.4 Compressive failure map showing three competing modes of failure. The failure map
has the axis of composite shear modulus G and the ratio between in-plane shear strength τ y and fibre
misalignment φ0 . Shown in a reconstructed edition (Fleck, 1997)

Fleck (1997) created several failure maps with appropriate axes to graphically observe the
competition between the different failure modes by using simple upper bound solutions for
the compressive failures. One of the failure maps constructed by Fleck (1997) is presented in
Fig. 1.4. The map has the in-plane composite shear modulus G as the first axis and the second
axis is the ratio between in-plane shear strength τ y and fibre misalignment φ0 . The two-
dimensional failure map distinguishes between three final failures: elastic micro-buckling,
plastic micro-buckling and fibre crushing. Material data for eight different carbon fibre-
reinforced composites is included in the map and for these materials, plastic micro-buckling
or kink band formation is the dominant mode of failure in compression. The following
section will further clarify the mechanisms of kink band formation.

1.2 Kink band formation

Kink band failure is also known as plastic micro-buckling or fibre kinking. In the current
dissertation, the failure will mainly be referred to as kink band formation. Two experiments
showing the formation of kink bands are shown in Fig. 1.5 for two different materials. Vogler
and Kyriakides (1997) considered the initiation and propagation of kink bands in a carbon
1.2 Kink band formation 7

(a) (b)

Figure. 1.5 Kink band formation in two different fibre composites. (a) AS4/PEEK composite (Vogler
and Kyriakides, 1997). (b) Cu/Nb nanolaminate composite (Nizolek et al., 2017).

fibre-reinforced PEEK composite shown in Fig. 1.5a. At the former stage, the kink band
width is approximately b ≈ 2000 µm: see Fig. 1.6. In an experiment conducted by Nizolek
et al. (2017) on a Cu/Nb nanolaminate composite, the kink band width is approximately
b ≈ 50 µm (Fig. 1.5b). The two orders of magnitude difference in kink band width is due
to the difference in fibre and layer sizes. Based on experimental findings (e.g. Fig. 1.5), an
idealised kink band geometry can be introduced, such as the one demonstrated in Fig. 1.6.
The idealised kink band geometry is composed of two material states: a kinked state and an
unkinked state. The kinked material, which is localised within a narrow band of width b,
has rotated an amount of φ relative to the base material. The localisation band is oriented at
angle β with respect to the x2 -axis, as presented in the figure. The material state within the
kink band has undergone large shear straining.
One of the studies earliest studies in which kink band formation was observed was con-
ducted by Paterson and Weiss (1966) on a phyllite that is a type of foliated metamorphic rock.
In his early work, Rosen (1965) derived an analytical expression based on the compressive
bifurcation load of beams surrounded by an elastic matrix. He suggested two bifurcation
modes: a tension and a shear mode. The shear mode gave the lowest compressive strength
cr at bifurcation and was equal to
σ11
cr
σ11 =G (1.1)

where G is the in-plane elastic shear modulus of the composite. The estimate by Rosen
(1965) overpredicts the compressive strength (Lager and June, 1969). Experiments show
8 Chapter 1 - Introduction

b u

σ11
σ11
β
x2
φ

x1 L0

Figure. 1.6 Idealised kink band geometry

compressive strengths down to 1/4 of the prediction by Rosen (1965). Later, Argon (1972)
suggested an upper bound solution concerning a composite with rigid plastic properties
in shear with a shear yield strength of τ y . He showed that the compressive strength was
influenced by initial fibre misalignments equal to

cr τy
σ11 = (1.2)
φ0

The estimate by Argon (1972) explains the knock-down in compressive strength compared
to the estimate by Rosen (1965). The two estimates by Rosen (1965) and Argon (1972)
given by Eqs. (1.1) and (1.2) were used by Fleck (1997) to create the two-dimensional failure
map presented in Fig. 1.4. Several other attempts have been made to develop analytical
expressions of the compressive strength of fibre composites and layered materials due to
kink band formation. Budiansky (1983) investigated the elastic-plastic shear behaviour of a
composite with rigid fibres. Fleck and Budiansky (1991) examined the influence of shear
stresses and later, Slaughter et al. (1993) considered transverse stresses. In all of the above
analytical expressions the kink band angle β is assumed to be zero. Christoffersen and Jensen
(1996) provided a method for finding the compressive bifurcation stress of a composite with
an arbitrary non-linear composite behaviour where the constituents can be described by two
independent constitutive relations. Furthermore, the formulation could take nonzero values
of the kink band angle β . Christoffersen and Jensen (1996) derived an analytical expression
of the bifurcation strength in the case of rigid fibres and found that the lowest compressive
strength was achieved with a kink band angle β = 0.
1.2 Kink band formation 9

1
2

Figure. 1.7 Shear strain in matrix constituent, ε12 , for five deformed stages of a unit cell. The
geometry has an initial imperfection φ0 = 2◦ and a band orientation β = 10◦

To further enhance the reader’s understanding of kink band instability, a compressive

response from a representative finite element (FE) simulation is presented. Instead of showing
the deformation pattern of a full kink band geometry, the deformations of a single unit cell
strip with periodic boundary conditions are shown in Fig. 1.7. The model set-up is described
in Chapter 2. A compressive displacement is applied in the direction of the fibres. The
FE-model is composed of two layers of fibre and a single matrix layer. In the current example,
material parameters for an AS4/PEEK composite are used, taken from Kyriakides et al.
(1995). Five deformed stages of the geometry are presented in Fig. 1.7. The normalised
compressive stress −σ11 /G as a function of end-displacement is shown in Fig. 1.8, and the
normalised compressive stress as a function of fibre rotation φ is shown in Fig. 1.9. The
geometric quantities L0 , u and φ are designated in Fig. 1.6. The dark regions in Fig. 1.7
correspond to zones of large shear strain.
As described by Liu et al. (1996), the formation of kink bands can be divided into different
stages: the linear response and incipient kinking 1 , peak stress 2 , transient kinking 3
and steady-state kink band broadening 4 and 5 . Initially, a linear response is observed.
With increasing displacement, the stress in the matrix approaches the yield strength at stage
1 . At stage 2 the matrix has undergone adequate material softening, so that the composite
becomes unstable and a kink band initiates. Following the peak load 2 , the load drops and
the compressive displacement decreases due to elastic relaxation at stage 3 . The snap-back
phenomenon can not be captured in experiments; rather a dynamic response will follow the
peak-load in a real experiment for this particular choice of initial imperfection. The load
drops further and asymptotically approaches a constant stress at stage 4 , known as the
steady-state kink band broadening stress σ11 ss or Maxwell stress. Going from stage 4 to 5
10 Chapter 1 - Introduction

2
0.18
1
0.16
3
0.14 5
4
−σ11 0.12
G
0.1
0.08
0.06
0.04
0.02

0.005 0.01 0.015 0.02 0.025 0.03 0.035

u/L0

Figure. 1.8 Applied normalised stress −σ11 /G as a function of normalised end shortening u/L0 for a
simulation with a band orientation β = 10◦ . The five deformation stages are designated in Fig. 1.7.

the end-displacement has increased and the kink band has broadened or propagated under
the constant applied stress σ11ss . Moreover, by focusing on Fig. 1.9 shows almost, nearly no

further fibre rotation is observed going from stage 4 to 5 .

The instability phenomenon, kink band formation, discussed in the current dissertation is
similar to a number of other physical phenomena seen on both structural and material scales.
Common to several of these phenomena is that a critical load is followed by a steady-state.
Examples of instabilities on the material scale include neck propagation in the drawing of
polymers, e.g. Hutchinson and Neale (1983); crushing of metal foams, e.g. Ashby et al.
(2000); and instabilities in elastic bars, e.g. Ericksen (1975). Other examples of propagating
phenomena include channelling of cracks (e.g. Beuth (1992); Jensen and Thouless (1995))
and phase transformation of mediums, e.g. Guggenheim (1959).
Similar instabilities on the structural level were discussed in the review by Kyriakides
(1998) and include bulging of internally pressured rubber cylinders (e.g. Kyriakides and
Yu-Chung (1991); Corneliussen and Shield (1961); Chater and Hutchinson (1984)), buckling
and propagation of externally loaded undersea pipelines (e.g. Mesloh et al. (1973); Palmer
and Martin (1972); Jensen (1988)), and buckling of long shallow arches (e.g. Timoshenko
and Young (1962); Schreyer and Masur (1966); Kyriakides (1993)).
1.3 Scope of work 11

2
0.18
1
0.16
3
0.14
4 5
−σ11 0.12
G
0.1
0.08
0.06
0.04
0.02

5◦ 10◦ 15◦ 20◦ 25◦

φ

Figure. 1.9 Applied normalised stress −σ11 /G as a function of fibre rotation φ for a simulation with
a band orientation β = 10◦ . The five deformation stages are designated in Fig. 1.7.

1.3 Scope of work

The current dissertation is written and structured based on the ’Guidelines for writing a
cumulative PhD thesis’ set out by the Department of Engineering, Aarhus University. The
three subsequent chapters include the methods and the theoretical foundation behind the
articles written during the PhD. Brief descriptions of the articles are attached prior to the
articles themselves and include the motivation, methods, contributions and main findings for
the given article. The dissertation includes the following chapters:
Chapter 2 presents the inspiration, assumptions and implementation of the finite element
models used and presents appropriate equations and illustrations. The chapter does not
include the results obtained using the models since these are presented in the articles.
Chapter 3 describes an idealised kink band model used in the publications [P1-P4]. The
kink band model was initially developed by Jensen and Christoffersen (1997). Variations
of the model were developed during the PhD so a substantial amount of time was spent
understanding, modifying and developing the models.
Chapter 4 describes an analytical model used to study the kink band broadening phenomenon.
The model was initially developed in [P3] and was used in a similar form in [P4].
Chapter 5 summarises the work done during the PhD. Furthermore, the chapter puts the
contributions in perspective and suggestions for further work are given.
12 Chapter 1 - Introduction

Chapter 6 includes the first article and deals with the development of a three-dimensional
constitutive model that can predict strain localisation in fibre-reinforced composites. The
constitutive model is implemented within a kink band model that is presented in Chapter 3.
Chapter 7 includes the second article, which deals with the development of a constitutive
model for a fibre composite that takes imperfect fibre/matrix interfacial bonding into account.
The constitutive model is used together with a kink band model that is presented in Chapter 3.
Chapter 8 includes the third article and deals with the study of steady-state kink band
propagation using the three different models presented in Chapters 2, 3 and 4.
Chapter 9 includes the fourth article, which again deals with kink band broadening. This
publication includes two case studies, of different length scales, and involves different
materials.
Chapter 10 presents the fifth article. This article is written based on the work done during
my visit to the Department of Mechanical Engineering in Cambridge, UK. It deals with the
study of load transfer in a bolted joint in a laminate made from ultra-high molecular weight
polyethylene.
Chapter 11 includes an article, mainly written by the master’s student Vedad Tojaga, for
which I am a co-author. The article deals with the development of kink bands in open-hole
fibre composites under compressive loading.
Chapter 2

Finite element models

This chapter will describe the finite element models used to study fibre-kinking during the
project. Different FE-models of varying complexity have been developed for the study of
kink band formation in the last couple of decades. The FE-models presented in this chapter
are comparable to the models by Kyriakides et al. (1995) and Hsu et al. (1999). Either two- or
three-dimensional models can be used. In the current work, both two- and three-dimensional
models were used when appropriate.
The models are further divided into discrete and homogenised models. In the discrete
models, the composite constituents are modelled explicitly as different geometry parts, where
different material behaviours are given to the constituents e.g. fibre and matrix. For the
homogenised models, the average properties of a representative volume element are used.
There are advantages and disadvantages to each method. The advantages of using discrete
models are: details about the fibre geometry and distribution can be included; the interaction
with the surrounding matrix can be studied. If commercial software such as Abaqus/CAE
is used, the built-in material behaviour can be utilised. When the constituents are modelled
separately, the interaction between the parts can be modelled explicitly by e.g. using cohesive
elements or contact friction. Moreover, using X-ray tomography, a real fibre configuration
can be built. Based on the X-ray scan, discrete FE-models can be built by reconstruction
followed by segmentation into fibre and matrix parts. Some disadvantages of using discrete
models can be the computational time due to the level of detail. It can be rather difficult
to build the fibre-matrix layout explicitly, even for simple geometries, which gives a key
advantage to the use of homogenised models.
Most of the composites of interest exhibit non-linear material behaviour in the form of
plastic deformation, at the deformations required for kink band formation. Furthermore, by
using the common geometrical configurations used to study fibre-kinking, snap-back and
14 Chapter 2 - Finite element models

snap-through, instabilities occur in the equilibrium path when using an implicit finite element
code. This leads to the use of non-linear arc-length solvers. In all of the FE-simulations,
time-independent formation of kink bands is studied using Abaqus/CAE.

2.1 Geometries
A typical composite structure consists of multiple layers. A popular choice is to use uni-
directional fibre plies (or lamina) and stack them together with different fibre orientations, as
seen in Fig. 2.1. An assembly of plies is called a composite laminate, or simply a laminate.
The average anisotropic constitutive behaviour of a laminate is used on a larger scale to
predict the overall structural behaviour. A representative volume element (RVE) of a laminate
structure will in the general case be exposed to a three-dimensional stress and strain state.
Classical laminated plate theory is frequently used in connection with composite laminates.
Using the theory, assumptions about the continuity of displacements are made. Therefore,
if the deformation of a ply is known based on the deformation of the laminate, and if the
constitutive relation is also known, the stress state can be determined. Multiple authors have
developed failure criteria for composite plies for different loading conditions e.g. shear-,
tensile-, compressive-, fatigue failure and others. In the current study, the focus is given to
compressive failure in the fibre direction
Fig. 2.1 shows a laminate together with an exploded view of the same laminate. The
principal strains at a point can be determined based on the general strain state. The direction
of maximum compressive strains is of interest when considering a laminate composed of
uni-directional plies. The plies with fibre orientations closest to the direction of the maximum
principal compressive strain will be exposed to the highest compressive stresses due to higher
stiffness in the fibre direction. The blue highlighted plies in Fig. 2.1 will be exposed to the
highest stresses and will therefore fail first. Thus, it is a valid simplification to study a single
ply when regarding the compressive failure of laminates.
A section of a ply is illustrated in Fig. 2.2 to the left. A hexagonal distribution of the
fibres is assumed based on the assumption of equal spacing between fibres and an even
distribution of the matrix. The effect of variations in the layout of fibre distribution upon
the constitutive behaviour was studied by e.g. Huang (2018). A full model as displayed in
Fig. 2.2 can be very computationally expensive to solve when with regard to the study of kink
band formation. Naya et al. (2017) considered a full model using approximately 50 fibres that
were randomly distributed. In the simulations, they included cohesive elements between the
fibre and matrix components. Furthermore, finite deformation and material non-linearities
were included. Using a cluster, the full model studied by Naya et al. (2017) took more than
72 hours to solve. This high computational time has led to the use of simplified models,
2.1 Geometries 15

σ
σ

σ
σ

Figure. 2.1 Illustration of a laminate section (left) and an exploded view (right) of the same laminate.
The lamina that have the same fibre directions as the direction of loading are highlighted with blue.

either by simplifying the FE-problem or by using alternative computational methods, e.g. the
constitutive models presented in Chapters 3 and 4.
Examples of possible simplifications of the full FE-model are demonstrated in Fig. 2.2.
Here the simplifications are divided into two- or three-dimensional versions. Moreover, two
levels of simplification are shown: a layered model and a unit cell model. The layered models
make it possible to study variations of the geometry in the plane i.e. both initial geometry and
post-buckling configuration. As an example, Fleck and Shu (1995) identified the influence
of imperfections confined within an elliptically-shaped domain in the plane upon kink band
formation. Hsu et al. (1998) used a three-dimensional layered model and compared it with a
two-dimensional version; only small variations between the results were observed. Wind et al.
(2014) used a two-dimensional layered model, inspired by the one introduced by Kyriakides
et al. (1995).
The next level of simplification after layered models is unit cell models. A three-
dimensional unit cell model is used in [P2] in order to study the influence of bonding between
the fibre and matrix. Prabhakar and Waas (2013) used a two-dimensional model to study
the influence of fibre-to-matrix bonding; however when using a two-dimensional model and
when low or no bonding is present, the shear straining in the matrix becomes unrealistic.
This fact led to the use of a three-dimensional model in [P2]. The model was simplified into
a unit cell model to decrease the computational time. A two-dimensional unit cell model is
the most computationally efficient discrete kink band FE-model and was used in [P3] and
[P4] to study kink band broadening. When using the unit cell models to study kink bands,
16 Chapter 2 - Finite element models

3D

2D

Full models Layer models Unit cell models

Figure. 2.2 An illustration of a section from an uni-directional lamina (left). Two simplification levels,
Layer models and Unit cell models, are illustrated from Full models for both 2D and 3D models.

periodic boundary conditions (PBC) are required, which will be described in Section 2.3, to
avoid structural Euler buckling.
An alternative simplification from the full model is to use a homogenised model. An
example of a homogenised FE-model is presented in Fig. 2.3. As described earlier, the overall
constitutive behaviour is needed for these models. An example of a constitutive formulation
that was developed by Christoffersen and Jensen (1996) will be presented in Chapter 3. The
latter constitutive formulation was implemented in an FE-model and was used in [C1] to
study the compressive response of fibre composites with open-holes.

2.2 Imperfection
The critical compressive stress at the initiation of kink bands is highly sensitive to geometrical
imperfections as confirmed in the review by Fleck (1997). There are several ways to
introduce imperfections into the analyses. One choice of initial imperfection is to use a
superposition of deformations obtained from a linear buckling analysis. Alternatively, an
assumed imperfect geometry can be implemented; e.g. Kyriakides et al. (1995) introduced a
uniform sinusoidal waviness of the fibres and Fleck and Shu (1995) considered an area of
fibre imperfections confined to an ellipse. Sørensen et al. (2009) studied kink band formation
using a homogenised model with the constitutive formulation developed by Christoffersen
and Jensen (1996). In the FE-model by Sørensen et al. (2009), an imperfection in the form of
2.2 Imperfection 17

Figure. 2.3 An example of an FE kink band simulation using a homogenised constitutive model. The
dark zone is an area of high plastic strain.

an infinitesimal kink band was used where the directions of fibres φ were described by the
analytical expression

    
1 2 π cos β
 φ0 cos

 (X1 + X2 tan β ) + 1 + α if X1,min < X1 < X1,max
2 b
φ (X1 , X2 ) =

(2.1)

α otherwise

where φ0 is the largest imperfection angle, β is the angle of the band, b is the band width and
α is the fibre angle outside the band. The equation above was presented in Sørensen et al.
(2009) and can also be presented in the equivalent form
  
π cos β
φ0 cos2 (X1 + X2 tan β ) + α if X1,min < X1 < X1,max


b
φ (X1 , X2 ) =

(2.2)

α otherwise

The geometric quantities are presented in Fig. 2.4. Hutchinson and Koiter (1970) wrote a
review on the bifurcation and post-buckling of plates and shells. According to Hutchinson
and Koiter (1970), an imperfection in the shape of the buckling mode results in the largest
reduction in buckling load. An imperfection in the form of a kink band will therefore be the
most critical imperfection. By assuming that the fibre imperfection angle φ , described by
Eq. (2.1) or (2.2), is equal to the derivative φ = dx2 /dx1 , the deformed coordinates x2 can be
determined. Small letters xα refer to deformed coordinates while capital letters Xα are initial
18 Chapter 2 - Finite element models

β
φ0 + α
X1,min X2 X1,max

α
X1
w0

L0

Figure. 2.4 Illustration of initial imperfection in the form of fibre misalignments. Sørensen et al.
(2009) introduced the current imperfection and is used in the simulation presented in Fig. 2.3.

coordinates. It is assumed that the deformed coordinate x1 is equal to the initial coordinate
(x1 = X1 ). The deformed coordinate x2 can be integrated into the form

  
−b
X2 + 12 φ0 2 cosβ − X2 tanβ + α X1 if X1 < X1,min








(2.3)
h   i
x2 (X1 , X2 ) = X2 + 1 φ0 b 2 π cosβ
 2 2 π cosβ sin b (X1 + X2 tanβ ) + X1 + α X1 if X1,min < X1 < X1,max



  

X2 + 1 φ0
 b
− X2 tanβ + α X1 if X1,max < X1
2 2 cosβ

where the boundaries X1,min and X1,max are introduced, which are simply two equations for
the two boundary lines

−b b
X1,min (X2 ) = − X2 tanβ X1,max (X2 ) = − X2 tanβ (2.4)
2 cosβ 2 cosβ

The imperfection, Eq. (2.3), is used in [P2], [P3] and [P4]. A geometrical representation
of an initial geometry and the corresponding imperfect geometry are illustrated in Fig. 2.5.
The imperfection angle used in the figure is φ0 = 15◦ where the initial imperfection angles
used in the simulations are in the range φ0 ≈ 0◦ -5◦ . In both Fig. 2.4 and Fig. 2.5, the initial
geometry width w0 and length L0 are presented.
2.3 Boundary conditions 19

β
X1,min X1,max x2
X2 φ0

X1 x1
w0

L0 b

Figure. 2.5 Illustration of an uni-directional lamina before (left) and after (right) the imperfection is
used given by Eq. (2.3). The fibre angle outside the band is α = 0◦ in the current illustration.

2.3 Boundary conditions

The most simple boundary conditions are applied to the geometries that still capture the kink
band phenomenon. The displacements ui are, introduced which are the displacements along
the corresponding axes Xi . In all of the simulations, the left boundary is simply supported
and the lower left corner is fully fixed. With a coordinate system placed in the geometry
centre, the boundary condition can be described by

L0
u1 = 0 on X1 = −
2 
L0 w0
u2 = 0 on (X1 , X2 ) = − , − (2.5)
2 2

The boundary conditions are graphically displayed in Fig. 2.6. Furthermore, in the
FE-simulations where a three-dimensional model is used, the front and back are simply
supported as well:

t0
u3 = 0 on X3 = −
2
(2.6)
t0
u3 = 0 on X3 =
2
where t0 is the initial thickness of the geometry. Besides the boundary conditions, loading is
applied in the form of applied stress or displacement. In [P2], [P3] and [P4], loading in the
form of an applied displacement u is used. When using an applied displacement field, the
stress can be calculated based on the reaction forces in the corresponding nodes.
20 Chapter 2 - Finite element models

x2
x2
x3
x1

Figure. 2.6 Visualisation of the boundary conditions used for the FE kink band model. The BCs on
the right does only apply for the 3D models.

In the simulations where a unit cell model is used (as in [P2], [P3] and [P4]), periodic
boundary conditions are used between the top and bottom boundary. The PBCs are applied
such that a single unit cell model represents an infinite array of fibres. The PBCs are enforced
using constraint equations in the form

(A) (B)
ui − ui =0 (2.7)

where (•)(A) and (•)(B) correspond to nodes on the top and bottom boundaries. An example
of the coupling between two nodes is presented in Fig. 2.7. When using PBCs, it is common
(A) (B) (A) (B)
to use nodes that share some coordinates, e.g. X1 = X1 and X3 = X3 , as in the model
used in [P2]. An alternative node coupling is introduced in [P3] to account for the band
inclination. In [P3] the coupled nodes are oriented with an angle β prior to deformation. This
forces the band to initiate and to propagate in a specified angle β .

2.4 Discretisation
In the FE-model presented in [P3], the influence of the number of elements per constituent
layer was investigated. It was concluded that a single element in the height per layer is
sufficient. The material behaviour of the constituents, presented in [P1]-[P4], is taken
from Kyriakides et al. (1995). The constituents in the latter study were AS4-carbon fibres
embedded in a PEEK matrix. In this composite, Young’s modulus of the fibres is 35 times
the modulus of the matrix. In a kink band simulation using the latter constituents, the fibres
will take most of the bending and the matrix will almost be in a state of pure shear.
2.4 Discretisation 21

β Node couples

=
Fiber Q8

+
Matrix Q8
or

Matrix Q4

Figure. 2.7 Illustration of the discretisation of the kink band geometry into separate elements. In the
top geometry, the node couplings are presented when using periodic boundary conditions. Furthermore,
the combination of elements are illustrated.

In the two-dimensional models plane-strain, 8-node biquadratic elements (Q8) with

full integration were used for the fibres. Biquadratic elements were chosen for the fibre
geometry instead of bilinear elements since Q8 elements give a more realistic response
during bending. The influence of using either 8-node biquadratic elements or 4-node bilinear
elements (Q4) with full integration for the matrix constituent was investigated in [P3]. The
idea of combining two incompatible elements (biquadratic and bilinear) was inspired by the
micro-mechanical model used by Sutcliffe and Fleck (1997). The discretisation scenarios
are presented graphically for a unit cell model in Fig. 2.7. No observable differences were
seen in the applied stress versus end-displacement curve when converting the elements from
biquadratic to bilinear. One of the primary benefits of using bilinear elements for the matrix
layer is a lower computational time; however more importantly, it was possible to obtain
a larger part of the post-bifurcation response. In [P3] and [P4] the propagating instability
known as kink band broadening was investigated and here the late post-buckling response is
of high importance.
In the three-dimensional unit cell model from [P2], a combination of 20-node quadratic
bricks and 15-node triangular prism elements with full integration was used for the fibre
constituents, while 20-node quadratic bricks with full integration were used for the matrix.
Again, the influence of mesh refinement was investigated for the three-dimensional model
22 Chapter 2 - Finite element models

Figure. 2.8 Illustration of the cross-section of the three-dimensional unit cell model showing the
different mesh refinements used.

Interaction

F2

M F3

F1 Interaction
x2
Interaction
x3

Figure. 2.9 The cross-section of the three-dimensional unit cell model (left) and an exploded view
(right) of the same section. The surfaces are designated where fibre-to-matrix interaction can occur.

and three examples of refinement of the cross-section are shown in Fig. 2.8. Visualisations
of the nodes is omitted for clarity. The coarsest mesh presented in Fig. 2.8 was used due to
negligible differences in the critical compressive stress and due to the lower computational
time.

2.5 Interaction
In [P2], the influence of fibre-to-matrix bonding upon the critical compressive strength was
investigated. A normal and an exploded view of the cross-section of the unit cell model are
presented in Fig. 2.9. The unit cell was composed of four parts: three fibre parts and one
matrix part. In the FE-model, the four parts were formulated as four different parts with
2.5 Interaction 23

the possibility of exploiting the fibre-to-matrix interaction by using e.g. contact friction
or cohesive elements. The FE-model in [P2] was compared with a constitutive model that
will be presented in Chapter 3. For simplicity’s sake, two simple interaction scenarios were
used: a tie constraint and a frictionless hard contact interaction. Based on the two interaction
possibilities, three bonding scenarios were created: a case of perfect bonding; a case of no
bonding; and a case of intermediate bonding. In the case of perfect bonding, tie constraints
were applied between the relevant parts and in the case of no bonding, the frictionless contacts
were used. In the case of intermediate bonding, tie constraints were used for half of the
surfaces (dashed lines in Fig. 2.9), while for the remaining surfaces, frictionless contacts
were used.
Chapter 3

Semi-analytical kink band model

This chapter will describe an idealised kink band model, which will be referred to as the semi-
analytical model. The kink band is idealised to consist of two material states: a state inside the
kink band and a state outside. However, in contrast to the FE-model, the current model does
not consider the transition from unkinked to the kinked configuration. The kink band model
that will be presented in the current chapter was developed by Jensen and Christoffersen
(1997) using a homogenised constitutive formulation introduced in Christoffersen and Jensen
(1996). The model requires continuity of displacements and traction equilibrium between
the base and kinked material and is solved incrementally. Full elastic-plastic constitutive
formulations can be used with the model without initial requirements of e.g. rigid fibres or
incompressibility.
The constitutive model is used in several of the publications [P1-P4] and variations of the
model so a substantial amount of time was needed during the project to understand, modify
and develop the models. Therefore, the assumptions and theories behind the models will
be described in detail in the current chapter. It is assumed that the reader has knowledge
about index notation and fundamental continuum mechanics (such as the equilibrium- and
compatibility equations).

3.1 General relations

In the current section, the continuum mechanics foundation is presented. Throughout the
chapter, the index notation and the summation convention are adopted. Latin indices i.e.
i, j, k take values 1, 2, 3 and Greek indices α, β , γ take values 1, 2. In general, the Latin and
Greek indices are used with three- and two-dimensional problems, respectively. A comma
26 Chapter 3 - Semi-analytical kink band model

t0 t + dt

P
P
ui vi

P
xi

x2
Xi

x1
x3

Figure. 3.1 Illustration of an arbitrary continuum body at three instants of time. The illustration shows
the important vectors founding the basis of the kinematics.

(•),i denotes a partial derivative with respect to the xi coordinate. A general continuum body
is illustrated in Fig. 3.1 in a three-dimensional Euclidean space with a Cartesian coordinate
system containing the three orthonormal directions x1 , x2 and x3 . The initial state of the
body at time t = 0 is described by the coordinates Xi with capital letters. The body may
include residual stresses σi j prior to deformation. The body in the current or deformed
configuration at time t is described by the coordinates xi . The displacement vector ui (Xi )
describes the displacements of every point going from the initial to the current configuration.
The configuration subsequent to a small time increment t = t + dt is illustrated as well. In
both time-dependent and -independent continuum mechanics the velocity vector vi = ẋi is
used, where the dot (•) ˙ refers to the material time derivative. In the current study, time-
independent constitutive formulations are used and here vi are regarded as infinitesimal
displacement increments taking place during a pseudo-time increment dt. In the proceeding
derivations, deformation gradients are used. The deformation gradients relate infinitesimal
vector elements in the initial to the current configuration

dxi = Fi j dX j (3.1)
3.1 General relations 27

v1,1 > 0 v2,2 > 0 v2,1 > 0 v1,2 > 0

v2
x2 v1
v1

x1 v2

Figure. 3.2 Five configurations for a square are shown: one initial state to the left and four deformed
configurations.

where the definition of the deformation gradients is

∂ xi
Fi j = (3.2)
∂ Xj

Furthermore, the velocity gradients vi, j are used in the constitutive model. The velocity
gradients give information about how the velocities change in the vicinity of a point in
the direction of the basis xi . Simple illustrations of squares undergoing homogeneous
deformations are shown in Fig. 3.2 with the relevant velocity gradients marked on the figure
above the deformations. The deformed configuration (dark area) is superimposed upon the
initial configuration (light area). The strain increment tensor ε̇i j and the spin tensor ωi j
are used in the model and are the symmetric and the anti-symmetric parts of the velocity
gradients

1 1
ε̇i j = (vi, j + v j,i ) ωi j = (vi, j − v j,i ) (3.3)
2 2

Here ε̇i j give the pure deformation increments and can be used with physical constitutive
formulations and ωi j gives the local rotation or spin of the points.
When setting up constitutive formulations of continuum bodies that include residual
stresses, it is important to distinguish between stress increments that arise from deformations
and those that arrive from the rotations of the local basis. The Cauchy stress tensor σi j is
used and gives the force per unit surface area in a given configuration. Furthermore, the
Kirchhoff stress tensor τi j is used and is related to the Cauchy stresses by

τi j = J σi j (3.4)
28 Chapter 3 - Semi-analytical kink band model

where J is the Jacobian, giving the volume ratio J = V /V0 or density ratio J = ρ0 /ρ between
the current and initial configuration. The models assume that the relation between the rate of
Kirchhoff stresses and strain increments is described by

τ̊i j = Li jkl ε̇kl (3.5)

where Li jkl are components of the elastic-plastic tangent moduli and τ̊i j are the Jaumann rate
of Kirchhoff stresses and are the work-conjugate of the strain rate tensor ε̇kl as confirmed by
Ji et al. (2013). The moduli Li jkl must satisfy the symmetries

Li jkl = L jikl = Li jlk (3.6)

due to the symmetries of the Kirchhoff stress (τ̊i j = τ̊ ji ) and strain increment tensor (ε̇kl = ε̇lk ).
The only restriction to the constitutive formulations giving the moduli Li jkl is that time-
independent material behaviour is used, whether the formulations are elastic or elastic-plastic
e.g. J2 -deformation theories or J2 -flow theories. For incompressible materials τ̊i j = σ̊i j
where σ̊i j is the Jaumann rate of Cauchy stresses. Both σ̊i j and τ̊i j are objective stress rates
that are independent of the local material rotation. As mentioned earlier, it is important
to distinguish between stress increments that have geometrical or physical origins. The
stresses after deformation in a coordinate system that stays stationary during an increment
are σi j (t + dt) = σi j (t) + σ̇i j . The relation between the Cauchy stress increments σ̇i j and σ̊i j
is
σ̇i j = σ̊i j + ω jk σik + ωik σk j (3.7)
|{z} | {z }
Physical origin Geometrical origin

The Jaumann rate of Cauchy stresses σ̊i j describes the stress increments in a coordinate
system that rotates with the deformation; for a more comprehensive explanation see Appendix
A. In the same manner the Jaumann rate of Kirchhoff stresses is given by

τ̇i j = τ̊i j + ω jk τik + ωik τk j (3.8)

|{z} | {z }
Physical origin Geometrical origin

Alternative stress quantities (nominal stresses or Second Piola-Kirchhoff stresses) oc-

cur when considering finite deformations. They are frequently used in connection with
total Lagrangian formulations where the state of strain is described based on the initial
configuration. The alternative stresses arrive as the work-conjugate of the choice of finite
strain measurements. The nominal stress tensor ti j is used to set up the constitutive models
presented in this chapter even though the models are formulated with an updated Lagrangian
approach. In Fig. 3.3 the concept of the nominal stress tensor ti j vs Cauchy stress σi j is
3.1 General relations 29

illustrated using a simple example. In Fig 3.3a a dog bone specimen is displayed in its initial
and deformed states. In this example, it is assumed that no residual stresses are present. The
surface area of the cross-sections before and after deformation is designated on the figure.
In Fig. 3.3b, two cuboids are presented, where the one to the left is a unit cube. The force
acting per unit surface area in the deformed configuration is given by σ11 = F1 /s1 , which is
the Cauchy stress or true stress. The nominal stress is the force per initial unit surface area
t11 = F1 /S1 . The nominal stresses can be seen as the forces acting on a cuboid’s faces (or in
the general case, a parallelepiped) that was initially a unit cube in the reference state.

σ11 = F1 /s1 t11 = F1 /S1

S1 σ11 t11

F1 F1
x2
x2 s1 σ11
x1
t11
x1 x3
(a) Uniaxial test (b) Cauchy and nominal stress

Figure. 3.3 Illustration that shows the difference between Cauchy stresses σi j and nominal stresses ti j
from a simple uniaxial loading example.

A more general case is illustrated in Fig. 3.4. In the figure, an initial and a deformed
configuration are displayed. From the deformed configuration, an infinitesimal cube and an
infinitesimal parallelogram are drawn from the same point. The cube represents the forces
acting per unit area. The parallelogram represents an infinitesimal piece that was initially a
unit cube in the reference configuration. The relation between nominal stresses and Cauchy
stresses is
∂ Xi
ti j = J σk j = J Fik−1 σk j (3.9)
∂ xk
The relation between the Cauchy stresses and nominal stresses can be obtained by considering
an infinitesimal force vector acting on an area in the current and reference configurations with
the equality d fi = ti ds = Ti dS, where ti and ds are a traction and an infinitesimal area around
a point in the current configuration. Furthermore Ti and dS are the corresponding nominal
traction vector and infinitesimal area in the reference configuration. Cauchy’s stress theorem
states that the tractions can be determined based on ti = σ ji n j and Ti = t ji N j . Nanson’s
formula dsi = J Fi−1j dS j relates surface areas in the deformed configuration to those in the
30 Chapter 3 - Semi-analytical kink band model

x2

x1 σ
t22

σ22 t21
t12
σ12
t11
σ12
σ11
σ11 t11

σ12 t12
t21
σ12
t22
σ22
Figure. 3.4 An initial and a deformed configuration. A single point showing the Cauchy- and nominal
stress tensor on an infinitesimal piece.
3.2 Homogenisation of constituents 31

reference configuration. The relation given by Eq. (3.9) can be derived by combining the
latter equalities. For a more comprehensive derivation see Appendix B. In addition, the
nominal stresses can be expressed with respect to Kirchhoff stresses:

∂ Xi
ti j = τk j = Fik−1 τk j (3.10)
∂ xk

As mentioned earlier, the kink band model is formulated with an updated Lagrangian
formulation. The rate of nominal stresses t˙i j are used in the models. Taking the material time
derivative d/dt(•) of Eq. (3.10) yields the rate of nominal stresses as a function of the rate of
Kirchhoff stresses:
t˙i j = τ̇i j − τ jk vi,k (3.11)

where τ̇i j are the rates of Kirchhoff stresses when following a point seen from a coordinate
system that does not rotate. The rate of Kirchhoff stresses can be written as a function
of the Jaumann rate of Kirchhoff stresses using Eq. (3.8), which is needed in constitutive
formulations.
t˙i j = τ̊i j + ω jk τik + ωik τk j − τ jk vi,k (3.12)

Expanding the spin tensor using Eqs. (3.3) and (3.5), the velocity gradients can be isolated in
the latter equation:
 
1 1 1 1
t˙i j = Li jkl − δil τk j − δik τl j − τil δk j + τik δl j vl,k (3.13)
2 2 2 2
| {z }
Ci jkl

where δi j is Kronecker delta and Ci jkl are components of the tensor of nominal moduli relating
nominal stress increments to velocity gradients through the Kirchhoff stress tensor. Details
behind the derivations resulting in Eq. (3.13) can be studied in Appendix B. The moduli Ci jkl
are used in the model set-up since they account for the deformations and tractions in the
updated deformed configuration. The following form will be used to set up the constitutive
models:
t˙i j = Ci jkl vl,k (3.14)

3.2 Homogenisation of constituents

In the current section, three models for homogenising the properties of fibre-reinforced
composites and layered materials will be described. First, a two-dimensional homogenised
model developed by Christoffersen and Jensen (1996) is described, followed by a three-
32 Chapter 3 - Semi-analytical kink band model

dimensional version developed in [P1]. Lastly, a model taking account of imperfect bonding
between fibres and the surrounding matrix is presented and developed in [P2]. The models
result in constitutive formulations in the form given by Eq. (3.14). In the current dissertation,
the focus is on kink band formation; however, the constitutive formulations can be used in
general for e.g. FE-analyses where there is a need for elastic-plastic composite behaviour.
The common factor and the unique point of the three formulations are that independent
constitutive equations are given to the constituents. This differs from regular constitutive
formulations used for composites, where the composite behaviour is determined based on the
overall response.

3.2.1 2D model
In Fig. 3.5, a general piece of a fibre-reinforced or layered composite is shown to the left.
When setting up the constitutive behaviour, a representative volume element (RVE) is needed,
as depicted on the right-hand side of the figure. The size of the RVE is determined based on
the phenomenon requiring attention. In kink band formation the localisation of deformation
needs information on the size of the fibres. In the two-dimensional model developed by
Christoffersen and Jensen (1996), volume fractions of the two constituents are needed.
The current formulation does not include a length scale which requires strain gradients
(higher order strains) as in the formulation developed by Poulios and Niordson (2016). The
two-dimensional formulation can be used for fibre-reinforced composites and assumes that
the composite behaviour can be adequately described with a layered material as indicated
in Fig. 3.5. The homogenisation assumes that the wavelengths λ of the fibre waviness
are much larger than the amplitudes λ ≫ a. Unit-thickness of the composite is assumed,
meaning the volume V and area A of the composite are equal: V = A. Two constituents
are introduced, indicated with superscripts (•) f and (•)m , indicating a fibre and matrix
constituent, respectively. Omission of superscripts refers to overall composite properties.
The volume fraction of fibres is c f = V f /V where V f is the fibre volume of the RVE. With
regard to two-dimensional models, the fibre volume fraction is given by c f = A f /A where
A f is the area of fibres as indicated in Fig. 3.5. The relation

c f + cm = 1 (3.15)

is required for the fibre and matrix volume fractions.

3.2 Homogenisation of constituents 33

Am
λ

Af

Figure. 3.5 A general piece from a two-dimensional layered composite (left) and a representative
volume element from the composite (right).

v1,1 > 0 v2,2 > 0 v2,1 > 0 v1,2 > 0

x2

x1

Figure. 3.6 A representative volume element of a layered composite in one undeformed and four
deformed configurations. The figure clarifies the kinematic equalities used in the homogenisation
formulation.

The constitutive formulation requires that the fibres of the RVE are aligned with the x1
axis prior to deformation. The homogenisation of the constituents is based on the assumptions
1. Material lines parallel with the fibres are subject to a common stretching and rotation.
2. Planes parallel with the fibres transmit identical tractions.
3. The material behaviour of the constituents is time-independent.
Five configurations of an RVE are presented in Fig. 3.6: one undeformed and four deformed
configurations. The first requirement, that lines parallel with the fibres are subject to a
common stretching and rotation, is illustrated through the second and fourth configurations.
The kinematic requirements are presented mathematically through the velocity gradients
34 Chapter 3 - Semi-analytical kink band model

f
v1,1 = v1,1 = vm
1,1
f
(3.16)
v2,1 = v2,1 = vm
2,1

which can be written with a compact notation using the index notation

f
vα,1 = vα,1 = vm
α,1 (3.17)

Here, Greek indices are used since the indices take values 1 and 2. Furthermore, the two
kinematic requirements

f
v1,2 = c f v1,2 + cm vm
1,2
f
(3.18)
v2,2 = c f v2,2 + cm vm
2,2

are required for overall compatibility. The equalities simply state that the overall deformations
v1,2 and v2,2 are average strains of the constituent deformations. The latter requirement can
be presented using index notation:

f
vα,2 = c f vα,2 + cm vm
α,2 (3.19)

The second requirement states that planes parallel with the fibres transmit identical tractions.
The requirement sets a restriction upon the stresses prior to deformation

f m
σ12 = σ12 = σ12
f m
(3.20)
σ22 = σ22 = σ22

as indicated in Fig. 3.7 to the left. The rate of nominal stresses can be used to enforce the
required equilibrium during the increments through

f m
t˙21 = t˙21 = t˙21
f m
(3.21)
t˙22 = t˙22 = t˙22

which states that the traction increments, on the plane with outward normal parallel to the
x2 -axis, are identical for the overall composite and for the constituents (Fig. 3.7 on the right).
The requirements presented with indices

f m
t˙2α = t˙2α = t˙2α (3.22)
3.2 Homogenisation of constituents 35

σ22 + t˙22

σ12 + t˙21
σ22

σ12 σ12 + t˙12

σ11 + t˙11
σ12

σ11

m + t˙m
σ22 22
m
σ22
m + t˙m
σ12 21
m
σ12 m
σ12 m + t˙m
σ12 12
m m + t˙m
σ11
σ11 11

f f f
σ22 σ22 + t˙22
f f
f
σ12 σ12 + t˙21
f f f
σ12 σ12 + t˙12
f f
f σ11 + t˙11
σ11

Figure. 3.7 An initial state with stress equilibrium of a representative volume element is shown (left)
and a (exaggerated) deformed configuration after an increment with equilibrium of tractions is shown
(right).
36 Chapter 3 - Semi-analytical kink band model

Overall equilibrium of stresses prior to deformation requires

f
σ11 = c f σ11 + cm σ11
m
(3.23)

The requirements of the traction increments are

f
t˙11 = c f t˙11 + cm t˙11
m

f
(3.24)
t˙12 = c f t˙12 + cm t˙12
m

and presented with indices

f
t˙1α = c f t˙1α + cm t˙1α
m
(3.25)

Based on the latter equalities a constitutive equation in the form

t˙αβ = Cαβ γδ vδ ,γ (3.26)

can be derived, where the homogenised moduli Cαβ γδ are functions of the constituent moduli
c
Cαβ γδ and the volume fractions. Using Eqs. (3.17), (3.19), (3.22) and (3.25), Christoffersen
and Jensen (1996) identified that the moduli can be calculated using
   
f f f
Cαβ γδ = c f Cαβ m m f m
γδ + c Cαβ γδ − c c Cαβ m m
2ε −Cαβ 2ε Hεζ C2ζ γδ −C2ζ γδ
(3.27)
f
Mαβ = cm C2α2β + c f C2α2β
m

where Hαβ is the inverse of Mαβ , complying to

Hαγ Mγβ = δαβ (3.28)

where δαβ is the Kronecker delta. The moduli Cαβ c

γδ for the constituents are calculated
c
using Eq. (3.13). The velocity gradients vα,2 for the constituents can be calculated using the
velocity gradients vα,β for the composite
 
f f
vε,2 =−cm Hεζ C2ζ 1δ
−C m m
2ζ 1δ vδ ,1 + Hεζ C2ζ 2δ vδ ,2
 
f f
vm f m
ε,2 = c Hεζ C2ζ 1δ −C2ζ 1δ vδ ,1 + Hεζ C2ζ 2δ vδ ,2 (3.29)

Due to differences in bulk moduli between the fibre and matrix constituent, the volume
fractions will change during deformation. The change in fibre volume fraction can be
3.2 Homogenisation of constituents 37

calculated using the velocity gradients

 
f
ċ f = c f cm v2,2 − vm
2,2 (3.30)

The derivations of Eqs. (3.27) and (3.30) are demonstrated in Appendix C using a vector
notation. The constitutive formulation presented in Eq. (3.27) was used in Christoffersen
and Jensen (1996) to study kink band formation at bifurcation. Furthermore, the relation
was used in Jensen and Christoffersen (1997) to study kink band formation with initial fibre
imperfections in the form of fibre misalignments. The kink band model is presented in
Section 3.4.

3.2.2 3D model
A three-dimensional version of the model was developed in [P1]. The assumptions behind the
model are described briefly; the article can be read for an extended explanation. In the article,
the emphasis is on fibre-reinforced composites, where fibres are surrounded by or embedded
in a resin material. Instead of simplifying an RVE into a two-dimensional domain with two
constituents, a three-dimensional cube with three constituents is introduced. An example of
one of the RVEs introduced in the article, is displayed in Fig. 3.8. Two matrices and one fibre
constituent were used. A methodology was developed to homogenise the properties between
two arbitrary constituents (•)a and (•)b with a shared plane that was parallel with either
the x1 -x2 or the x1 -x3 plane. The homogenisation of the fibre-matrix model was obtained
through two homogenisation steps as indicated in Fig. 3.8. Three assumptions are used for
the homogenisation between two arbitrary materials in 3D:
1. Material lines lying on the common plane between two materials are subjected to a
common stretching and rotation.
2. Planes on the intersection between two materials transmit identical tractions.
3. The material of the constituents is elastic or elastic-plastic.
When the shared plane between the constituents lies in the x1 -x3 plane, the first assumption
leads to the restriction of the velocity gradients

vi,1 = vai,1 = vbi,1

(3.31)
vi,3 = vai,3 = vbi,3

where the Latin indices (•)i are used, indicating that they take values 1, 2 or 3. To have
overall compatibility, the average of the strains in the x2 direction requires

vi,2 = ca vai,2 + cb vbi,2 (3.32)

38 Chapter 3 - Semi-analytical kink band model

= +
x2

x3 x1

Figure. 3.8 A representative volume element of a three-dimensional cube consisting of two homogeni-
sations.

The second assumption leads to the stress equilibrium

a b
σ12 = σ12 = σ12
a b
σ22 = σ22 = σ22 (3.33)
a b
σ23 = σ23 = σ23

prior to a deformation increment and the equilibrium of tractions on the shared plane requires

a b
t˙2i = t˙2i = t˙2i (3.34)

Overall equilibrium on the two other planes prior to deformation requires

σ11 = ca σ11
a
+ cb σ11
b

σ33 = ca σ33
a
+ cb σ33
b
(3.35)
σ13 = ca σ13
a
+ cb σ13
b

and during an increment, the rate of nominal stresses

t˙1i = ca t˙1i
a
+ cb t˙1i
b
(3.36)
t˙3i = ca t˙3i
a
+ cb t˙3i
b
3.2 Homogenisation of constituents 39

is required. Using the latter equalities, [P1] demonstrates that the rate of nominal moduli can
be written in a form similar to
   
Ci jkl = ca Ciajkl + cb Cibjkl − ca cb Ciaj2m −Cibj2m Hmn C2nkl
a b
−C2nkl
(3.37)
Mi j = ca C2i2
b b a
j + c C2i2 j

where Hi j is the inverse of Mi j complying to the equality:

Hik Mk j = δi j (3.38)

In [P1], the equations were presented using a matrix format similar to the one used in
Christoffersen and Jensen (1996) to simplify the derivations. The moduli given by Eq. (3.37)
give the homogenised properties between two materials that share a common plane with
outward normals parallel with the x2 axis. Similarly, the homogenised moduli can be
formulated for two arbitrary materials that share planes with outward normals parallel with
the x3 axis. In this case the moduli are given by
   
Ci jkl = ca Ciajkl + cb Cibjkl − ca cb Ciaj3m −Cibj3m Hmn C3nkl
a b
−C3nkl
(3.39)
Mi j = ca C3i3
b b a
j + c C3i3 j

By using a combination of the homogenisation equations Eqs. (3.37) and (3.39), the
overall homogenised properties of a fibre-reinforced composite can be obtained as presented
in [P1]. It is emphasised that general time-independent elastic-plastic material behaviour can
be given to the constituents using the current three-dimensional constitutive formulations. The
models are formulated in a rate form and can be used in formulations with large deformations
and rotations.

3.2.3 Imperfect bonding

The article [P2] presents the development of a two-dimensional constitutive formulation
accounting for imperfect bonding between fibre and matrix. The formulation is inspired by the
two-dimensional model by Christoffersen and Jensen (1996) assuming perfect bonding, and
the version introduced in Jensen (1999b) assuming complete decohesion. In the constitutive
formulation presented in [P2] a variable µ is introduced to control the amount of bonding
where µ = 1 is a case of perfect bonding and µ = 0 is a completely debonded case.
40 Chapter 3 - Semi-analytical kink band model

v1,1 > 0 v2,2 > 0 v2,1 > 0 v1,2 > 0

Figure. 3.9 A representative volume element of a layered composite with imperfect bonding in one
undeformed and four deformed configurations. The figure clarifies the kinematic assumptions used in
the formulation.

In Fig. 3.9 an initial undeformed configuration together with four deformed configurations
of an RVE are displayed. The figure is presented in [P2] to show how the deformations vary
between the matrix and fibre constituent in composites with poor bonding.
Once again, the constitutive formulation assumes that material lines parallel with the
fibres are subject to a common stretching and rotation. In contrast, the strains in the x2
direction are not simply averages of the strains as in the formulation by Christoffersen and
Jensen (1996). In the extreme case of complete debonding, the shear strains in the matrix are
equal to the overall shear strains and no shear strains are transmitted to the fibres. Similarly,
the shear tractions are not transmitted to the fibres in the case of complete debonding. In [P2]
it is shown that the constitutive formulation given by
   
f f f
Cαβ γδ = c f Cαβ γδ + cm m
Cαβ γδ − c f m
c κ Cαβ 2ε − µ C m
αβ 2ε Hεζ κ C2ζ γδ
− µ C m
2ζ γδ
f
Mαβ = cm κ 2 C2α2β + c f µ 2 C2α2β
m

(3.40)

gives the relation for an imperfect composite, where the variables µ and κ are introduced to
control the amount of bonding. The variable κ is dependent on µ according to
 
1 1
κ(µ) = 1 − m µ + m (3.41)
c c

In the case of perfect bonding µ = 1, the constitutive formulation, Eq. (3.40), equals the
two-dimensional version from Christoffersen and Jensen (1996), Eq. (3.27). In the case of
3.3 Kink band bifurcation 41

complete debonding µ = 0, the formulation simplifies to

µ =0→
f f f f
Cαβ γδ = c f Cαβ m m
γδ + c Cαβ γδ − c Cαβ 2ε Hεζ C2ζ γδ (3.42)
f
Mαβ = C2α2β

The latter form of the constitutive formulation was proposed by Jensen (1999b). Details
behind the derivations and the use of the constitutive formulation can be explored in [P2].
The use of the constitutive formulation is explored in [P2] using a kink band model. The
model predictions are compared with a three-dimensional unit-cell model with different
amounts of bonding as presented in Section 2.5.

3.3 Kink band bifurcation

The constitutive formulations presented for fibre-reinforced and layered composites are used
studying kink band formation. Initially, a methodology is introduced to investigate kink
band formation at bifurcation, i.e. no initial fibre misalignment. In Section 3.4, a numerical
formulation is introduced to enable the study of post-bifurcation behaviour, e.g. initial fibre
imperfections.
Kink band formation in the absence of fibre misalignments can be predicted as localisation
of deformation at bifurcation where ellipticity of the incremental equilibrium equations is
lost as explained in Jensen and Christoffersen (1997). A similar approach was used by
Hutchinson and Neale (1978) to study the localisation phenomenon known as sheet necking
that can form under tensile loading. They were inspired by the fundamental work by Rice
(1976) regarding localisation of plastic deformation. The current bifurcation formulation
within kink band formation was initially used by Christoffersen and Jensen (1996) together
with the two-dimensional constitutive formulation. The methodology is used in [P2] to study
fibre kinking in imperfect composites.
Fig. 3.10 depicts a perfect composite. The fibres are aligned in the x1 -direction. A general
stress field σαβ can be present prior to localisation. An alternative velocity field is searched
for where the fibres localise in a band according to

vα = fα (n1 x1 + n2 x2 ) = fα (nβ xβ ) (3.43)

where vα are the velocity components and fα are functions describing the velocity field and
are dependent on the normal nβ to an assumed band and the coordinates xβ . In this way
42 Chapter 3 - Semi-analytical kink band model

σ22

σ12

σ12
nα
σ11 σ11

σ12 β

x2
σ12

x1 σ22

Figure. 3.10 Kink band geometry before localisation. The angle β is the orientation of the localisation
band searched for in the bifurcation analysis.

the velocity fields are constrained to vary across the band. The state at bifurcation requires
equilibrium of the nominal stress rates

t˙αβ ,α = 0 (3.44)

Replacing the rate of nominal stresses t˙i j with the form given by Eq. (3.14) yields

Cαβ γδ vδ ,γα = 0 (3.45)

and when using the assumed velocity field, Eq. (3.43), the equilibrium equation can be
rewritten using the product rule of differentiation

∂ 2 fδ (nε xε )
Cαβ γδ nα nγ =0 (3.46)
∂ (nζ xζ )2

The equation above gives two equations since β is the only free index. Thus, bifurcation is
possible when a non-trivial solution to the latter equation exists (where the equation loses
ellipticity), i.e. when the determinant of the coefficient matrix of the function fδ vanishes.

det(Cαβ γδ nα nγ ) = 0 (3.47)
3.4 Kink band model and numerical scheme 43

No assumptions were made regarding the constitutive behaviour of the solid. By using the
two-dimensional constitutive equation Eq. (3.27), and by assuming rigid fibres, Christoffersen
and Jensen (1996) showed that the state at kink band bifurcation is

1 m 2 1 m
m
− c f σ12 tanβ + m C2222 tan2 β


0 = σ11 − σ22 + m
C2121 + m L1222 (3.48)
c c c

As [P2] shows, the localisation of fibres at bifurcation for an imperfect composite with rigid
fibres is

σ m − κ σ22
 
1 m 2 m
− 11 − c f κ µ σ12 tanβ


0 = σ11 − σ22 + L1212 + m 2 L1222
c κ2
m 2 c κ
1 m
+ m 2 (L2222 − κ σ22 ) tan2 β (3.49)
c κ
Furthermore, [P2] shows that the latter equation is identical to Eq. (3.48) with µ = 1. In
contrast to the results in the proceeding sections, the current formulation can result in a closed
form solution. The solutions give valuable information regarding the effect of the stress field,
constituent volume fractions and kink angle β upon the critical stress σ11 cr at bifurcation.

3.4 Kink band model and numerical scheme

The constitutive formulations introduced in Section 3.2 can be used in a numerical scheme to
study kink band formation. This scheme is presented in the current section. The numerical
formulation is developed by Jensen and Christoffersen (1997) (two-dimensional) and is used
in [P2], [P3] and [P4]. The formulation is extended to a three-dimensional version in [P1]
and is used to validate the constitutive formulation presented in Section 3.2.
Common to both the two- and three-dimensional numerical formulations is that a kink
band forms either by bifurcation or due to the presence of an initial imperfection φ0 . The kink
band develops by prescribing a rotation increment φ̇ to the fibres within the kink band. Two
material states are present: a state inside the kink band (band material) and a state outside
the kink band (base material). Equilibrium of stresses prior to deformation together with
equilibrium of traction increments is required on the band boundary between the base and
kinked material. Furthermore, displacement continuities are required across the band during
deformation.
44 Chapter 3 - Semi-analytical kink band model

o
σ22
o
σ12

o
tα σ12
βi nα
xi2 o
σ11
xi1
o
σ11
o βo
σ12
φ

o
σ12
o
σ22
o
σ22
o
σ12

o
σ12
c βo
σ 12 c o
σ 11 σ11
c
i βi σ 11 c
σ 11
i
σ 12
σ 12

i
σ 12
i
σ 22

Figure. 3.11 Kink band geometry showing the two material states with two coordinate system (top).
Two small pieces of the material states together with the stresses on the band (bottom).
3.4 Kink band model and numerical scheme 45

3.4.1 2D model
An illustration of a two-dimensional kink band is presented in Fig. 3.11. Two material
states are shown. The superscripts (•)o and (•)i are associated with the outer base and inner
kink band material, respectively. Furthermore, a superscript (•)c is introduced which in
this section refers to the common system located at the band. The stress states in the three
systems can be observed in the bottom part of Fig. 3.11.
A rotation increment φ̇ is prescribed to the kink material system. The constitutive
formulations require that the x1 -axis of the coordinate systems are aligned with the fibres
prior to the deformation increment. Rotation of the coordinate systems is therefore required
during deformation. There are two ways of implementing the rotation: the coordinate systems
stay stationary during the rotation increment; the coordinate systems follow the rotation.
Both implementations are explored by Jensen and Christoffersen (1997) and give identical
results. The implementation where the coordinate systems follow the rotation is used in [P1 -
P4]. Using this approach, rotation of the stresses and instantaneous moduli is avoided. When
the coordinate systems follow the rotation of the fibres, then

vo2,1 = 0 vi2,1 = 0 (3.50)

are the consequences upon the velocity gradients. The velocity gradients wii, j observed in a
system that stays stationary during a rotation increment are

wii, j = vii, j + ωi j (3.51)

where ωi j is the relative spin tensor with the components

ω21 = −ω12 = φ̇ ω11 = ω22 = 0 (3.52)

In the latter equation, the prescribed spin increment φ̇ is introduced. Two tensors Rii j and Roij
that contain the directional cosines are introduced

Rii j = cos ∡(Oxic , Oxij ) Roij = cos ∡(Oxic , Oxoj )

(3.53)

where Rii j and Roij give the directional cosines between the inner kink band system xii or outer
system xio to the common system xic . Continuity of velocities across the band requires that a
line element lying on the band is stretched vc2,2 and rotated vc2,1 equally, when it is calculated
from the two systems. The directional cosines can be used to evaluate the velocity gradients
46 Chapter 3 - Semi-analytical kink band model

in another system. Continuity of velocities requires

wii, j Ri2i Ri2 j = voi, j Ro2i Ro2 j wii, j Ri1i Ri2 j = voi, j Ro1i Ro2 j (3.54)

In addition, the equalities can be expressed using the normal and tangent to the band

wii, j tii t ij = voi, j tio t oj wii, j nii t ij = voi, j noi t oj (3.55)

where noi and tio are the unit normal and tangent expressed in the outer base material system.
Equilibrium of traction increments require the tractions t˙11c and t˙c , expressed in the two
12
systems to be equal

tii j Ri1i Ri1 j = tioj Ro1i Ro1 j tii j Ri1i Ri2 j = tioj Ro1i Ro2 j (3.56)

which can be expressed with velocity gradients using Eq. (3.14)

Cii jkl wil,k Ri1i Ri1 j = Ciojkl vol,k Ro1i Ro1 j Cii jkl wil,k Ri1i Ri2 j = Ciojkl vol,k Ro1i Ro2 j (3.57)

When shear strains or stresses are present in the base material, the band will rotate. The band
rotation increments are

β̇ o = −voi, j Ro1i Ro2 j β̇ i = −vii, j Ri1i Ri2 j (3.58)

The necessary constraints are given by Eqs. (3.50), (3.54) and (3.57), which gives six
equations. The velocity gradients of the inner and outer material contain eight components
in total that need to be specified in every increment. Furthermore, two boundary conditions
are required. In the majority of the simulations presented in [P1 - P4], the following two
boundary conditions are used: no shear and transverse stresses are present in the base material

o o
σ̇22 =0 σ̇12 =0 (3.59)

The Cauchy stress increments σ̇αβ are calculated based on

σ̇αβ = Cαβ γδ vδ ,γ − σαβ vγ,γ + vα,γ σγβ (3.60)

o in the outer base material can be determined in every increment. The

Using the model, σ11
maximum value obtained for σ11 o during the fibre increments φ̇ is the critical compressive
cr
stress σ11 .
3.4 Kink band model and numerical scheme 47

3.4.2 3D model
A three-dimensional numerical formulation is developed in [P1] to study kink band formation
and to validate the developed three-dimensional homogenised constitutive equation. The nu-
merical scheme is inspired by the two-dimensional version and is generalised. An illustration
of a developed three-dimensional kink band is presented in Fig. 3.12. Again, two material
states are present and a common system lies on the band. The stresses in the systems are
illustrated at the bottom of the figure.
The velocity gradients vi, j in 3D contain nine components. Since two material states are
present, 18 components need to be determined in every increment and therefore 18 equations
are required. For simplicity and for the comparison with the two-dimensional formulation, in
[P1] we chose to let the kink band develop in the x1 -x2 plane. Furthermore, the x3o and x3i
bases coincide before and during deformation. A formulation where the coordinate systems
follow the rotation is once again chosen:

vo2,1 = 0 vi2,1 = 0 (3.61)

Continuity of velocities across the band require six equations in 3D and again, the directional
cosines are used to relate the velocity gradients to the common system

wii, j Ri2i Ri2 j = voi, j Ro2i Ro2 j wii, j Ri3i Ri3 j = voi, j Ro3i Ro3 j
wii, j Ri1i Ri2 j = voi, j Ro1i Ro2 j wii, j Ri1i Ri3 j = voi, j Ro1i Ro3 j (3.62)
wii, j Ri3i Ri2 j = voi, j Ro3i Ro2 j wii, j Ri2i Ri3 j = voi, j Ro2i Ro3 j

The first two equations ensure stretching continuity on the band. The third and fourth
equations ensure continuity of out-of-plane rotations of the kink band plane. The fifth and
sixth equations ensure continuity of shear straining and rotation of the kink band plane.
Continuity of traction increments requires three equations:

Cii jkl wil,k Ri1i Ri1 j = Ciojkl vol,k Ro1i Ro1 j

Cii jkl wil,k Ri1i Ri2 j = Ciojkl vol,k Ro1i Ro2 j (3.63)
Cii jkl wil,k Ri1i Ri3 j = Ciojkl vol,k Ro1i Ro3 j

The equalities given by Eqs. (3.61), (3.62) and (3.63) are necessary and give eleven equations.
The last seven equations set the boundary conditions. Five boundary conditions are used in
[P1] to enforce plane strain conditions
48 Chapter 3 - Semi-analytical kink band model

βi xi2
xi1

βo
xi3 φ

x2o

x1o
o
x3o σ22
o
σ12
o
σ23 o
c σ12
σ 12 c
σ 11
c βo o
σ11
σ 13
i
σ 13 c o
σ 13 σ13
βi c
σ 11 c
σ 12
i
σ 11
i i
σ 12 σ 23
i
σ 12
i
σ 22

Figure. 3.12 Kink band geometry showing the two material states with two coordinate systems in a
three-dimensional case (top). Two small pieces of the material states together with the stresses on the
common plane (bottom).
3.4 Kink band model and numerical scheme 49

vo3,1 = 0 vo3,2 = 0 vo1,3 = 0

(3.64)
vo2,3 = 0 vo3,3 = 0

In [P1], the influence of multi-axial loading is explored. Two variables ρ and ψ are introduced,
to control the proportional amount of multi-axial loading.
o
σ̇22 o
σ̇12
ρ= o ψ= o (3.65)
σ̇11 σ̇11

The last two boundary conditions for the three-dimensional scheme are

o o o o
σ̇22 − ρ σ̇11 =0 σ̇12 − ψ σ̇11 =0 (3.66)

The Cauchy stress increments σ̇i j are calculated based on

σ̇i j = Ci jkl vl,k − σi j vk,k + vi,k σk j (3.67)

With the introduced formulation, it is possible to compare the two- and three-dimensional
kink band models using the homogenised constitutive formulations.

3.4.3 Numerical scheme

Many equations were presented and to enhance the transparency of the implementations,
the steps within the numerical scheme are outlined here. The relative spin increment φ̇ is
prescribed. The simulations progress from an initial fibre imperfection φ0 to an end rotation
φ decided by the user. The following steps are required for every increment:
1. Update instantaneous moduli for the constituents Licjkl using an elastic or elastic-plastic
time-independent material law.
2. Calculate the nominal moduli for the constituents Cicjkl using Eq. (3.13).
3. Calculate the nominal moduli for the composite Cαβ γδ or Ci jkl based on Eq. (3.27),
Eq. (3.40) or Eq. (3.39).
4. A linear system in the form [A] V = R is created, where V is a vector containing the
unknown velocity gradients for the band and base material vii, j , voi, j and [A] is a matrix
containing the coefficients. The linear system is solved with respect to the unknown
vector V. The linear system is based on
(a) In 2D – 8 equations given by Eqs. (3.50), (3.54), (3.57) and (3.59) using
Eq. (3.60).
50 Chapter 3 - Semi-analytical kink band model

(b) In 3D – 18 equations given by Eqs. (3.61), (3.62), (3.63), (3.64) and (3.66) using
Eq. (3.67).
5. Calculate the velocity gradients for the constituents vci, j within the band and base
material based on the overall velocity gradients vi, j using Eq. (3.29) in 2D. The
velocity gradients in the 3D case are presented in [P1].
6. Update the stresses σi j = σi j + σ̇i j where the stress increments σ̇i j are calculated using
Eq. (3.60) or (3.67).
7. Update the geometric quantities β , φ and c f using Eq. (3.58) and Eq. (3.30).

3.5 Kink band broadening

The propagating instability known as kink band broadening or steady-state kink band prop-
agation is investigated in [P3] and [P4]. In the publications, the semi-analytical model
presented in this chapter is used. Propagating instabilities initiate from local imperfections
and propagate under a constant applied load, that can be several magnitudes lower than
the load required to initiate the instability as explained by Kyriakides (1993). Within the
framework of fibre-kinking, a kink band will initiate and rotate until a steady-state is reached
where the fibres stop rotating and broadening of the kink band is energetically preferable.
In the current section, a method is introduced to study kink band broadening using the
semi-analytical model that is developed by Jensen (1999a). The methodology is based on a
Maxwell construction of work equilibrium. Chater and Hutchinson (1984) proposed a similar
methodology to study steady-state propagation in structural applications. The internal work
W I and external work W E per unit volume are calculated in every increment and when the
equilibrium W I = W E is obtained, the steady-state is obtained. The external work per unit
volume is calculated according to

W E = Sαβ
o o
∆Eαβ (3.68)

where ∆Eαβ o are the differences in Lagrangian strains inside and outside the kink band
o are stress components of the work-conjugate of
evaluated in the base material system. Sαβ
the Lagrangian strain i.e. Second Piola Kirchhoff stresses. The internal work per unit volume
is calculated based on the integral
Z εi
WI = σαβ dεαβ (3.69)
εo
3.5 Kink band broadening 51

where ε o and ε i are strain states in the base and kink band material. The simulations
initiate from a small fibre imperfection φ0 ≈ 0 and progress with fibre rotation increments
φ̇ . The simulations run until the work equilibrium W I = W E is reached. Further details and
illustrations are presented in [P3] and [P4].
Chapter 4

Analytical kink band model

The phenomenon known as kink band broadening was briefly introduced in the previous
chapter. The concept will be further discussed in the current chapter using an analytical
model developed in [P3] and is used in a similar form in [P4]. Regarding an idealised
kink band, the analytical model assumes stress equilibrium on the band and displacement
continuity across the band. In the current chapter, it is formulated using the total stress
and strain states instead of a rate formulation. Similarly, a Maxwell construction of work
equilibrium is assumed, resulting in one transcendental equation with the fibre rotation φ
being the unknown variable. Moran et al. (1995) derived a similar analytical expression of
kink band broadening where they used the experimentally observed relation φ = 2β from
Evans and Adler (1978). Furthermore, Moran et al. (1995) assumed incompressibility of the
composite. In contrast, the analytical model developed in [P3] does not assume φ = 2β but
does account for compressibility. In the current formulation the kink band inclination β is
seen as a free parameter. The final value of β during kink band broadening is discussed in
[P4] with the hypothesis that β is influenced by the boundary conditions. In contrast to the
semi-analytical model (Chapter 3), the analytical model makes several prior assumptions
regarding e.g. the fibre and matrix material behaviour and the overall stress state in the base
material.

4.1 Stresses and equilibrium

Two material states are considered: an outer base material and an inner kink band material.
This is equivalent to the assumptions introduced for the semi-analytical model. Equilibrium
of stresses is required on the band between the base and band material, which gives a statically
54 Chapter 4 - Analytical kink band model

admissible stress field. To have equilibrium of stresses, both stress states are evaluated in the
same coordinate system as shown in Fig. 3.11. The directional cosines are used once again:

σiij Ri1i Ri1 j = σioj Ro1i Ro1 j σiij Ri1i Ri2 j = σioj Ro1i Ro2 j (4.1)

c , expressed through the stress field from the band

where the first equation gives the stress σ11
material (left of equality) and the outer material (right of equality). Similarly, the second
equation equals the shear stress σ12 c in the common system. In the analytical model, it is
o
assumed that only σ11 is present (stress along the fibres) in the base material. In [P3] it is
demonstrated that the stress σ11o can be expressed as a function of the inner stresses σ i and
12
i according to
σ22

o i i
σ11 = −σ12 [cot φ + tan β ] − σ22 [cot φ tan β − 1] (4.2)

where the band inclination β is regarded as a free parameter that needs to be specified. In
[P3], simple assumptions are made regarding the constitutive behaviour and are inspired
by the stress-strain response in the semi-analytical model. It is assumed that the stress σ22
can be determined from the total strain ε22 alone and that the response is linear-elastic. The
expression
E (1 − ν)
σ22 = ε22 (4.3)
(1 + ν) (1 − 2ν)
is used with the assumption of plane strain conditions, where E is the transverse composite
Young’s modulus and ν is the Poisson’s ratio. A simple bilinear relation is assumed for the
shear stress-strain behaviour:

y
2 G1 ε12

 for ε12 < ε12
σ12 =   (4.4)
y G2
σ12 = 2 G2 ε12 + σ12 1 −

 otherwise
G1

where G1 is the initial shear modulus prior to yielding and G2 is the shear modulus beyond
y y
yielding. σ12 and ε12 are the shear yield stress and shear yield strain.

4.2 Strains and continuity

In the current section, the strain state εi j in the base and band material is determined as a
function of fibre rotation φ . Fig. 4.1 will be used to illustrate the kinematic requirements
on the kink band boundary. Initially, the focus is given to the two points lying on the band
boundary (black circles in Fig. 4.1). It is assumed that there is a compatible strain field prior
4.2 Strains and continuity 55

c xo2
x 2 xc 1
c
x 2 xc 1 xo1
i
Initial x2 i
x1

i,c o,c
v 1,2 v 1,2

i,c o,c
v 2,2 v 2,2
Deformed

Figure. 4.1 Two points on kink band boundary seen from the two material states (top). Initial (black)
and deformed configuration (grey) of the two points on the band boundary seen from the two states
(bottom).

to deformation i.e. material lines lying on the band boundary have stretched and rotated the
same amount when expressed through the inner and outer strains. To have compatibility
during a deformation increment, material lines on the band boundary need to have the same
stretch and rotation increment when expressed through the strains in the two systems. This
requirement is illustrated in Fig. 4.1, where the grey circles represent a material piece lying
on the boundary after a deformation increment.
The velocity gradient vo,c 1,2 represents a velocity gradient in the common system (x )
c

and is calculated based on the velocity gradients voi, j in the outer material system. In the
analytical model, the fibres are assumed to be inextensible and when the fibres are aligned
with the x1 -axis during deformation, this entails vo1,1 = vi1,1 = 0. Furthermore, only σ11 stress
is applied along the fibres in the base material, which implicitly yields no transverse and
shear straining voi, j = 0 in the base material. Continuity of displacements across the band is
formulated using velocity gradients

wii, j Ri2i Ri2 j = 0 wii, j Ri1i Ri2 j = 0 (4.5)

56 Chapter 4 - Analytical kink band model

β o = 20◦

2 ε22
ε12 = φ /2

1 1
φ = 60◦ 2

Figure. 4.2 Strain state in outer material 1 and inside kink band 2 (left). The strain state inside the
kink band is rotated 60◦ to show the total strains (right).

where the velocity gradients wii, j are given by Eq. (3.51). In [P3] it is demonstrated that the
finite strains in the band material can be derived based on the latter equations, and are equal
to
 
i φ i cos (β − φ )
ε12 = ε22 = ln (4.6)
2 cos β
The same expressions of the strains were obtained by Budiansky and Fleck (1993). To
enhance understanding of the final strain states, Fig. 4.2 is introduced in [P3]. A material
piece of the base material is designated as 1 and a piece in the band material is designated
as 2 . In the initial configuration, the material piece 2 is equal to 1 . To illustrate the total
deformation going from 1 to 2 , the piece 2 is rotated an amount of φ in the right-hand
i , zero
illustration in Fig. 4.2. By inserting the kinematic relation φ = 2 β into Eq. (4.6) for ε22
i
transverse straining ε22 = 0 is obtained.

4.3 Kink band broadening

An illustration of a kink band that has undergone broadening is displayed in Fig. 4.3. It is
assumed that a steady-state is obtained when the work done by external forces equals the
work required to deform an infinitesimal piece into the band. The externally applied stress
σ11 works on an area A. The specimen boundary displaces at the uniform rate −u. Positive
values work in the direction of the basis. The work done by the external force is

W E = −u A σ11 (4.7)
4.3 Kink band broadening 57

σ11
φ

σ11
β u
x2o
l

x1o

Figure. 4.3 Illustration of the model of kink band broadening. A propagated configuration is illustrated
with dashed lines.

Using the geometric relation u = l (1 − cos φ ), see Fig. 4.3, the relation can be written in the
form:
W E = −l A σ11 (1 − cos φ ) (4.8)

During the uniform rate −u, a material volume of l A undergoes internal work equal to
Z εi
WI = l A σαβ dεαβ (4.9)
εo

When comparing the internal and external work, the volume l A is present in both equations.
The stresses and strains within the equations of work can be substituted by the expressions
presented in the two previous sections. [P3] demonstrates that the internal and external work
per unit volume can be expressed by

W E = (2C2 φ +C3 ) [cot φ + tan β ] (1 − cos φ )


cos (β − φ )
 (4.10)
+ 2C4 ln [cot φ tan β − 1] (1 − cos φ )
cos β

cos (β − φ ) 2
   
W I = C1 (φ y )2 +C2 φ 2 − (φ y )2 +C3 (φ − φ y ) +C4 ln (4.11)
cos β
58 Chapter 4 - Analytical kink band model

where constants are introduced as follows:

 
G1 G2 y G2 E (1 − ν)
C1 = C2 = C3 = σ12 1 − C4 = (4.12)
2 2 G1 2 (1 + ν) (1 − 2ν)

The latter constants are based on material properties of the composite and as mentioned,
the band inclination β is a free parameter. The only unknown variable upon evaluating the
work balance W E = W I is the fibre angle φ . The work balance during steady-state kink band
broadening is determined numerically.
Chapter 5

Summary and outlook

The dissertation is chiefly concerned with the study of kink band formation, which is the
dominant compressive failure mechanism for many fibre-reinforced composites loaded in the
direction of the fibres. The study covers both numerical and analytical approaches. Several
numerical models and constitutive relations were developed during the project.
Secondly, contributions to the field of joint mechanics within composites were made and
were carried out at the Department of Engineering at the University of Cambridge.
The contributions to the field are elaborated in the following section.

5.1 Contributions
Regarding the failure mechanism, kink band formation, this project has contributed with:
• An exploration of kink bands using an idealised three-dimensional, semi-analytical,
kink band model.
• Investigations of the effect of multi-axial loading upon the critical peak strength for a
case study involving a carbon fibre-reinforced composite.
• An investigation of the influence of fibre-to-matrix bonding quality upon the critical
peak strength at kink band formation.
• A developed understanding of the propagating instability known as kink band broad-
ening, including the influence of relevant material parameters upon the lock-up angle
and propagating stress.
• An investigation of size effects upon kink band broadening using an FE-model.
60 Chapter 5 - Summary and outlook

• An investigation of kink band formation in open-hole fibre-composites, including the

sensitivity of loading direction upon compressive strength.

As part of the project, the following models were developed:

• A three-dimensional constitutive model that can be used to explore plastic deformation
with large rotations and deformations for fibre-reinforced composites. The model is
based on two independently described constitutive formulations for the constituents.
• A two-dimensional constitutive model that includes information about fibre-to-matrix
bonding. Once again, this model is based on two independently described constitutive
relations of the constituents.
• A three-dimensional, unit-cell finite element model where the fibre and matrix con-
stituents are modelled explicitly. The model makes it possible to investigate the effect
of the fibre-to-matrix interaction upon kink band formation.
• A two-dimensional finite element model with periodic boundary conditions on a
skewed mesh, used to study kink band broadening.
• An analytical model for the estimation of kink band broadening. The model was based
on a Maxwell condition for the steady-state.

As part of the visit to the Department of Engineering at the University of Cambridge, the
following contributions were made regarding the investigation of the load transfer mechanism
in a Dyneema composite:
• Construction of an experimental set-up to study the load transfer mechanism.
• Identification of the dominant failure mechanism through X-ray tomography and
optical images. It was found that the failure was dominated by inter-laminar shearing
between the 0◦ and 90◦ fibre layers.
• Development of a simple analytical model of the dominant failure.
• Construction of a failure mechanism map with the specimen geometries as the inde-
pendent variables.

5.2 Perspective
In the previous section, all the contributions were stated as bullet points and in the following,
the contributions are put into perspective for real applications.
The development of the three-dimensional constitutive model makes it possible to simu-
late the failure of real three-dimensional structures made from composite materials. Previous
3D models were based on the measured homogenised response of the overall composite. The
5.3 Recommendations for further work 61

current model takes two independently described materials and homogenises the properties
instead of taking the overall measured properties. Furthermore, no information is lost with
respect to the stress and strain states of the individual constituents during deformation.
The results obtained using the two-dimensional constitutive model taking imperfect
bonding into account can be used to give an upper and lower bound for the actual compressive
strength of fibre composites. The actual strength lies between the two boundaries of perfect
and no bonding. The latter model can be implemented as a user subroutine in Abaqus/CAE.
This gives the possibility of having areas of imperfect bonding as e.g. around drilled open-
holes in composite laminates where the composite will be damaged in the vicinity of the hole
due to the machining process.
A laminated composite structure will, in general, be exposed to different multi-axial
stress states throughout the material. Previous studies have mainly focused on idealised cases
with only stresses acting in the direction of the fibres. The results obtained with respect to
multi-axial loading can be used to estimate the actual compressive strength of a composite
with a realistic stress state.
As mentioned previously, the stress at steady-state kink band broadening can be seen
as the lower bound of the compressive strength. In structures where the design needs to be
insensitive to fibre-imperfections, parametric studies with respect to kink band broadening
can be used to develop new composite materials with improved properties with compressive
failure in mind.

5.3 Recommendations for further work

The three-dimensional constitutive model developed in [P1] was used to study kink band
formation using an idealised model. The constitutive relation is formulated in a general
framework and can be implemented as a user subroutine in a finite element software as e.g. a
UMAT into Abaqus/CAE. This enables the study of failures of complex structures and fibre
layouts where plasticity plays a crucial role.
The constitutive formulation developed in [P2] uses a phenomenological parameter µ
going from zero to unity to specify the quality of fibre-to-matrix bonding. A connection is
missing between the phenomenological parameter µ and the measured fracture toughness
obtained from single-fibre push-out experiments e.g. Jäger et al. (2015).
Hybrid composites are currently being developed i.e. composites with more than two
constituents; for example, glass and carbon fibres embedded in a resin material. The company
Owens Corning has produced glass fibre laminates since 1944 and has recently started
producing hybrid laminate composites composed of glass and carbon fibres. Limited work
has been done within the field of compressive failure of hybrid composites.
Chapter 6

P1
Three-dimensional constitutive model for elastic-plastic
behaviour of fibre-reinforced composites

6.1 Motivation
As described in the Introduction, the estimation of the behaviour of fibre composites under
large deformations, including plasticity, can be challenging due to the heterogeneous structure
of the composites. Both discrete and homogeneous finite element models can be used to study
the deformation as explained in Chapter 2. The use of discrete models, where the fibre and
matrix constituents are modelled explicitly, is limited to simple and small geometries. When
the fibre layout becomes complex, the use of homogeneous models become advantageously.
Previous developed homogeneous, elastic-plastic, formulations have mainly been based on
the composite response instead of the constitutive response of the constituents.
The current publication regards the development of a three-dimensional constitutive
formulation for fibre-reinforced composites. The formulation takes as input, two constitutive
relations of two constituents and the volume fractions of the constituents. The formulation
can be used in a general elastic-plastic framework with large displacements and rotations.
The formulation is inspired by the two-dimensional constitutive model developed by Christof-
fersen and Jensen (1996). The original source of the publication is: (Skovsgaard and Jensen,
2018b).
64 Chapter 6 - P1

6.2 Method
The homogenization of the constituents is based on three assumptions
• Material lines on the common plane between two materials are subjected to common
stretching and rotation.
• Planes on the intersection between two materials transmit identical tractions.
• The material of the constituents is elastic of elastic-plastic.
The thoughts and theory behind the formulation were described in Chapter 3 and the
essence is repeated here for completeness. The model is formulated in an updated Lagrangian
framework using velocity gradients and rates of nominal stresses. The final three-dimensional
constitutive relation is based on two homogenisations to obtain a fibre that is embedded in
a matrix. The relation is implemented and used in a semi-analytical kink band simulation
using Fortran 77 programming. The model implementation is described in Section 3.4.

6.3 Contribution
The contributions to the publication comprise: the development of a novel constitutive for-
mulation; expansion of a two-dimensional kink band model to a three-dimensional version;
conducting kink band simulations with and without multi-axial loading and writing the paper.

6.4 Main findings

One of the main findings is the development of the constitutive formulation itself. Several
different versions of the three-dimensional relations are proposed and results from a kink
band simulations using the models are compared with the results from a two-dimensional
model introduced by Jensen and Christoffersen (1997). The influence of initial fibre im-
perfection upon the critical compressive strength is investigated using the different models.
Furthermore, the influence of transverse and shear loading upon the critical compressive
stress is investigated using the two- and three-dimensional relations.
The developed constitutive formulation is used in a kink band study using a semi-
analytical approach, but is far from limited to this application. The formulation can e.g. be
implemented as a user subroutines into Abaqus/CAE and be used to study the response and
failure of complex fibre domains.
 6.5 Article P1 65

International Journal of Solids and Structures 139–140 (2018) 150–162

Contents lists available at ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

Three-dimensional constitutive model for elastic-plastic behaviour of

fibre-reinforced composites
Simon P.H. Skovsgaard∗, Henrik Myhre Jensen
Department of Engineering, Aarhus University, Inge Lehmanns Gade 10, Aarhus C 8000, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: A formulation of a three-dimensional homogenized constitutive model is developed that can predict
Received 8 August 2017 strain localization in fibre-reinforced composites. A constitutive equation for the model is developed in
Revised 23 January 2018
a general rate form based on independent constituent behaviour which can be either elastic or elastic-
Available online 2 February 2018
plastic. The performance of the constitutive model is investigated by performing an infinite kink band
Keywords: analysis. The development of a kink band is investigated for several initial imperfections and compared
Fibre reinforced with previously published results.
Anisotropic
© 2018 Elsevier Ltd. All rights reserved.
Constitutive laws
Finite deformation
Instability

1. Introduction tic matrix material. The failure was caused by kink band forma-
tion in a geometry with an initial fibre imperfection. As a contin-
The increased use of fibre composites has given rise to a thor- uation, Budiansky and Fleck (1993) included the effects of finite
ough investigation of the behaviour and failure of composites fibre stiffness. They concluded that the assumption of rigid fibres
based on several load types. Especially the compressive failure has could be justified for many kinking problems. It is now commonly
received attention due to its sudden instability which can cause accepted that the micro-buckling compression failure is primarily
a catastrophic failure. Compressive failure of fibre composites is dominated by fibre imperfections and the elastic-plastic shear be-
caused by several competing failure mechanism. Several experi- haviour of the composite.
ments show that a frequently observed failure is due to plastic Numerical investigations of the failure of long fibre-reinforced
micro-buckling which leads to the development of kink bands. composites can roughly be divided into two categories. The first
Wadee et al. (2004) observed plastic micro-buckling in a compres- approach is by creating a finite element analysis where fibre and
sion experiment and compared the results with a simple mechan- matrix are treated individually. This approach is commonly known
ical model. Liu et al. (2014) investigated the collapse of composite as discretized models or individual fibre models. The approach
beams made from ultra-high molecular polyethylene fibres where was initially used by Guynn et al. (1992) and several authors
they observed kink band formation. Zhou et al. (2013) investigated have used a similar approach, including: Vogler et al. (2001a) who
the compressive response of unidirectional fibres under axial and investigated the compressive response due to local imperfec-
off-axis compressive loading and kink band formation dominated tion near a free edge and was compared with own experiments
the failure for small off-axis angles. Nizolek et al. (2017) investi- presented in Vogler et al. (2001b). Romanowicz (2013) inves-
gated stable kink band propagation in Cu-Nb nanolaminates. They tigated both uniform and non-uniform sinusoidal imperfections.
observed that kink bands were initiated and propagated from local Wind et al. (2014) compared a discretized and homogenized model
imperfections. in a kink band study. Hsu et al. (1999a) conducted an experi-
In the early work of micro-buckling (Rosen, 1965) investi- ment where they investigated steady-state axial propagation of
gated the compressive strength of a polymer matrix composite kink bands and compared their findings with a three-dimensional
where the fibres were treated as beams on an elastic foundation. discretized finite element model presented in Hsu et al. (1999b).
Budiansky (1983) introduced a composite with elastic ideally plas- Recent investigations using the current approach with random fi-
bre distribution were performed by Zhang et al. (2016). More com-
prehensive methods including damage and cohesive friction ele-
∗ ments were studied by Naya et al. (2017). The use of discretized
Corresponding author.
E-mail addresses: [email protected] (S.P.H. Skovsgaard), [email protected] (H.M. finite element models can be challenging and time consuming
Jensen). when concerning large or complex geometries. Especially the deci-

https://doi.org/10.1016/j.ijsolstr.2018.01.032
0020-7683/© 2018 Elsevier Ltd. All rights reserved.
66 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 151

Fig. 1. Geometries for the constitutive formulations.

sions on how the direction should be varied for every single fibre 2. Geometrical representation of constitutive model
are complicated when the geometry becomes comprehensive. The
method, where the fibres are discretized and are embedded in a In this section, a geometric representation of the representative
matrix material, is limited to simple geometries. volume elements is shown and described. The constitutive model
A second approach is by homogenizing or smear out the is based on similar assumptions used to derive the model used
properties of the fibre and matrix constituents. Early formu- in Christoffersen and Jensen (1996). Their constitutive model was
lations of two-dimensional constitutive models were presented based on a two-dimensional domain of fibres and matrix and was
by Fleck and Shu (1995) and Christoffersen and Jensen (1996). formulated based on three assumptions outlined here for com-
Fleck and Shu (1995) formulated their constitutive model using pleteness
Cosserat couple stress theory, where they were able to include fi-
bre bending effects. Christoffersen and Jensen (1996) formulated a 1. Material lines parallel with the fibres are subject to a common
general rate constitutive equation where the homogenized prop- stretching and rotation.
erties were based on independent constitutive laws for both con- 2. Planes parallel with the fibres transmit identical tractions.
stituents (matrix and fibre). Poulios and Niordson (2016) combined 3. The material of the constituents is elastic or elastic-plastic.
the qualities of the two latter models. They developed a two-
The first assumption correspond to a Voigt estimate in the fibre
dimensional constitutive model which is described by two inde-
direction and is commonly accepted for fibre-reinforced compos-
pendent constitutive laws for the constituents and is able to in-
ites. The second assumption implies that the shared plane between
clude intrinsic size effects using higher order strain gradients.
the fibre and matrix transmit the same traction and thereby is in
The present paper introduces a general three-dimensional con-
equilibrium across the boundaries. The third assumption is a spec-
stitutive model which is largely inspired by the formulation done
ification of the time-independent materials for the constituents.
by Christoffersen and Jensen (1996). Only few attempts have
The current model uses both the first and third assumption di-
been done to introduce three-dimensional effects into a gen-
rectly and partially the second assumption which will be outlined
eral framework to simulate the compressive failure of compos-
later. An illustration of the representative volume elements (RVE)
ites. Grandidier et al. (1992) introduced three-dimensional ef-
is shown in Fig. 1. Two different geometries are shown which the
fects to investigate elastic micro-buckling by introducing a non-
derivations are based on. Common for both geometries is a fibre
uniform distribution of the displacements through the thickness.
constituent f and two matrix constituents m1 and m2 , representing
Gutkin et al. (2016) investigated kink bands under 3D stress states
a fibre surrounded by a resin material. Two length scales are in-
using a damage model created in Gutkin and Pinho (2015).
troduced wf and hf which is the width and height of the simplified
Three-dimensional constitutive models including plasticity is
fibre geometry. In the case of a uniform distribution of fibre and
needed to investigate the failure of more complex geometries and
matrix, the length scales can be calculated based on
fibre layouts created by fibre-reinforced composites. The current
paper introduces a three-dimensional homogenized constitutive
p
wf = hf = cf (1)
model which is based on independent constitutive laws for the
constituents. The model is created in a rate form using nominal cf
where is the volume fraction of fibres. The fibre and matrix vol-
stresses. The model is based on kinematic and static continuities ume fraction cf and cm comply to the identities
and are averaged over a representative volume element. Lastly, the
constitutive model is investigated using an infinite kink band anal- cf = wf hf
ysis and is compared with results obtained by Jensen and Christof- c m = ( 1 − w f )h f + ( 1 − h f ) = ( 1 − w f ) + ( 1 − h f )w f (2)
fersen (1997) and Jensen (1999).
1 = c f + cm
The paper is organized in nine sections. Initially in Section 2,
an overview of the representative volume elements is presented The only difference between the two geometries shown in
which the constitutive model is based on. The general relations Fig. 1 is the volume fractions of m1 and m2 . When the mod-
are introduced in Section 3. Section 4 together with Section 5 de- els are derived, they behave orthotropically. Unidirectional fibre
rives the constitutive model. Section 6 presents the formulation of composites with fibres aligned in the x1 direction are usually as-
the infinite kink band simulation. In Section 7, the constitutive law sumed to have transverse isotropic behaviour. To obtain a trans-
used for the constituents is presented. Section 8 gives an overview verse isotropic behaviour of the constitutive model, a combination
of the steps required in the numerical simulation. Section 9 shows of the properties between the two geometries, illustratively shown
results for kink band formation where different constitutive mod- in Fig. 2, is used.
els are compared. Section 10 concludes the paper. The constitutive models are constructed based on a homoge-
nization in two steps. For the first geometry shown in Fig. 1(a), the
 6.5 Article P1 67

152 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

Combining Eq. (5) and Eqs. (9)–(11) yields the relation between
nominal stress rate and velocity gradients

1 c 1 1 1
³ ´
t˙icj = Lci jkl + σ δ − σ c δ − σ c δ − σ c δ + σicj δlk vcl,k
2 ik jl 2 il jk 2 k j il 2 l j ik
(12)
where δ ij is Kronecker’s delta. Comparing the latter equation with
Eq. (4) results in the desired moduli
1 c 1 1 1
Fig. 2. Geometric interpretation of the average constitutive properties. Cicjkl = Lci jkl + σ δ − σ c δ − σ c δ − σ c δ + σicj δlk (13)
2 ik jl 2 il jk 2 k j il 2 l j ik
Using the nominal moduli Cicjkl gives the opportunity to include
initial homogenization is done between f and m1 . After the homog- residual stresses which were investigated in a two-dimensional
enization, the lower part of the geometry is treated as a homo- kink band study by Jensen (2002). In the derivations below, it is
geneous constituent which is illustrated in Fig. 3(a). In the same convenient to represent the stress rates and velocities in vectors.
manner, the second geometry is initially homogenized between f The nine components of the nominal stress rates are represented
and m2 as shown in Fig. 3(b). in the three vectors
t˙11 t˙21 t˙31
à ! à ! à !
3. General relations t˙ 1 = t˙12 t˙ 2 = t˙22 t˙ 3 = t˙32 (14)
t˙13 t˙23 t˙33
The constitutive model is based on a rate form of the constitu-
Each of the stress vectors represents the traction increment acting
tive behaviour of the three constituents. A formulation in the fol-
on the deformed configuration which initially was a unit cube. The
lowing form is pursued
vector v will denote the velocity vector
t˙i j = Ci jkl vl,k i, j, k, l = {1, 2, 3} (3) Ã !
v1
where t˙i j are components of the rate of nominal stresses, vl, k v= v2 (15)
are components of the velocity gradients and Cijkl are components v3
of the tensor of nominal moduli. The summation convention is Using the vectors above, the constitutive relation Eq. (4) for the
adopted for repeated index. Knowing the constitutive relations and composite and the constituents may be written as
the velocity gradients for the individual constituents, similar rela-
tions can be set up t˙ i = Ci j v, j i, j = {1, 2, 3}
(16)
t˙ c = Cc vc
i ij ,j
i, j = {1, 2, 3}
t˙icj = Cicjkl vcl,k i, j, k, l = {1, 2, 3} (4)
Where Cij and Cci j
represents matrices of the size 3x3. The matrices
1 2
here (•)c represent either (•)f , (•)m or (•)m . The relation between includes the constitutive moduli Cijkl and Cicjkl .
the nominal stress rate and the Cauchy stress rate is given by
4. Homogenization between arbitrary materials
t˙icj = σ˙ icj + σicj vck,k − σ jk
c c
vi,k (5)
It is assumed that the elastic-plastic relation between the rate of The homogenization between two material planes are based on
Cauchy stress and strain rate are given in the form the assumptions:

σ̊icj = Lci jkl ε˙ klc (6) 1. Material lines lying on the common plane between two mate-
rials are subjected to a common stretching and rotation.
here σ̊icj are components of the Jaumann rate of Cauchy stresses 2. Planes on the intersection between two materials transmit
which follows the local rotated coordinate system during defor- identical tractions.
mation. Lci jkl are components of the elastic-plastic tangent moduli 3. The material of the constituents is elastic or elastic-plastic.
which satisfies the symmetries
The first assumption results in 3 stretching continuities and 3
Lci jkl = Lci jlk = Lcjikl (7) rotation continuities. The second assumption expresses local equi-
librium.
which is required by objectivity. The strain rates can be calculated To obtain the homogenized models presented in Section 2, two
using the velocity gradients homogenization operations are performed. The homogenization is
1¡ c ¢ done for two materials who share a plane parallel with the x1 − x2
ε˙ icj = v + vcj,i (8) plane and another situation where they share a plane parallel with
2 i, j
the x1 − x3 plane.
Eq. (6) can be rewritten using the symmetries of Lci jkl combined
On Fig. 4 two materials a and b are shown who share a plane
with the latter equation
parallel with the x1 − x3 plane. As a consequence of the first as-
σ̊icj = Lci jkl vcl,k (9) sumption, the velocity gradients vc,1 and vc,3 are common for both
constituents based on the geometry shown in Fig. 4
The relation between the Jaumann rate of Cauchy stress and the
Cauchy stress rate seen from a stationary coordinate system is va,1 = vb,1 = v,1
(17)
given by va,3 = vb,3 = v,3

˙ icj c
σ = σ̊ + ω σ + ω σ
ij
c
jk
c
ik
c
ik
c
kj (10) where superscripts (•)a and (•)b represents the two arbitrary mate-
rials, and velocity gradients with no superscript are overall quanti-
where ωicj is the spin tensor given by ties. For overall compatibility of the representative volume element
1¡ c ¢
ωicj = v − vcj,i (11) ca va,2 + cb vb,2 = v,2 (18)
2 i, j
68 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 153

Fig. 3. Geometries after first homogenization.

Combining the latter equation with the first part of Eq. (20)
¡ ¡ ¢ ¡ ¢¢
t˙ 1 = ca Ca11 + cb Cb11 − ca cb Ca12 − Cb12 C∗−1
22
Ca21 − Cb21 v,1
¡ a
¡ ¢ ¡ ¢¢
+ c Ca12 +c b
Cb12 −c c a b
Ca12 − Cb12 ∗−1
C22 Ca22 − Cb22 v,2
¡ ¡ ¢ ¡ ¢¢
+ ca Ca13 + cb Cb13 − ca c Ca12 − Cb12 C22 b ∗−1
Ca23 − Cb23 v,3
(24)
Using the same arguments for the tractions t˙ c3 using the second
part of Eq. (20) yields
¡ ¡ ¢ ¡ ¢¢
Fig. 4. Representative volume element, where two arbitrary materials share a com- t˙ 3 = ca Ca31 + cb Cb31 − ca cb Ca32 − Cb32 C22
∗−1
Ca21 − Cb21 v,1
mon plane with outward normals which are parallel with the x2 -axis. ¡ ¡ ¢ ¡ ¢¢
a
+ c Ca32 +c b
Cb32 −c c a b
Ca32 − Cb32 C∗−1
22
Ca22 − Cb22 v,2
¡ ¡ ¢ ¡ ¢¢
+ ca Ca33 + cb Cb33 − ca cb Ca32 − Cb32 C22
∗−1
Ca23 − Cb23 v,3
where ca and cb are the length scales presented in Fig. 4 which
also are the volume fractions of the two materials. (25)
The second assumption leads to the equality
By inserting the velocity gradients given by Eq. (21) into
t˙ a2 = t˙ b2 = t˙ 2 (19) Eq. (19) results in two expressions for the traction t˙ 2
¡ ¡ ¢¢
Overall equilibrium of the RVE entails t˙ 2 = Ca21 − cb Ca22 C22
∗−1
Ca21 − Cb21 v,1
¡ ¡ ¢¢
ca t˙ a1 + cb t˙ b1 = t˙ 1 + Ca22 − cb Ca22 C∗−1
22
Ca22 − Cb22 v,2
(20)
ca t˙ a3 cb t˙ b3 = t˙ 3
¡ ¡ ¢¢
+ + Ca23 − cb Ca22 C∗−1
22
Ca23 − Cb23 v,3
¡ ¡ ¢¢ (26)
Based on Eqs. (17)–(20) the representative volume element will t˙ 2 = Cb21 + ca Cb22 C∗−1
22
Ca21 − Cb21 v,1
comply to equilibrium and compatibility. Combining Eqs. (17)–(20) ¡ ¡ ¢¢
together with Eq. (16), as detailed in Appendix A, yields the fol- + Cb22 + ca Cb22 C22 ∗−1
Ca22 − Cb22 v,2
lowing expressions for vc2
¡ ¡ ¢¢
+ Cb23 + ca Cb22 C22
∗−1
Ca23 − Cb23 v,3
1 ∗ Since the two previous solutions for t˙ 2 yield identical results, a
³¡ ¢ ³ ´ ¡ ¢ ´
va,2 = −cb C∗−1 Ca21 − Cb21 v,1 + Ca22 − Cb22 − C v,2 + Ca23 − Cb23 v,3
22
cb 22 combination of the two, multiplied by ca and cb will also give an
1 ∗
³¡ ³ ´ ´
vb,2 = ca C∗−1
¢
Ca21 − Cb21 v,1 + Ca22 − Cb22 + C
¡ ¢
v,2 + Ca23 − Cb23 v,3 identical solution
22
ca 22
t˙ 2 = ca t˙ 2 + cb t˙ 2 (27)
(21)
Using the latter property, the tractions t˙ 2 can be written as
Using Eq. (16) the traction forces for the constituents can be deter- ¡ ¡ ¢ ¡ ¢¢
mined based on the velocity gradients. Initially, the focus is on the t˙ 2 = ca Ca21 + cb Cb21 − ca cb Ca22 − Cb22 C22
∗−1
Ca21 − Cb21 v,1
tractions t˙ c1 ¡ a
¡ ¢ ¡ ¢¢
+ c Ca22 +c b
Cb22 −c c a b
Ca22 − Cb22 C∗−1
22
Ca22 − Cb22 v,2
t˙ a1 = Ca11 v,1 + Ca12 va,2 + Ca13 v,3
¡ a
¡ ¢ ¡ ¢¢
+ c Ca23 +c b
Cb23 −c c a b
Ca22 − Cb22 C∗−1
22
Ca23 − Cb23 v,3
(22)
t˙ b1 = Cb11 v,1 + Cb12 vb,2 + Cb13 v,3 (28)
Inserting the results for va,2 and vb,2 yields Comparing the coefficients from Eqs. (24), (25) to (28) with the
¡ ¡ ¢¢ desired relation given by Eq. (16), the matrices may be written in
t˙ a1 = Ca11 − cb Ca12 C22
∗−1
Ca21 − Cb21 v,1 the form
¡ ¡ ¢¢
+ Ca12 −c b
Ca12 ∗−1
C22 Ca22 − Cb22 v,2
¡ ¢ ¡ ¢
Ci j = ca Cai j + cb Cbi j − ca cb Cai2 − Cbi2 C22
∗−1
Ca2 j − Cb2 j i, j = {1, 2, 3}
¡ ¡ ¢¢
+ Ca13 − cb Ca12 C22 ∗−1
Ca23 − Cb23 v,3 (29)
¡ ¡ ¢¢ (23)
t˙ b1 = Cb11 + ca Cb12 C22
∗−1
Ca21 − Cb21 v,1 The constitutive equation gives the homogenized properties of the
¡ ¡ ¢¢ overall nominal tractions and overall velocity gradients. In an in-
+ Cb12 + ca Cb12 C22
∗−1
Ca22 − Cb22 v,2
¡ ¡ ¢¢ cremental scheme, the velocity gradients for the constituents can
+ Cb13 +c a
Cb12 C∗−1
22
Ca23 − Cb23 v,3 be obtained based on Eqs. (17) and (21). The previous derivations
 6.5 Article P1 69

154 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

share a common plane with outward normal parallel with the x3 -

axis and for this reason Eq. (34) is used

C˜ i j =w f Cifj +(1 − w f )Cm −w f (1 − w f )(Cif3 −Cim3 )C33 (C3f j −Cm

1 1 ∗−1 1
ij 3j
)
i, j = {1, 2, 3}
(36)
where C∗33 is given by
1 f
C∗33 = w f Cm f
33 + (1 − w ) C33 (37)
Fig. 5. Representative volume element, where two arbitrary materials shares a Based on the homogenized velocity gradients v˜ , j for the initial ho-
common plane with outward normals which are parallel with the x3 -axis. mogenization the velocity gradients for the constituents can be cal-
culated as
1
were based on a representative volume with two arbitrary mate- v,f1 = v,m1 = v˜ ,1
rials sharing a plane with an outward normal parallel to the x2 - 1
axis. In Fig. 5, two materials share a common plane with outward v,f2 = v,m2 = v˜ ,2
normal parallel to the x3 -axis. The derivation of the homogenized ³¡
1 1
v,f3 f f
¢ ¡ ¢
properties of such a geometry is similar to the previous and there- = −(1 − w f ) C∗−1
33 C31 m
− C31 v˜ ,1 + C32 m
− C32 v˜ ,2
fore details are omitted.
1
³ ´ ´
f 1
m
In this case the velocity gradients vc,1 and vc,2 are common for + C33 − C33 − C∗ v˜ ,3
1 − w f 33
both constituents ³¡
1 f 1 f
¢ ¡ 1 ¢
va,1 = vb,1 = v,1 v,m3 = w f C33
∗−1
C31 m
− C31 v˜ ,1 + C32 − Cm ˜ ,2
32 v
(30)
va,2 = vb,2 = v,2
1 ∗
³ ´ ´
f 1
m
For overall compatibility of the representative volume element
+ C33 − C33 + C v˜ ,3 (38)
w f 33
ca va,3 + cb vb,3 = v,3 (31) The last homogenization is between (•˜ ) and The current con- m2 .
The second assumption leads to the equality stituents share a common plane with an outward normal parallel
with the x2 -axis and therefore Eq. (29) is used
t˙ a3 = t˙ b3 = t˙ 3 (32)
2 2 2
C1i j =h f C˜ i j +(1 − h f ) Cm −h f (1 − h f ) C˜ i2 −Cim2 C˜ ∗−1 C˜ 2 j −C2mj
¡ ¢ ¡ ¢
By overall equilibrium ij 22
i, j = {1, 2, 3}
ca t˙ a1 + cb t˙ b1 = t˙ 1
(33) (39)
ca t˙ a2 + cb t˙ b2 = t˙ 2
where C˜ ∗22 is
Based on the assumptions and the equalities for the current ge-
2
ometry, the constitutive relations between nominal tractions and C˜ ∗22 = h f Cm f ˜
22 + (1 − h ) C22 (40)
velocity gradients is given by
¡ ¢ ¡ ¢ Based on the overall velocity gradients v, j the velocity gradients
Ci j = ca Cai j + cb Cbi j − ca c Cai3 − Ci3 C33
b b ∗−1
Ca3 j − Cb3 j i, j = {1, 2, 3} for the constituents can be calculated.
(34) 2
v˜ ,1 = v,m1 = v,1
The only difference between the latter constitutive equation and v˜ ,3 = v,m3
2
= v,3
Eq. (29) is the last part which is a consequence of the direction of
1
³¡ ³ ´
m2 2
C˜ 21 −C21 v,1 + C˜ 22 −Cm
¢
the outward normal of the shared plane between the constituents. v˜ ,2 = −(1 − h f )C22
∗−1
22 − C ∗ v,2
The velocity gradients vc3 can be calculated based on the overall 1 − h f 22
´
velocity gradients m2
+ C˜ 23 − C23
¡ ¢
¡¡ ¢ ¡ ¢ v,3
va,3 = −c b ∗−1
C33 Ca31 − Cb31 v,1 + Ca32 − Cb32 v,2
1 ∗
³¡ ³ ´
2
m2 2
C˜ 21 − C21 v,1 + C˜ 22 − Cm
¢
³ 1 ∗
´ ´ v,m2 = h f C22
∗−1
22 + C v,2
+ Ca33 − Cb33 − C v,3 h f 22
cb 33 ´
m2
+ C˜ 23 − C23
¡ ¢
v,3 (41)
¡¡ ¢ ¡ ¢
vb,3 = ca C33
∗−1
Ca31 − Cb31 v,1 + Ca32 − Cb32 v,2
1
³ ´ ´
+ Ca33 − Cb33 + a C∗33 v,3 (35) Combining the latter expression for the velocity gradients with
c Eq. (38), the velocity gradients for the constituents can be ob-
tained.
5. Constitutive models
v,f1 = v,1
To derive the constitutive properties of the geometries shown in
1
³¡ ³ ´
m2 m2
v,f2 = −(1 − h f )C∗−1 C˜ 21 − C21 v,1 + C˜ 22 − C22
¢
Fig. 1, a combination of the constitutive formulations Eqs. (29) and − C∗ v,2
(34) is to be used. Initially, the focus is on the first geometry
22
1 − h f 22
´
shown in Fig. 1(a). + C˜ 23 − Cm
¡ 2 ¢
23 v,3

5.1. Constitutive model 1 ³¡

1 1
v,f3 = −(1 − w f ) C∗−1 f f
m
¢ ¡ ¢
33 C31 − C31 v,1 − (1 − h f ) C32 − Cm ∗−1
32 C22
An initial homogenization is performed between f and as m1
1
³¡ ³ ´
m2 2
C˜ 21 − C21 v,1 + C˜ 22 − Cm
¢
shown in Fig. 3(a), where the symbol (•˜ ) is used. The materials 22 − C∗ v,2
1 − h f 22
70 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 155

1 used
³ ´ ´
2 f 1
+ C˜ 23 − C23
¡ m
¢ ¢
v,3 + C33 − Cm − C∗ v,3 (42)
33
1 − w f 33 1 1 1
C2i j = w f Cˆ i j +(1 − w f )Cimj −w f (1 − w f ) Cˆ i3 −Cm Cˆ ∗−1 Cˆ 3 j −Cm
¡ ¢ ¡ ¢
i3 33 3j
i, j = {1, 2, 3}
1
v,m1 = v,1 (48)

1
³¡
2
³ 1 2
´ where Cˆ ∗33 is given as
C˜ 21 −C21 v,1 + C˜ 22 −Cm
¢
v,m2 = −(1 − h f )C22
∗−1 m
22 − C∗ v,2
1 − h f 22 Cˆ ∗33 = w f Cm
1
f ˆ
´ 33 + (1 − w ) C33 (49)
2
+ C˜ 23 − Cm
¡ ¢
23 v,3 Based on the overall velocity gradients v, j , the velocity gradients
³¡ for the constituents can be obtained
1 f 1 ¢f
¡ 1 ¢
v,m3 = w f C33
∗−1
C31 m
− C31 v,1 − (1 − h f ) C32 − Cm ∗−1
32 C22 vˆ ,1 = vm
1
,1 = v,1
1
1 vˆ ,2 = vm = v,2
³¡ ³ ´
m2 m2
C˜ 21 − C21 v,1 + C˜ 22 − C22
¢ ,2
− C∗ v,2
1 − h f 22 ³¡
m1 1
vˆ ,3 = −(1 − w f ) Cˆ 33 Cˆ 31 − C31 v,1 + Cˆ 32 − Cm
∗−1
¢ ¡ ¢
1 32 v,2
³ ´ ´
2 f 1
+ C˜ 23 − C23
¡ m
¢ ¢
v,3 + C33 − Cm − C ∗ v,3 (43)
33
1 − w f 33 ³
1 1
´ ´
+ Cˆ 33 − Cm
33 − Cˆ ∗ v,3
1 − w f 33
1 1¢
f ˆ m1
Cˆ 31 − Cm ˆ
¡¡ ¡ ¢
2 vm
,3 = w C33
∗−1
31 v,1 + C32 − C32 v,2
v,m1 = v,1
1 ∗
³ ´ ´
1
1 ∗ + Cˆ 33 − Cm Cˆ v,3 (50)
³¡ ³ ´
33 +
2
m2 2
C˜ 21 − C21 v,1 + C˜ 22 − Cm
¢
v,m2 = h f C∗−1 22 + C v,2 w f 33
22
h f 22
m2 Combining the latter expression with Eq. (47), the velocity gradi-
+ C˜ 23 −
¡ ¢ ¢
C23 v,3
ents for the constituents can be determined.
2
v,m3 = v,3 (44) v,f1 = v,1
³¡
2
v,f2 = −(1 − h f ) C∗−1 f
¢
22 C21 − Cm
21 v,1
5.2. Constitutive model 2
1
³ ´
2
f m2 ˆ ∗−1
m
¡ f ¢
+ C22 − C22 − C∗ v,2 − (1 − w f ) C23 − C23 C33
The second constitutive model shown in Fig. 1(b) is derived in a 1 − h f 22
1
³¡ ³ ´ ´´
similar manner, where the first plane for performing homogeniza- m1
ˆC31 −C31 ˆ m1 ˆ m1 ˆ
¢ ¡ ¢ ∗
v,1 + C32 −C32 v,2 + C33 −C33 − C v,3
tion is shifted. The initial homogenization is done between f and 1−w f 33
m2 as shown in Fig. 3(b), where the symbol (•ˆ ) is used. The mate-
³¡
1
m1
v,f3 = −(1 − w f ) Cˆ ∗−1 Cˆ 31 − Cm ˆ
¢ ¡ ¢
rials share a common plane with outward normal parallel with the 33 31 v,1 + C32 − C32 v,2
x2 -axis and for this reason Eq. (29) is used
1
³ ´ ´
1
+ Cˆ 33 − Cm
33 − Cˆ ∗ v,3 (51)
1 − w f 33
Cˆ i j =h f Cifj + (1 − h f ) Cm − h f (1 − h f ) Cif2 − Cm C2f j − Cm
2 2 2
¡ ¢ ∗−1
¡ ¢
ij i2
C22 2j
i, j = {1, 2, 3} vm
1
,1 = v,1
(45) 1
vm
,2 = v,2
1 1
where C∗22 is given by f ˆ ∗−1 m1
Cˆ 31 − Cm ˆ
¡¡ ¢ ¡ ¢
vm
,3 = w C33 31 v,1 + C32 − C32 v,2
1 ∗
³ ´ ´
1
+ Cˆ 33 − Cm Cˆ
2 f
C∗22 = h f Cm f
22 + (1 − h ) C22 (46) 33 + v,3 (52)
w f 33
Based on the homogenized velocity gradients vˆ , j for the initial ho- 2
mogenization, the velocity gradients for the constituents can be vm
,1 = v,1
obtained as 2
³¡
f 2 ¢
vm f ∗−1
,2 = h C22 C21 − Cm
21 v,1
2
v,f1 = vm ˆ ,1
,1 = v 1 ∗
³ ´
f 2 ¡ f 2¢
2 + C22 m
− C22 + C v,2 − (1 − w f ) C23 − Cm ˆ ∗−1
23 C33
v,f3 = vm ˆ ,3
,3 = v h f 22
1
ˆC31 − Cm ˆ m1
¡¡ ¢ ¡ ¢
1 31 v,1 + C32 − C32 v,2
³¡ ³ ´
2 2
v,f2 = −(1−h f )C22 f f
¢
∗−1
C21 −Cm ˆ ,1 + C22
21 v − Cm
22 − C∗ vˆ ,2
1 − h f 22 ³
1 1
´ ´´
+ Cˆ 33 − Cm
33 − Cˆ ∗ v,3
1 − w f 33
´
f m2
¡ ¢
+ C23 − C23 vˆ ,3 ³¡
2
f ˆ ∗−1 m1 1
Cˆ 31 − C31 v,1 + Cˆ 32 − Cm
¢ ¡ ¢
³¡ ³ 1
´ vm
,3 = − (1 − w ) C33 32 v,2
2 f 2 ¢ f 2
v,m2 =h f
C∗−1
22 C21 − Cm
21 vˆ ,1 + C22 − Cm
22 + f C∗22 vˆ ,2
h ³
1 1
´ ´
+ Cˆ 33 − Cm
33 − Cˆ ∗ v,3 (53)
1 − w f 33
´
f
¡ 2 ¢
m
+ C23 − C23 vˆ ,3 (47)
5.3. Average constitutive model
The last homogenization for the model is done between (•ˆ ) and
m1 . The current constituents share a common plane with an out- By examination of the entries in the matrices given by
ward normal parallel with the x3 -axis and therefore Eq. (34) is Eqs. (39) and (48) it was observed that the models did not result
 6.5 Article P1 71

156 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

either a stress- or velocity gradient increment and the rest of the

unknown quantities can be calculated based on equilibrium and
displacement continuities across the localized band. The velocity
gradients voi, j and vii, j need to be fully determined for every incre-
ment, this is a total of 18 velocity gradients, therefore 18 equations
are to be determined.
In Jensen and Christoffersen (1997) two different formulations
for the simulation were introduced. One where the coordinate sys-
tems followed the rotation of the fibres during an increment, and a
second where the coordinate system inside the kink band was sta-
tionary during the increment. The two formulations were shown to
give identical results. In the current simulation, the first approach
is adapted due to its convenience when implemented in a numer-
ical scheme. When the coordinate systems rotate, with the fibres,
Fig. 6. Geometric representation of kink band formation.
this entails

vo2,1 = vi2,1 = 0 (57)

in transverse isotropic moduli. An average of the two models was
observed to give the wanted transverse isotropic properties the velocity gradient v2, 1 represents the rotation of the fibres and
1¡ 1 since the coordinate systems rotates with the fibres no relative ro-
C + C2i j
¢
Ci j = (54) tation is seen. In a coordinate system that initially coincides with
2 ij
the band coordinate system but stays stationary during an incre-
The constitutive expression shown above is the key expression pro-
ment, seen from the outer coordinate system, has the velocity gra-
posed for three-dimensional simulations. In addition, the veloc-
dients vii, j + Äi j . Here, Äij is the relative spin tensor between the
ity gradients for the constituents should be the average quantities
band and base material, where
based on Eqs. (42)–(44) and (51)–(53)
1 ¡ c (1 ) Ä21 = −Ä12 = φ˙ (58)
+ vc, j(2)
¢
vc, j = v (55)
2 ,j and all other components of Äij are zero. The quantity φ˙ is the
If the length scale wf is set to unity in the first, second and average relative spin increment. In the simulation, φ˙ will be a prescribed
constitutive equation, given by Eqs. (39), (48) and (54), then the quantity and based on this increment, the velocity gradients and
constitutive equations simplifies to stresses are to be determined. To ensure continuity across the
Cifj
2
Cif2
2
C2f j
2 band, the second-order tensors (tractions and velocity gradients)
f m
¡ ¢ ¡ ¢
Ci j = c +c Cimj f
−c c m
− Cim2 ∗−1
C22 − C2mj
(56) need to be rotated into a common coordinate system. The tensors
i, j = {1, 2, 3}
Roi j and Rii j are introduced which gives the directional cosines into
where the equality given by Eq. (2) is used. The matrix is given C∗22 a common coordinate system where their first axis will coincide
by Eq. (46). The latter constitutive equation has similarities with with the outward normal n. The components of the rotation ten-
the formulation derived by Christoffersen and Jensen (1996), where sor Roi j can be illustrated in a matrix format
in their formulation the free index i, j varies between the integers
1 and 2 and their matrices Cci j have dimensions 2 × 2 and not 3 × 3 Ro11 Ro12 Ro13 cos(β o ) sin(β o )
à ! à !
0
as in the present study. Ro21 Ro22 Ro23 = −sin(β o ) cos(β o ) 0 (59)
Ro31 Ro32 Ro33 0 0 1
6. Localization of deformation
where the orientation β o is shown in Fig. 6. The calculation of
Rii j is equivalent using the orientation β i . Displacement continuity
To validate the constitutive equation given by Eq. (54), a kink
band simulation is set-up. The analysis is created as a numeri- across the band entails six equations
cal simulation similar to the one done by Jensen and Christof- ¡ i ¢ ¡ ¢
fersen (1997) with some changes in the formulation to account for v + Äi j Ri2i Ri2 j = voi, j Ro2i Ro2 j vii, j + Äi j Ri3i Ri3 j = voi, j Ro3i Ro3 j
¡ i,i j ¢ i i ¡ i ¢
the three-dimensional boundary conditions. v + Äi j R1i R2 j = vi, j R1i R2 j vi, j + Äi j Ri1i Ri3 j = voi, j Ro1i Ro3 j
o o o

In the analysis, two composite states are introduced: A material ¡ i,i j ¢ i i ¡ i ¢

vi, j + Äi j R3i R2 j = vi, j R3i R2 j vi, j + Äi j Ri2i Ri3 j = voi, j Ro2i Ro3 j
o o o
inside the kink band and a surrounding base material as illustrated
in Fig. 6. In the figure (•)o represents quantities associated with the (60)
outer base material and (•)i is associated with the material inside
The first two equations are stretching continuities of the material
the kink band. The relative rotation of the fibres within the kink
on the kink band plane. The third and fourth equation are continu-
band is given by the angle φ . The rotated fibres are located in a
ities of the band plane rotation. The last two equation describe the
localized band with an outward normal n. The angle between the
shear strain and rotation of the material on the kink band plane.
first axis xo1 of the base material and the outward normal n is given
Continuity of tractions entails three equations
by β o and the angle β i is the angle from the first axis of the inner ¡ i ¢
material xi1 to the normal. Cii jkl + Älk Ri1i Ri1 j = Ciojkl vol,k Ro1i Ro1 j
¡vl,k ¢ i i
To compare the analysis with the two-dimensional simulation Cii jkl v i
+ Ä o o o o
lk R1i R2 j = Ci jkl vl,k R1i R2 j (61)
done by Jensen and Christoffersen (1997) the third axis xo3 and xi3 ¡ l,k ¢
Cii jkl vil,k + Älk Ri1i Ri3 j = Ciojkl vol,k Ro1i Ro3 j
coincide initially and during the simulation. This is done for the
purpose of illustration, and is not a restriction in general. The out- Where the first equation is the force normal to the band plane and
ward normal n lies in the xo1 − xo2 plane which entails that it also the last two equations are shear stresses. In the simulations done
lies in the xi1 − xi2 plane. The angle φ is formed during the simu- by Jensen and Christoffersen (1997), different boundary conditions
lation either by bifurcation (for φ0 = 0) or as a consequence of an were investigated including plane stress, plane strain, no σ˙ 22 incre-
initial imperfection φ 0 . The simulation is performed by prescribing ments and no vo2,2 velocity gradient. The simulations are performed
72 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 157

o = 0,
with plane strain condition and unless stated otherwise, σ22 Only small differences were seen in the post-buckling regime be-
which entails the boundary condition tween the formulations. The Stören and Rice (1975) version of J2 -
o deformation theory is presently adapted for all constituents. The
σ˙ 22 =0 (62)
formulation is outlined here for completeness. The components of
Some simulations are performed with biaxial loading e.g. com- the tensor of instantaneous moduli are given by
bined σ 11 and σ 22 stresses. In those simulations the biaxial load-
¢ ³ 2 4
¡ ´
ing ratio, ρ , is introduced Li jkl = Gs δik δ jl + δil δ jk + K − Gs δi j δkl − (Gs − Gt )mi j mkl
3 3
o
σ˙ 22 (73)
ρ= (63)
σ˙ 11o
mij is given in terms of Cauchy stresses
In the simulations with biaxial loading the boundary condition
1 1
³ ´
o
σ˙ 22 o
− ρ σ˙ 11 =0 (64) mi j = σi j − δi j σkk (74)
2 σeq 3
is used. Furthermore, simulations with combined compression and where σ eq is the equivalent von Mises stress
shear stresses in the base material are performed, where the load- r
ing ratio, ψ , is introduced 3 1
σeq = σi j σi j − σii σ j j (75)
σ o
˙ 12 2 2
ψ= o
(65)
σ ˙ 11 The remaining terms introduced in Eq. (73) is the bulk modulus, K,
the secant shear modulus, Gs , and the tangent shear modulus, Gt .
together with the boundary condition
The two latter moduli can be calculated based on the secant and
o o tangent moduli Es and Et
σ˙ 12 − ψ σ˙ 11 =0 (66)
The rate of Cauchy stress for the composite can be calculated based 1 3 1 − 2ν 1 1 1 − 2ν E
on the nominal stress rates and the velocity gradients using
= − = − K=
Gs Es E Gt Et E 3 (1 − 2 ν )
σ˙ i j = Ci jkl vl,k − σi j vk,k + vi,k σk j (67) (76)

Furthermore, five boundary conditions are applied to the base ma- where ν is Poisson’s ratio and E is Young’s modulus. The nonlinear
terial to obtain plan strain conditions relationship between the equivalent uniaxial logarithmic strain ε eq
and equivalent uniaxial stress σ eq is given by a Ramberg–Osgood
vo3,1 = 0 , vo3,2 = 0 , vo1,3 = 0 relation
(68)
vo2,3 = 0 , vo3,3 = 0 µ ¶n
σeq 3 σy σeq
Using the continuity- and boundary conditions presented, the εeq = + (77)
E 7E σy
velocity gradients and stresses can be calculated for the base and
band material numerically. The relative rotation of the kink band where σ y is a reference yield stress and n is a hardening exponent.
is updated according to Both the secant modulus Es and tangent modulus Et can be deter-
mined based on the Ramberg–Osgood curve using the relations
β˙ o = −voi, j Ro1i Ro2 j β˙ i = −vii, j Ri1i Ri2 j (69)
1 εeq 1 ∂εeq
In line with the formulation done by Jensen and Christof- = = (78)
Es σeq Et ∂σeq
fersen (1997), the fibre volume fraction cf inside and outside the
kink band is different. The number of fibres penetrating a unit area
8. Numerical scheme
A = A f /c f of an arbitrary section with the unit normal ni is

cf The steps required in the numerical scheme are presented in

N = ni ei f (70) this section. The relative spin increment φ˙ is prescribed. The simu-
A
lation goes from an initial fibre imperfection φ 0 to an end rotation
where ei is a unit vector parallel with the fibres, Af is the area of φ decided by the user. The following steps are required in every
a single fibre. In any region not containing fibre ends the latter increment:
equation entails
1. Update instantaneous moduli for the constituents Lci jkl using
c fo cfi
noi eoi = nii eii f i (71) Eq. (73).
Afo A 2. Calculate nominal moduli for the constituents Cicjkl using
Based on the latter equation the volume fraction inside the band Eq. (13).
can be determined based on the volume fraction in the base ma- 3. Form the matrices Cci j for the constituents based on Cicjkl so they
terial. The length scales wf and hf are updated according to fulfil the form tci = Cci j vc, j given by Eq. (16).
h˙ f = h f (1 − h f ) (v2f ,2 − vm 4. Calculate the nominal moduli for the composite Cij based on
2
2,2 )
(72) Eq. (54) using Eqs. (39) and (48).
w˙ f = w f (1 − w f ) (v3f ,3 − v3m,3 )
1

5. Set-up a linear system of equations based on the 18 equations

Jensen and Christoffersen (1997) used a similar formulation with given by Eq. (57), (60), (61), (64), (66) and (68). A linear system
the main difference that two length scales wf and hf are needed in in the form [A] V = R can be set up where V is a vector con-
the current formulation. taining the unknown velocity gradients for the band and base
material vii, j , voi, j . The linear system is solved with respect to
7. Elastic-plastic behaviour of the constituents the unknown vector V.
6. Calculate the velocity gradients for the band and base con-
In Jensen (1999) three constitutive formulations for the incre- stituents vci, j based on the overall velocity gradients vi, j using
mental stiffness tensor Lci jkl for the constituents were compared. Eqs. (42)–(44), (51)–(53) and (55).
 6.5 Article P1 73

158 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

o
Fig. 7. Applied stress −σ11 in base material normalised with the elastic shear modulus G, as a function of fibre rotation φ , shown for four different homogenized formulations.
Initial imperfection φ 0 ≈ 0° and initial orientation β0 = 10◦ .

Table 1 and will be referred to as the 2D model. The results obtained us-
Material parameters.
ing the Simple 3D model and the 2D model show results that coin-
νm 0.356 cides. The reason for the coinciding results can be explained by the
σym /E m 0.013 simplified constitutive equation given by Eq. (56) and the similari-
nm 4
ties to the constitutive equation developed by Christoffersen and
Ef /E m 35
νf 0.263 Jensen (1996). The main difference between the two mentioned
σyf /E f 0.019 constitutive equation is the size of the matrices, which is 3 × 3
nf 1.9 in the present study and not 2 × 2 as in the equation derived by
cf 0.6 Christoffersen and Jensen.
The results obtained using the three-dimensional model given
by Eq. (54) will be denoted the 3D model which is illus-
7. Update the stresses σi j = σi j + σ˙ i j where the stress increments trated as a dashed line. The 3D model shows the same char-
σ˙ i j is calculated using Eq. (67). acteristics during deformation based on the results shown in
8. Update the geometric quantities β , φ , hf and wf using Figs. 7 and 8. The 3D model gives slightly higher bifurca-
Eqs. (69) and (72). tion stress and higher stresses during deformation. The higher
stresses can be explained by the slightly stiffer shear response
9. Results observed in the current three-dimensional model. Several au-
thors including Budiansky (1983), Budiansky and Fleck (1993) and
The material parameters used in the present study and also Christoffersen and Jensen (1996) have shown that kink band for-
used in Jensen (1999), are taken based on experimental results on mation is dominated by the shear response of the composite inter-
APC-2/AS4 carbon fibre composite from Kyriakides et al. (1995). acting with fibre misalignments. Skovsgaard and Jensen (2017) in-
The parameters are summarized in Table 1. Aligned with stud- vestigated the kink band response for a three-dimensional dis-
ies done by previous authors, then the composite behaviour cretized finite element model and a constitutive model. In that
is described through a rate independent J2 -deformation theory, study a comparison between the shear response was conducted
even though the matrix constituent is a polymer. Authors who initially. The matrix material was softened in the finite element
have investigated the influence of the rate and pressure de- model so the shear response between the models was similar. The
pendency in fibre-reinforced polymers include: Vogler and Kyri- material characteristics of the matrix material (AS4-PEEK) change
akides (1999a) experimentally investigated the nonlinear behaviour during the curing process, due to this fact the matrix characteris-
of AS4/PEEK under combined transverse compression and shear tics are in several cases predicted based on the composite response
and they compared it with simulations with different constitutive (Hsu et al., 1998; Ng et al., 2010). Since the shear response of the
models in Vogler and Kyriakides (1999b). 3D model and 2D model differs, a new calibration of the material
The results presented are used to investigate the performance parameters of the matrix material is proposed. A fourth model is
of the present three-dimensional constitutive model. illustrated in the figures and will be referred to as the 3D model
In Figs. 7 and 8, results are shown for simulations with an ini- with matrix correction. This is a model where the elastic modulus
tial imperfection of φ 0 ≈ 0° and φ0 = 3◦ , respectively. For both sim- of the matrix material Em is changed. The modulus Em is calcu-
ulations, the initial kink band angle is set to β0 = 10◦ . In the fig- lated so the elastic shear response of the composite is identical to
ures, four different results are presented. The first result, which the elastic shear response of the 2D model.
is presented as a solid line, is a simplified version of the three- The 3D model with matrix correction is illustrated as a dashed-
dimensional constitutive model, where w f = 1. In the case where dotted line. The simulations with an initial imperfection of φ0 =
the length scale w f = 1, the three-dimensional formulation con- 3◦ show that the present model, where the shear response is
verses towards Eq. (56). This model will be denoted the Simple 3D aligned, yield results closer to the response obtained by the two-
model. The results obtained by Jensen (1999) is presented as circles
74 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 159

o
Fig. 8. Applied stress −σ11 in base material normalised with the elastic shear modulus G, as a function of fibre rotation φ , shown for four different homogenized formula-
tions. Initial imperfection φ0 = 3◦ and initial orientation β0 = 10◦ .

c
Fig. 9. Critical kink band stress −σ11 in base material normalised with the elastic shear modulus G, versus initial imperfection φ 0 , shown for four different homogenized
formulations. The critical stress is presented for the most critical initial kink band orientation β 0 .

dimensional model. Hsu et al. (1998) made a comparison between imental data. In the current results, a simulation was created with
a two- and three-dimensional micro-mechanical kink band model the enhanced elastic response of the matrix material. Since kink
where they simulated the buckling response using a finite ele- band formation is highly sensitive to the plastic shear response
ment model in ABAQUS. They arrived at the same conclusion that (Budiansky, 1983), the elastic and plastic parameters should be cal-
the three-dimensional model showed similar response as the two- ibrated to match the composite shear response for strains relevant
dimensional model. Equivalently, they also used different elastic- for the simulation. In the current simulation, only the elastic pa-
plastic parameters for the two-dimensional and tree-dimensional rameter Em was changed in the 3D model with matrix correction.
model. Furthermore, different simulation cases have been conducted
In Fig. 9, the critical stress is shown as function of the initial with biaxial loading (combined compression and tension and com-
fibre imperfection φ 0 . In line with the simulations performed by bined compression and compression) and cases of combined com-
Jensen and Christoffersen (1997), the critical stress is illustrated pression and shear with different sign of shear. In the simulations
for the most critical initial kink band orientation β 0 . The Simple to come both the 2D model and the 3D model with matrix correction
3D model, the 3D model with matrix correction and the 2D model have been used. In Figs. 10 and 11 the normalized critical stress
show similar results. The 3D model, with the same material param- −σ11c /G is shown as function of initial fibre imperfection, φ , for
0
eters as used for the two-dimensional model, shows higher critical different biaxial loading ratios. Fig. 10 gives the results from the
stresses for all initial imperfections. As mentioned earlier, the in- 2D model and Fig. 10 are results from the 3D model with matrix
situ material parameters of the matrix material are in several cases correction.
fitted so the overall shear response of the composite match exper-
 6.5 Article P1 75

160 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

c
Fig. 10. Normalized critical kink band stress −σ11 /G versus initial imperfection φ 0 for different ratios of biaxial loading. The critical stress is presented for the most critical
initial kink band orientation β 0 . The results are generated based on the 2D-model.

c
Fig. 11. Normalized critical kink band stress −σ11 /G versus initial imperfection φ 0 for different ratios of biaxial loading. The critical stress is presented for the most critical
initial kink band orientation β 0 . The results are generated based on the 3D-model with matrix correction.

Positive values of the biaxial loading ratio, σ22 o /σ o > 0, rep- the results obtained from the 3D model with matrix correction. The
11
resents a case of combined compression and compression. With boundary condition σ˙ 22 o = 0 is used in the simulations including

the current material parameters, then the loading ratio σ22 o /σ o = shear stresses in the base material.
11
0.1 leads to an increase in the critical stress for all imperfec- For fibre imperfections φ 0 > 1° a clear tendency is seen be-
tion angles compared to the case of σ22 = 0. The loading ratio tween the critical stresses and the compression-shear ratio. Posi-
o /σ o = 0.2 leads to a reduction in the critical stress for imper-
σ22 o /σ o > 0 lead to an increase in the critical compres-
tive ratios σ12
11 11
fection angles φ 0 < 1°, which is due to matrix yielding due to the o < 0. The result that neg-
sive stress, e.g. negative shear stresses σ12
increased stress level. In all cases of combined compression and ative shear stresses increase the compressive strength is in line
tension, e.g. σ 11 < 0 and σ 22 > 0, a decrease in the critical stress with the expression derived in Christoffersen and Jensen (1996).
is obtained. The current findings are in line with the analytical For small fibre imperfections a cross-over is seen between the re-
expression for kink band bifurcation found by Christoffersen and sults with positive ratios σ12o /σ o > 0 e.g. the bifurcation stress for
11
Jensen (1996) and Slaughter et al. (1993). By comparing the results o /σ o = 0.02 is smaller than in the case of pure compression. The
σ12 11
obtained using the two-dimensional and three-dimensional consti- yield surface is based on a von Mises yield criterion and the equiv-
tutive model then good correlation is seen. alent von Mises stress is higher in the case combined compression
The critical stress as function of initial fibre imperfection in the and shear. It has been observed that very small positive ratios (e.g.
case of combined compression and shear is shown in Figs. 12 and o /σ o = 0.001) will lead to an infinitesimal increase in bifurca-
σ12 11
13. Fig. 11 shows the results from the 2D model and Fig. 13 gives tion stress, but this is not shown in the figure.
76 Chapter 6 - P1

S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162 161

c
Fig. 12. Normalized critical kink band stress −σ11 /G versus initial imperfection φ 0 for different ratios of combined compression and shear. The critical stress is presented
for the most critical initial kink band orientation β 0 . The results are generated based on the 2D-model.

c
Fig. 13. Normalized critical kink band stress −σ11 /G versus initial imperfection φ 0 for different ratios of combined compression and shear. The critical stress is presented
for the most critical initial kink band orientation β 0 . The results are generated based on the 3D-model with matrix correction.

By comparing Figs. 10 and 11 for biaxial loading and by com- constitutive model is built up using three material blocks: one fi-
paring Figs. 12 and 13 for combined compression and shear, these bre and two matrix.
results further validates the present three-dimensional constitutive The current constitutive model can be used to predict strain
model. localization of fibre composites including the localization phe-
nomenon known as kink band formation. The constitutive model is
used in a simple kink band simulation where the response is com-
10. Conclusion pared with results obtained by previous authors. Similar results
for plane strain are observed in the post-buckling regime when
A three-dimensional constitutive model has been proposed comparing the constitutive models. The critical stress is seen to
which give the material response of a fibre-reinforced composite be sensitive to fibre imperfections and to the shear response of
in a general framework. The model is based on independent con- the composite which is consistent with previously published re-
stitutive relations for the constituents where the constituents can sults. The same elastic-plastic parameters are used for the present
behave either elastic or elastic-plastic. The model is formulated in three-dimensional model and higher stresses are observed in gen-
a rate form using nominal stress rates and velocity gradients which eral. The higher stresses observed, can be explained by the stiffer
gives the model opportunity to be used in the finite strain regime shear response in the present model. The three-dimensional ef-
with large deformations and rotations. The present paper empha- fects from the constitutive model due not alter the critical values
sises the derivations behind the model, where homogenizations predicted by previous two-dimensional models when the shear re-
are based on continuum mechanical continuities and averages. The sponses for the two models are aligned with the present bound-
 6.5 Article P1 77

162 S.P.H. Skovsgaard, H.M. Jensen / International Journal of Solids and Structures 139–140 (2018) 150–162

ary conditions. This is consistent with conclusions done by other Fleck, N.A., Shu, J.Y., 1995. Microbuckle initiation in fibre composites : afi-
authors (Hsu et al., 1998), who have compared three-dimensional nite element study. J. Mech. Phys. Solids 43 (12), 1887–1918. doi:10.1016/
0 022-5096(95)0 0 057-P.
and two-dimensional models. To fit the full elastic-plastic shear re- Grandidier, J.C., Ferron, G., Potier-Ferry, M., 1992. Microbuckling and strength in
sponse of the proposed constitutive model to experimental data long-fiber composites: theory and experiments. Int. J. Solids Struct. 29 (14-15),
has not been in the scope of the current research. Instead, focus is 1753–1761. doi:10.1016/0020-7683(92)90168-S.
Gutkin, R., Costa, S., Olsson, R., 2016. A physically based model for kink-band growth
given to the overall tendencies and performance of a new three- and longitudinal crushing of composites under 3D stress states accounting for
dimensional constitutive model. friction. Compos. Sci. Technol. doi:10.1016/j.compscitech.2016.09.002.
The simulations performed verifies the derived three- Gutkin, R., Pinho, S.T., 2015. Combining damage and friction to model compressive
damage growth in fibre reinforced composites. J. Compos. Mater. 49 (20), 2183–
dimensional constitutive model. The current infinite kink band
2495. doi:10.1177/0021998314549614.
simulation only includes plane strain conditions and in-plane Guynn, E.G., Ochoa, O.O., Bradley, W.L., 1992. A parametric study of variables that
rotations. Future simulations can include three-dimensional effects affect fiber microbuckling initiation in composite laminates: part 1-analyses. J.
Compos. Mater. 26 (11), 1594–1616. doi:10.1177/002199839202601103.
e.g. deformation out-of-plane and inclined kink bands through
Hsu, S.-Y., Vogler, T., Kyriakides, S., 1999. On the axial propagation of kink bands in
thickness in a similar set-up. fiber composites : part I experiments. Int. J. Solids Struct. 36, 575–595. doi:10.
1016/S0 020-7683(98)0 0 030-4.
Acknowledgement Hsu, S.-Y., Vogler, T., Kyriakides, S., 1999. On the axial propagation of kink bands in
fiber composites : part II analysis. Int. J. Solids Struct. 36, 575–595. doi:10.1016/
S0 020-7683(98)0 0 030-4.
This work was supported by the Thomas B. Thriges fund. Hsu, S.-Y., Vogler, T.J., Kyriakides, S., 1998. Compressive strength predictions for fiber
composites. J. Appl. Mech. 65 (1), 7. doi:10.1115/1.2789050.
Jensen, H.M., 1999. Analysis of compressive failure of layered materials by
Appendix A. Velocity gradients vc,2
kink band broadening. Int. J. Solids Struct. 36 (23), 3427–3441. doi:10.1016/
S0 020-7683(98)0 0158-9.
Noting Eqs. (16) and (17) together with Eq. (19) yields Jensen, H.M., 2002. Residual stress effects on the compressive strength of
uni-directional fibre composites. Acta Mater. 50, 2895–2904. doi:10.1016/
Ca21 v,1 + Ca22 va,2 + Ca23 v,3 = Cb21 v,1 + Cb22 vb,2 + Cb23 v,3 (A.1) S1359- 6454(02)00114- 3.
Jensen, H.M., Christoffersen, J., 1997. Kink band formation in fiber reinforced ma-
Isolating the term related vb,2 terials. J. Mech. Phys. Solids 45 (7), 1121–1136. doi:10.1016/S0022-5096(96)
00126-3.
Kyriakides, S., Arseculeratne, R., Perry, E.J., Liechti, K.M., 1995. On the compressive
¡ ¢ ¡ ¢
Cb22 vb,2 = Ca21 − Cb21 v,1 + Ca22 va,2 + Ca23 − Cb23 v,3 (A.2)
failure of fiber reinforced composites. Int. J. Solids Struct. 32 (6-7), 689–738.
Using the equality given by Eq. (18) and isolating the term related doi:10.1016/0 020-7683(94)0 0157-R.
Liu, G., Thouless, M.D., Deshpande, V.S., Fleck, N.A., 2014. Collapse of a composite
to vb,2 yields beam made from ultra high molecular-weight polyethylene fibres. J. Mech. Phys.
Solids 63 (1), 320–335. doi:10.1016/j.jmps.2013.08.021.
cb vb,2 = v,2 − ca va,2 (A.3) Naya, F., Herraez, M., Gonzalez, C., Lopes, C., der Veen, S.V., Pons, F., 2017. Com-
putational micromechanics of fiber kinking in unidirectional FRP under differ-
Multiplying the latter equation with Cb22 yields ent environmental conditions. Compos. Sci. Technol. 144, 26–35. doi:10.1016/j.
compscitech.2017.03.014.
cb Cb22 vb,2 = Cb22 v,2 − ca Cb22 va,2 (A.4) Ng, W.H., Salvi, A.G., Waas, A.M., 2010. Characterization of the in-situ non-linear
shear response of laminated fiber-reinforced composites. Compos. Sci. Technol.
Comparing the equation with Eq. (A.2) yields 70 (7), 1126–1134. doi:10.1016/j.compscitech.2010.02.024.
¡ ¢ ¡ ¢ Nizolek, T., Begley, M., McCabe, R., Avallone, J., Mara, N., Beyerlein, I., Pollock, T.,
cb Ca21 − Cb21 v,1 + cb Ca22 va,2 + cb Ca23 − Cb23 v,3 2017. Strain fields induced by kink band propagation in Cu-Nb nanolaminate
composites. Acta Mater. 133, 303–315. doi:10.1016/j.actamat.2017.04.050.
= Cb22 v,2 − c a
Cb22 va,2 (A.5) Poulios, K., Niordson, C.F., 2016. Homogenization of long fiber reinforced composites
including fiber bending effects. J. Mech. Phys. Solids 94, 433–452. doi:10.1016/j.
Isolating the term va,2 yields jmps.2016.04.010.
¡ ¢ Romanowicz, M., 2013. Initiation of kink bands from regions of higher misalignment
ca Cb22 + cb Ca22 va,2 in carbon fiber-reinforced polymers. J. Compos. Mater. 0 (0), 1–13. doi:10.1177/
¡ ¢ ¡ ¢ 0021998313498106.
b
= −c Ca21 − Cb21 v,1 + Cb22 v,2 − cb Ca23 − Cb23 v,3 (A.6) Rosen, B.W., 1965. Mechanics of Composite Strengthening in: Fiber Composite Ma-
terials. American Society for Metals, Metals Park, OH, pp. 37–75.
Introducing the matrix C∗22 Skovsgaard, S.P.H., Jensen, H.M., 2017. Constitutive model for imperfectly bonded fi-
bre-reinforced composites. Compos. Struct. 192, 82–92.
C∗22 = ca Cb22 + cb Ca22 (A.7) Slaughter, W.S., Fleck, N.A., Budiansky, B., 1993. Compressive failure of fibre com-
posites: the roles of multi-axial loading and creep. J. Eng. Mater. Technol. 115
The velocity gradients va,2 can be calculated as (3), 308–313.
Stören, S., Rice, J.R., 1975. Localized necking in thin sheets. J. Mech. Phys. Solids 23
(6), 421–441. doi:10.1016/0 022-5096(75)90 0 04-6.
1 b
³¡ ¢ ¡ ¢ ´
va,2 = −cb C22
∗−1
Ca21 − Cb21 v,1 − C v,2 + Ca23 − Cb23 v,3 Vogler, T., Hsu, S.-Y., Kyriakides, S., 2001. On the initiation and growth of kink bands
cb 22 in fiber composites. Part II: analysis. Int. J. Solids Struct. 38 (15), 2653–2682.
doi:10.1016/S0 020-7683(0 0)0 0175-X.
(A.8) Vogler, T.J., Hsu, S.Y., Kyriakides, S., 2001. On the initiation and growth of kink bands
in fiber composites. Part I: experiments. Int. J. Solids Struct. 38 (15), 2653–2682.
In a similar manner the velocity gradients vb,2 can be determined doi:10.1016/S0 020-7683(0 0)0 0175-X.
as Vogler, T.J., Kyriakides, S., 1999. Inelastic behavior of an AS4 / PEEK composite un-
³¡ ¢ 1 a ¡ ¢ ´ der combined transverse compression and shear . Part I : experiments 15 (8),
vb,2 = ca C∗−1 Ca21 − Cb21 v,1 + C v,2 + Ca23 − Cb23 v,3 (A.9) 807–836.
22
ca 22 Vogler, T.J., Kyriakides, S., 1999. Inelastic behavior of an AS4 / PEEK composite under
combined transverse compression and shear . Part II : modeling 15.
It is convenient to rewrite the velocity gradients to the form pre- Wadee, M.A., Hunt, G.W., Peletier, M.A., 2004. Kink band instability in layered struc-
sented in Eq. (21). tures. J. Mech. Phys. Solids 52 (5), 1071–1091. doi:10.1016/j.jmps.2003.09.026.
Wind, J.L., Steffensen, S., Jensen, H.M., 2014. Comparison of a composite model and
References an individually fiber and matrix discretized model for kink band formation. Int.
J. Non-Linear Mech. 67, 319–325. doi:10.1016/j.ijnonlinmec.2014.10.005.
Budiansky, B., 1983. Micromechanics. Comput. Struct. 16 (1-4), 3–12. doi:10.1016/ Zhang, L., Zhang, S., Jiang, Y., Tao, J., Chen, X., 2016. Compressive behaviour of fibre
0045- 7949(83)90141- 4. reinforced plastic with random fibre packing and a region of fibre waviness. J.
Budiansky, B., Fleck, N., 1993. Compressive failure of fibre composites. J. Mech. Phys. Reinf. Plast. Compos. doi:10.1177/0731684416674070.
Solids 41 (I), 183–211. doi:10.1016/0 022-5096(93)90 068-Q. Zhou, H.W., Yi, H.Y., Gui, L.L., Dai, G.M., Peng, R.D., Wang, H.W., Mishnaevsky, L.,
Christoffersen, J., Jensen, H.M., 1996. Kink band analysis accounting for the mi- 2013. Compressive damage mechanism of GFRP composites under off-axis load-
crostructure of fiber reinforced materials. Mech. Mater. 24 (4), 305–315. doi:10. ing: experimental and numerical investigations. Compos. Part B 55, 119–127.
1016/S0167-6636(96)0 0 052-X. doi:10.1016/j.compositesb.2013.06.007.
Chapter 7

P2
Constitutive model for imperfectly bonded fibre-reinforced
composites

7.1 Motivation
The behaviour and failure of fibre composites are influenced by the fibre-to-matrix interfacial
bonding. Imperfect fibre/matrix bonding will inevitably be present and can be influenced by
the fibre treatment when producing the fibres (different sizing). One way of including the
behaviour of the interface properties, when doing simulations, is by using a three-dimensional
discrete FE-model including interfacial behaviour through e.g. cohesive elements and contact
constraints as explained in Section 2.5. The use of micro-mechanical models (discrete
models) become inexpedient compared to homogenised constitutive models when advanced
fibre layups or complex geometries are considered.
The present publication introduces a novel two-dimensional constitutive formulation that
includes information about imperfect fibre-to-matrix bonding, that can be used in a general
elastic-plastic framework. The constitutive model is used to study kink band formation that
is formed by bifurcation or due to the presence of initial geometric fibre imperfections. The
original source of the publication is: (Skovsgaard and Jensen, 2018a).
80 Chapter 7 - P2

7.2 Method
The thoughts and theory behind the formulation are described in Chapter 3 and the essence is
repeated here for completeness. In the case of perfect bonding, the homogenised properties
are based on the assumptions that
• Material lines parallel with the fibres are subject to a common stretching and rotation.
• Planes parallel with the fibres transmit identical tractions.
• The material of the constituents is elastic or elastic plastic.
In the other extreme case of complete debonding, the model assumes that the transverse and
shear behaviour is dominated by the matrix properties as if the fibres were not present. The
formulation includes a parameter µ taking values between zero and one. Where a value µ = 1
is perfect bonding and µ = 0 is no bonding. The developed constitutive formulation is used
together with a semi-analytical kink band model described in Section 3.4, and was solved
using Fortran 77. Furthermore, a three-dimensional finite element model was developed
that includes different amounts of fibre-to-matrix bonding. The FE-model was developed to
investigate the influence of fibre/matrix bonding and to validate the homogenised constitutive
formulation.

7.3 Contribution
The contributions to the publication comprise: the development of a novel constitutive formu-
lation including information about fibre/matrix bonding; development of a three-dimensional
unit cell finite element model; conducting kink band simulations using the different models
and writing the paper.

7.4 Main findings

One of the main findings is the development of the constitutive formulations itself. In the
publication, the homogenised constitutive model is compared with the FE-model through
a kink band study, and good agreement between the models was observed despite the
remarkable difference in the model set-up. The constitutive model was used to investigate
the influence of both geometrical fibre imperfections and fibre-to-matrix imperfections upon
the critical compressive strength of unidirectional fibre composites and layered materials.
The formulation can furthermore be implemented as a user subroutines into Abaqus/CAE
and thereby be used to study the response and failure of complex fibre domains e.g. investigate
the influence of open-holes as done in [C1].
7.5 Article P2 81

&RPSRVLWH6WUXFWXUHV  ²

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Constitutive model for imperfectly bonded fibre-reinforced composites

Simon P.H. Skovsgaard , Henrik Myhre Jensen
⁎

Department of Engineering, Aarhus University, Inge Lehmanns Gade 10, 8000 Aarhus C, Denmark

A R T I C L E I N F O A B S T R A C T

Keywords: A constitutive model for fibre-reinforced composites has been developed that takes imperfect fibre/matrix in-
Imperfect bonding terfacial bonding into account. The model can predict the nonlinear material response of a composite in the large
Strain localisation strain regime. Independent constitutive laws can be used for the constituents behaving elastic or elastic-plastic. A
Bifurcation constitutive equation is derived for the composite moduli relating stresses to strains. The model is used to predict
Constitutive model
the development of the compressive failure; fibre kinking that is either formed by bifurcation or due to an initial
Kink band
fibre imperfection. A three-dimensional finite element model for kink band formation is used to validate the
results obtained using the constitutive model with varying levels of interfacial bonding.

1. Introduction account. Zhou et al. (2013) [4] investigated the compressive strength of
unidirectional glass fibre reinforced-polymers (GFRP) from different
Fibre-reinforced composites (FRC) are used in many applications angles and fibre kinking was the dominant failure for small angles.
where high stiffness and low weight is desirable. These materials are Nizolek et al. (2017) [5] observed kink band initiation and stable band
composed of fibres bonded together using a resin material also known broadening in a Cu-Nb nanolaminate exposed to compressive loading. It
as the matrix material. In many of the frequently used FRCs as for ex- was observed by Nair et al. (2017) [6] that a 75 % reduction in com-
ample glass- and carbon fibre-reinforced composites, the fibres have pressive strength could be achieved by introducing a severe fibre wa-
high stiffness and strength and the matrix is more ductile and has high viness into the unidirectional GFRP and failed by kink band formation.
toughness but with a lower stiffness. These fibre composites have high Several attempts have been done towards developing analytical
strength in the direction of the fibres but when loaded in compression expressions predicting the kink band initiation. In the early work of
the critical stress can be considerable lower due to instabilities caused Rosen (1965) [7] an analytical expression was derived based on the
by imperfections coupled with matrix yielding. The compressive failure compressive bifurcation load of beams surrounded by an elastic matrix.
of multidirectional laminates is composed of several competing failure Argon (1972) [8] treated the kink band failure as a plastic event, with a
mechanisms including: fibre kinking, fibre splitting, matrix cracking, composite behaving rigid perfectly plastic with an initial imperfection.
delamination, fibre/matrix interfacial debonding. In an experimental Budiansky (1983) [9] extended the expression by assuming rigid fibres
study conducted by Bishara et al. (2017) [1], together with finite ele- and elastic perfectly plastic shear response of the composite. Fleck and

with 16 layers initiated due to kink band formation in the 0° ply near an
ment analyses, they showed that the compressive failure of a laminate Budiansky (1991) [10] included shear stresses and later Slaughter et al.
(1993) [11] introduced transverse stresses in an analytical expression.
imperfection. The fibre kinking was triggered by matrix yielding. Sev- Christoffersen and Jensen (1996) [12] developed a method to find the
eral of the other mentioned failure mechanisms was seen to occur close kink band bifurcation load for a composite including fibre and matrix
to the kink band after the initiation. material nonlinearities and with multiaxial loading. In the case of rigid
The kink band failure has been observed experimentally by several fibres an analytical equation was developed where the effect of residual
authors including: Kyriakides et al. (1995) [2] who made a thorough stresses could be included. Later Jensen (1999) [13] developed an
investigation of the compressive failure of unidirectional AS4/PEEK analytical expression for the kink band bifurcation in the extreme case
composites. Kink band formation were seen both experimentally and in of no bonding between fibre and matrix. The expression developed in
their two-dimensional micromechanical finite element model (FEM). Christoffersen and Jensen (1996) [12] assumed perfect bonding.
Wadee et al. (2004) [3] conducted a compression test on a FRC and The kink band initiation can be investigated numerically either by
fibre kinking was observed and compared with a simple mechanical setting up a simplified kink band analysis as done by Jensen and
model taking bending, friction, membrane and foundation energy into Christoffersen (1997) [14] and Wadee et al. (2004) [3] or by creating a

⁎
Corresponding author.
E-mail addresses: [email protected] (S.P.H. Skovsgaard), [email protected] (H.M. Jensen).

https://doi.org/10.1016/j.compstruct.2018.02.053

$YDLODEOHRQOLQH)HEUXDU\
Received 28 September 2017; Accepted 19 February 2018

‹(OVHYLHU/WG$OOULJKWVUHVHUYHG
82 Chapter 7 - P2

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

finite element model. The FEMs used to study kink bands can roughly Jensen (1996) [12] and Jensen (1999) [13]. The latter constitutive
be divided into two categories: discretized FEMs also known as mi- models are extrema where the model derived in Christoffersen and
cromechanical models or FEMs using homogenised constitutive models. Jensen (1996) [12] assumes perfect bonding between the constituents
The fibre and matrix constituents are modelled discretely in the mi- and the model in Jensen (1999) [13] assumes complete decohesion. The
cromechanical models. Examples of authors who have used this ap- present constitutive model can display the transition between the two
proach are Hsu et al. (1999) [15] who used a three-dimensional FEM previous mentioned extrema using a factor µ going from zero to unity.
with a hexagonal fibre distribution, and Wind et al. (2014) [16] who The constitutive model is implemented in a kink band analysis and is
compared results obtained using a discretized model and a constitutive compared with a three-dimensional micromechanical finite element
model. When advanced fibre layups or complex geometries are con- model for verification.
sidered then discretized models become inexpedient compared to Simple representations of perfect and imperfect bonding are illu-
homogenized constitutive models. Fleck and Shu (1995) [17] devel- strated in Figs. 1 and 2 to enhance the understanding behind the as-
oped a constitutive model using finite strain continuum theory com- sumptions used in the constitutive models. The model with perfect
bined with Cosserat couple theory. The constitutive model gives the bonding is based on the assumptions.
smeared out properties of a composite including fibre bending effects.
The constitutive model developed by Christoffersen and Jensen (1996) 1. Material lines parallel with the fibres are subject to a common
[12] is based on independent constitutive equations for the constituents stretching and rotation.
instead of smeared out properties of the composite. Poulios and 2. Planes parallel with the fibres transmit identical tractions.
Niordson (2016) [18] developed a two-dimensional constitutive model 3. The material of the constituents is elastic or elastic–plastic.
based on independent constituent behaviour and included intrinsic size
effects using higher order strain gradients. Skovsgaard and Jensen As outlined in Christoffersen and Jensen (1996) these assumptions
(2018) [19] developed a three-dimensional constitutive model for a leads to the restriction on the velocity gradients

= v1,1 = v1,1, = v2,1 = v2,1,

FRC with independent elastic-plastic behaviour of the constituents si- m f m f
v1,1 v2,1
+ c mv1,2 = v1,2, + c mv2,2 = v2,2,
milar to the two-dimensional model developed in Christoffersen and
f m f m
Jensen (1996) [12]. c f v1,2 c f v2,2 (1)
Dève (1997) [20] observed in a compression experiment on an

c f + c m = 1. A comma (•) , denotes partial derivative. Superscripts (•)m

aluminium matrix composite reinforced by Al2O3 fibres that the inter- where cm
and cf
are volume fractions of matrix and fibre fulfilling
facial bonding between matrix and fibre has a severe influence on the
compressive strength. In the experiment the fibres were coated so the and (•) f will refer to quantities associated with the matrix and fibre
interface was ”weak” and lower critical stresses were observed. To the constituent and omission of superscript refers to overall composite
authors’ knowledge, this is the only compression experiment of kink properties. This convention will be adopted in the current article. The
second assumption together with overall equilibrium entails

̇ = t21
̇ f = t21̇ , ̇ = t22
̇ f = t22̇ ,
band formation done where composites with strong and weak bonding
are compared. The longitudinal and transverse response of an alumi- m m
t21 t22
c f t11̇f + c mt11
̇ = t11̇ , c f t12̇f + c mt12
̇ = t12̇ ,
nium composite was investigated by Zhang et al. (2008) [21] and they m m
concluded that the transverse properties depends strongly on interfacial (2)
bonding between matrix and fibre. It is commonly accepted that the where tij̇ is the nominal stress rates.
kink band instability is sensitive to the composite shear response, and In Fig. 2 shear and transverse deformation of a fibre composite with
since the shear response is altered by the interfacial bonding then so are imperfect bonding is shown. The constitutive model suggested by
the critical kink band stress. Jiang et al. (2014) [22] investigated the Jensen (1999) [13] also assumes that material lines parallel with the
composite properties using a finite element representative volume fibres are subject to a common stretching and rotation. Furthermore it
element with different levels of bonding between matrix and fibre and is assumed that the matrix as a whole is subject to the overall strains

= v1,1 = v1,1, = v2,1 = v2,1,

made the same conclusions as Zhang et al. (2008) [21]. The interface
m f m f

= v1,2 = v2,2.
was modelled with cohesive contact surfaces. A comprehensive mi- v1,1 v2,1
m m
cromechanical kink band analysis with cohesive-frictional interfaces v1,2 v2,2 (3)
were conducted by Naya et al. (2017) [23].
Further, the rate of nominal stress was given by

c f t11̇f + c mt11
̇ = t11̇ , c f t12̇f + c mt12
̇ = t12̇ ,
The interface bonding between fibre and matrix has a severe in-
fluence on the critical compressive stress, and the need to have con- m m

̇ = t21̇ , t21
̇ f = 0, c mt22
̇ = t22̇ , t22
̇ f = 0.
stitutive models to investigate complex geometries and fibre layouts m m
c mt21 (4)

The suggestion t1̇ α = c f t1̇fα + c mt1ṁ α is an average of the tractions

leads to the main focus of this paper. The current paper focuses on the
development of a novel constitutive model for imperfectly bonded

c t2̇ α = t2̇ α is an average where only the matrix constituent contributes.

fibre-reinforced composites. The constitutive model is validated using a where both the fibre and matrix transmit traction. The assumption
kink band analysis and is compared with results obtained using a three- m m
dimensional FE micromechanical kink band model. The paper is orga-
nised in 7 sections. The constitutive model and a novel analytical kink
2.1. General relations
band expression is derived in Section 2. A semi-analytical kink band
analysis with the constitutive model implemented is introduced in
The index notation and the summation convention is adopted. Latin
Section 3. The constituent behaviour used in the FE and semi-analytical
indices i.e. i,j,k take values 1,2,3, and Greek indices i.e. α,β,γ take values
analysis is presented in Section 4. The FEM together with the boundary
1,2 . The relation between nominal stress rates and velocity gradients is
conditions are shown in Section 5. Finally the results from the analyses
given by
tij̇ = Cijkl vl,k,
are shown, compared and discussed in Section 6 and Section 7 con-
cludes the paper. (5)
where tij̇ are components of the nominal stress rate, vl,k are the velocity
2. Constitutive model
gradients and Cijkl are components of nominal moduli. The moduli Cijkl
can be calculated using

Cijkl = Lijkl − τkj δil− τlj δik− τil δkj + τik δlj,
In the following section a constitutive model is derived that can take
imperfect cohesion between fibre and matrix into account. The model is 1 1 1 1
inspired by the two constitutive models derived by Christoffersen and 2 2 2 2 (6)


7.5 Article P2 83

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

Fig. 1. Perfect bonding between constituents. Three-dimensional (a–c) and two-dimensional (d–f) illustrations of the initial state, shear- and transverse deformation are shown.

Fig. 2. Imperfect bonding between constituents. Three-dimensional (a–c) and two-dimensional (d–f) illustrations of the initial state, shear- and transverse deformation are shown.

where Lijkl are components of a tensor of elastic–plastic tangent moduli Plane strain conditions is used for the moduli Lαβγδ . The relation be-

τ ˚ij = Lijkl εkl . δij is Kronecker’s delta and τij are Kirchoff stresses. Lijkl
relating the Jaumann rates of Kirchoff stresses to strain rates tween Kirchhoff stresses and Cauchy stresses is given by

τij = J σij, (8)

Lijkl = Ljikl = Lijlk , Lijkl = Lklij .
must satisfy the minor and major symmetries

giving the volume ratio J = V / V0 .

(7) where σij are components of the Cauchy stresses and J is the Jacobian
When the tangent moduli are used in two dimensions then Latin
indices are replaced by Greek indices Lαβγδ and so are the moduli Cαβγδ .


84 Chapter 7 - P2

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

det(M ) = (δαγ δβδ−δαδ δβγ ) Mγα Mδβ.

2.2. Constitutive model with imperfect bonding 1
2 (20)
Inspired by the assumptions from the latter constitutive models with It is convenient to rewrite the velocity gradients v∊f,2 and v∊m,2 in the
perfect and imperfect bonding, a transition between the two models is equivalent forms

vεf,2 = −c m κ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1−c m κ Hεζ (κ C2fζ 2δ−µ C2mζ 2δ ) vδ,2 + vε,2,
suggested. Introducing the factors µ and κ , where µ takes values be-
tween 0 and 1 and κ is calculated based on µ according to
vεm,2 = c f µ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + c f µ Hεζ (κ C2fζ 2δ−µ C2mζ 2δ ) vδ,2 + vε,2.
κ (µ) = ⎛1− m ⎞ µ + m ,
⎝ c ⎠
1 1
c (9)

which is a linear relation going from κ (µ = 0) = 1/ c m to κ (µ = 1) = 1.

(21)
Using Eq. (13) and (16) the overall nominal stress rates t1̇ α can be
The present constitutive model also assumes that material lines parallel determined

t1̇ β = c f C1fβ1δ vδ,1 + c f C1fβ2δ vδf,2 + c m C1mβ1δ vδ,1 + c m C1mβ2δ vδm,2.

with the fibres are subjected to a common stretching and rotation.
Further the fibres is aligned with the x1-axis prior to deformation this (22)
entails Replacing vδf,2 and vδm,2 using Eq. (21) and collecting terms related to
vαf,1 = vαm,1 = vα,1,
vα,β yields

t1̇ β = [c f C1fβ1δ + c m C1mβ1δ−c f c m (κ C1fβ2ε−µ C1mβ2ε ) Hεζ (κ C2fζ 1δ−µ C2mζ 1δ )] vδ,1
(10)

+ [c f C1fβ2δ + c m C1mβ2δ−c f c m (κ C1fβ2ε−µ C1mβ2ε ) Hεζ (κ C2fζ 2δ

where the index (•)α takes values 1 and 2. Inspired by the assumptions
from Eq. (1) and (3) and using the introduced factors µ and κ suggest

vα,2 = c f µ vαf,2 + c m κ vαm,2. (11) −µ C2mζ 2δ )] vδ,2. (23)

Using the last part of Eq. (12) then t2̇ α can be determined as
when µ = 1 and imperfect bonding for µ = 0 . Tractions on planes
Using the proposed equation yields the case of perfect bonding
t2̇ β = [c f C2fβ1δ + c m C2mβ1δ−c f c m (κ C2fβ2ε−µ C2mβ2ε ) Hεζ (κ C2fζ 1δ−µ C2mζ 1δ )] vδ,1
+ [c f C2fβ2δ + c m C2mβ2δ−c f c m (κ C2fβ2ε−µ C2mβ2ε ) Hεζ (κ C2fζ 2δ
parallel with the fibres can be calculated according to

t2̇ fα= µ t2̇ α,

̇ α= κ t2̇ α,
t2m −µ C2mζ 2δ )] vδ,2. (24)
t2̇ α= cf t2̇ fα + ̇ α.
c m t2m (12) The latter two equations may be written in a similar form as Eq. (5)
with

Cαβγδ = c f Cαβγδ + c m Cαβγδ −c f c m (κ Cαβ 2ε−µ Cαβ 2ε ) Hεζ (κ C2ζγδ −µ C2ζγδ ).

To have overall equilibrium

t1̇ α = c f t1̇fα + c m t1ṁ α.

f m f m f m

(13)
(25)
Using the relation between the Jaumann rate of Cauchy stresses and This is the final equation for the constitutive model relating the
nominal stress rates
σ ˚ij = tij̇ −ωjk σik−ωik σkj−σij vk,k + σjk vi,k,
overall nominal stress rates to velocity gradients. Using the constitutive
(14) equation, the overall response of a composite can be determined with
information about imperfect bonding based on µ . When the overall
where σ˚ij are components of the Jaumann rate of Cauchy stresses and deformations are determined in an incremental scheme then the velo-

(21). In the case of perfect bonding µ = 1, κ = 1 Eq. (25) simplifies to

ωij is the spin tensor, then it can be proven that the components of the city gradients for the constituents can be retrieved using Eqs. (10) and
Cauchy stresses are

σ11= c f σ11f + c m σ11 µ= 1→

Cαβγδ= c f Cαβγδ + c m Cαβγδ −c f c m (Cαβ 2ε−Cαβ 2ε ) Hεζ (C2ζγδ −C2ζγδ ),
m
,
σ12f = µ σ12, = κ σ12,
m f m f m f m
σ12
σ22f = µ σ22, m
σ22 = κ σ22, Mαβ= c m C2fα 2β + c f C2mα 2β, (26)
(15)

where moment equilibrium σij = σji is satisfied. The relation between which is identical to the expression derived by Christoffersen and

µ = 0, κ = 1/ c m the constitutive equation simplifies to

rates of nominal stresses and velocity gradients given by Eq. (5) can Jensen (1996) [12]. In the other extremum of decohesion

µ= 0→
equivalently be used for the two constituents. The relation given by Eq.

Cαβγδ= c f Cαβγδ + c m Cαβγδ −c f Cαβ

(5) together with Eq. (10) can be written as
= Cαβ1δ vδ ,1 + Cαβ 2δ vδ ,2,
ċ f m f f
2ε Hεζ C2ζγδ ,
c c c
tαβ
Mαβ= C2fα 2β,
(16)

where the superscript (•)c indicates one of the constituents. As derived (27)
in Appendix A, the velocity gradients vεf,2 and vεm,2 can be expressed
which is identical to the suggested expression presented in Jensen
through overall composite velocity gradients
(1999) [13]. In both of the simplified expressions, the inverse Hαβ can
vεf,2 = −c m κ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + µ Hεζ C2mζ 2δ vδ,2, be calculated using Eq. (18). The elastic-plastic tangent moduli can be

vεm,2 = c f µ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + κ Hεζ C2fζ 2δ vδ,2,

determined using Eq. (6) to be

Lαβγδ = Cαβγδ + τγβ δαδ + τδβ δαγ + ταδ δγβ− ταγ δδβ.
(17) 1 1 1 1
where Hαβ denotes 2 2 2 2 (28)

= (δαβ δγδ−δαδ δβγ ) Mδγ ,

1 Combining this with the derived constitutive equation (Eq. (25))
Hαβ and utilizing the properties from Eq. (15) yields the tangent moduli for
det(M ) (18)
the composite

Lαβγδ = c f Lαβγδ + c m Lαβγδ −c f c m (κ Cαβ 2ε−µ Cαβ 2ε ) Hεζ (κ C2ζγδ −µ C2ζγδ ).

which is the inverse of Mαβ given by

Mαβ = c m κ 2 C2fα 2β + c f µ2 C2mα 2β,

f m f m f m

(19) (29)
and det(M ) denotes the determinant calculated according to The tangent moduli for the composite satisfy the minor and major


7.5 Article P2 85

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

det(det(M ) Cαβγδ nα n γ ) = 0.
symmetries given by Eq. (7). rewritten

(37)
2.3. Kink band bifurcation
When calculating the determinant in Eq. (37) a polynomial form is
The constitutive equation is now applied in a kink band study, thus obtained
0 = (E f )5 d5 + (E f ) 4 d4 + O (E f )3.
where no initial fibre waviness is present and thereby kink band for-
(38)
mation at a bifurcation point is searched for. The applied method for

The solution to d5 = 0 gives the closed form solution

finding the bifurcation load associated with strain localisation was The term d5 will dominate the solution for high fibre stiffness E f .
formulated by Rice (1976) [24] and was used in a kink band analysis in

1 ⎛ m σ m−κ σ22 ⎞
0 = σ11−σ22 + L1212− 11 + m 2 (L1222 −c f κ µ σ12)tanβ
Christoffersen and Jensen (1996) [12]. The equilibrium equation can be
2
cm κ2 ⎝ ⎠
m

̇ ,α = 0.
written in terms of the rate of nominal stresses ⎜ ⎟

2 c κ
+ m 2 (L2222 −κ σ22)tan2 β,
tαβ (30) 1 m
c κ (39)

Cαβγδ vδ,γα = 0.
Combining this with Eq. (5) yields
(31) which can be compared to the expressions obtained by Christoffersen
and Jensen (1996) [12] and Slaughter et al. (1993) [11] with no initial
In the bifurcation analysis a nonhomogeneous and sufficiently fibre imperfection. Evaluating Eq. (39) for perfect bonding and using
smooth velocity field is searched for where the fibres starts to rotate and

µ= 1→
Eq. (6) leads to
overall strains localize into a band with orientation β as shown in Fig. 3.

0= σ11−σ22 + + −c f σ12)tanβ +
vα = fα (nβ xβ ),
The velocity field searched for takes the form
1 2 1
Cm
c m 2121 cm
m
(L1222 C m tan2 β,
c m 2222 (40)
(32)
where nα = (cos(β ), sin(β )) is the unit normal to the localization band.
= L2212
which is identical to the expression derived by Christoffersen and
m m
Using the introduced velocity field in the equilibrium equation yield Jensen (1996) [12] where the symmetry L1222 has been utilized.

= 0. µ= 0→
In the case of imperfect bonding Eq. (39) yields
d 2fδ (nε x ε )

0= σ11−σ22 + c mC2121 + 2 c m L1222 tanβ + c m C2222

Cαβγδ nα n γ
d (n ζ x ζ )2 (33) m m m
tan2 β, (41)
The bifurcation point is when the determinant of the latter equation
which is nearly identical to the expression derived in Jensen (1999)

det(Cαβγδ nα n γ ) = 0.
vanishes
[13] with the only difference that in the current expression no terms are
(34) present in front of σ22 . This is due to the assumptions of the stress
equality given by Eq. (15). In the current case of imperfect bonding, it is
Until this point, no assumptions have been made regarding the
assumed that no shear- and transverse stresses are transmitted to the
elastic–plastic behaviour of the constituents. When concerning the
fibres. The same conclusion can be made as done by Jensen (1999) [13]
idealized situation of infinitely rigid fibres then a closed form solution
that the two expressions Eq. (40) and (41) corresponding to the two
for the bifurcation point is obtainable. The method to obtain the solu-
extreme cases; perfect bonding and no bonding essentially differs by a
tion follows the method introduced in Jensen (1999) [13]. The com-
factor of (c m)2 . In the experiments conducted by Dève (1997) [20] the
ponents of the tangent moduli for the fibre constituent are written in
compressive strength of coated and uncoated fibres were compared.

∼
the form

= E f Lαβγδ , ≈ −4000 [MPa] was obtained. Only one ex-

The uncoated fibres had good interfacial bonding and a mean com-
f cr
Lαβγδ (35) pressive strength of σ11

= −2600 [MPa] was obtained. Assuming

periment with coated fibres (weakly bonded) was conducted where a
where the moduli are proportional to the stiffness modulus E f . Multi-
that the uncoated fibres were perfectly bonded µ = 1, a matrix volume
cr
compressive strength of σ11
plying Eq. (25) with the determinant det(M ) given by Eq. (20) will not
fraction of c = 0.5 and only σ11 stresses were present then this corre-
sponds to an interfacial bonding µ = 0.76 using Eq. (39).
change the bifurcation point. By doing so, the constitutive equation can m
be written as polynomials of E f

det(M ) Cαβγδ = (E f )3 Cαβγδ

(3)
+ (E f )2 Cαβγδ
(2)
+ E f Cαβγδ
(1)
+ Cαβγδ
(0)
, (36)
3. Kink band formation with the constitutive model
(3)
where the superscripts Cαβγδ means the moduli associated with the given
polynomial order of E f . The condition for bifurcation Eq. (34) can be The developed constitutive model is implemented in a semi-analy-
tical kink band analysis with an initial fibre imperfection. The method
was introduced in Jensen and Christoffersen (1997) [14]. Since the
implementation is similar to the previous some of the details are
skipped. Two material states are introduced, one inside the kink band
(band system) and one outside (base system) as illustrated in Fig. 4. The
fibres are localized in a band with an inclination β o and β i when re-
presented in the base- and band system respectively. The fibres are
aligned with the x1-axis prior to the deformation increment. A small
fibre rotation ϕ0 is introduced initially as an imperfection. The dis-
placement gradients for the composite is updated numerically due to a
prescribed rotation increment of the fibres in the band ϕ .̇ The governing

Cαβγδ vδ ,γ −σαβ vγ ,γ + vα,γ σγ ,β = σ̇αβ.

equation for the material states is
(42)
Two different numerical implementations in the incremental
Fig. 3. Geometry of composite model with fibres oriented in the x1 direction. The angle β loading procedure leading to identical predictions were presented by
is the orientation of the localization band searched for in the bifurcation analysis.
Jensen and Christoffersen (1997) [14]. The first version is utilized here


86 Chapter 7 - P2

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

σeq = σij σij− σii σjj .

3 1
2 2 (52)

The remaining terms introduced in Eq. (50) is the bulk modulus, K,

the secant shear modulus, Gs , and the tangent shear modulus, Gt . The
two latter moduli can be calculated based on the secant and tangent
moduli Es and Et
3 1−2 ν 3 1−2 ν
= − = − K=
3(1−2 ν )
1 1 E
, , ,
Gs Es E Gt Et E (53)

where ν is Poisson’s ratio and E is Young’s modulus. The nonlinear

Fig. 4. Kink band geometry. relationship between logarithmic strain εeq and the stress σeq is given by
the Ramberg-Osgood relation

3 σy ⎛ σeq ⎞
εeq = +
n
where the basis rotates with the fibres, leading to σeq
= 0, = 0, 7 E ⎝ σy ⎠
⎜ ⎟ ,
o
v2,1 i
v2,1 (43) E (54)

where superscripts (•)o and (•)i refer to quantities described in the base where σy is a reference yield stress and n is a hardening exponent. Both
and band system, respectively. A spin tensor is introduced Ωαβ with the the secant modulus Es and tangent modulus Et can be determined based
components on the Ramberg-Osgood relation. The J2 -deformation theory is used for

Ω21 = −Ω12 = ϕ,̇ Ω11 = Ω22 = 0.

the constitutive model and is also implemented as a UMAT in Abaqus
(44) for the finite element model presented in Section 5.
Using the spin tensor then the components of the velocity gradients, The material parameters used for the constitutive model and par-

formation, are vαi ,β + Ωαβ . The unit normal nα and unit tangent tα to the
seen from a basis in the band system that stays stationary during de- tially the FEM are normalised quantities of the parameters obtained by
Kyriakides et al. (1995) [2] and are given by

c f = 0.6, E f / E m = 35, ν f = 0.263,

band for the two systems are calculated using

(n1o,n2o) = (cosβ o,sinβ o), (t1o,t2o) = (−sinβ o,cosβ o), v m = 0.356, σ y,m/ E m = 0.013, nm = 4.
= (cosβ i,sinβ i ), = (−sinβ i,cosβ i ).
(55)
(n1i,n2i ) (t1i,t2i ) (45) The fibres are assumed to behave linear elastic although Kyriakides
Continuity of velocities along the boundary of the band requires et al. (1995) [2] observed nonlinear behaviour of the fibres. The fibre

et al. (1995) [2] used an initial imperfection of ϕ0 ≈ 1.5° and compared

that nonlinearity have a minor effect on the kink band instability. Kyriakides

(vαi ,β + Ωαβ ) tαi tβi = vαo,β tαo tβo, (vαi ,β + Ωαβ ) nαi tβi = vαo,β nαo tβo. (46) results obtained using linear and nonlinear fibres. The tendencies were
not influenced significantly based on their results. Jensen and Chris-
Continuity of tractions across the boundary requires that

(vδi ,γ + Ωδγ ) nαi nβi= Cαβγδ

toffersen (1997) [14] concluded that nonlinear fibres have a significant
i o
Cαβγδ vδo,γ nαo nβo, influence on the bifurcation stress but goes asymptotically towards the

(vδi ,γ + Ωδγ ) nαi tβi= Cαβγδ

i o same stress for increasing imperfections ϕ0 .
Cαβγδ vδo,γ nαo tβo. (47)

The relative rotation increment of the band seen from the two bases 5. Finite element model
are calculated according to

β ̇ = −vαo,β nαo tβo, β ̇ = −vαi ,β nαi tβi.

To validate the usability of the constitutive model developed in
o i
(48) Section 2, then a three-dimensional finite element model is created. In
the FEM the fibre and matrix constituents are discretized and are as-
The volume fraction of fibres inside and outside the band change
sumed to be hexagonally distributed comparable to the set-up in-
due to transverse deformation according to

c ̇ f = c f c m (v2,2−v2,2
troduced by Hsu et al. (1998) [26]. Utilizing the symmetries of the
f m
). (49) assumed fibre distribution, the cross section used in the simulation can

= 0 is used,
be seen in the unit cell shown in Fig. 5. The diameter of the fibres

furthermore σ22 = 0 together with Eq. (46) and (47) give five boundary
o
Plane strain conditions and the boundary condition v1,2 entering the simulation is

d = w0
conditions for the last five unknown velocity gradients. 6c f
,
π 3 (56)

the width t0 = 3/2 w0 . ABAQUS 6.14 is chosen to solve the finite

4. Constituent behaviour where w0 is the width of the cross section, and the thickness is related to

The Stören and Rice (1975) [25] version of J2 -deformation theory is element problem. 20-node quadratic brick- and 15-node quadratic tri-

of Kirchoff stresses to strain rates τ ˚ij = Lijkl εkl̇ . The formulation is out-
chosen as the elastic-plastic tangent moduli relating the Jaumann rate angular prism elements are chosen with full integration. Different mesh
configurations are investigated for the cross section as seen in Fig. 6.
lined here for completeness. The components of the tensor of in- Based on a convergence analysis, it is chosen to use the first mesh
stantaneous moduli are given by presented in Fig. 6a. The starting point for the material parameters was

Lijkl = Gs (δik δjl + δil δjk ) + ⎛K − Gs ⎞ δij δkl− (Gs−Gt ) mij mkl,
⎝ 3 ⎠
2 4 presented in Section 4. The kink band instability is sensitive to the
3 (50) elastic–plastic shear response of the composite as confirmed by Argon
(1972) [8], Budiansky (1983) [9], Slaughter et al. (1993) [11] and
where mij is given in terms of Cauchy stresses several others since then. In the FEM created by Hsu et al. (1998) [26]

mij = ⎛σij− δij σkk ⎞,

2 σeq ⎝ ⎠
1 1 they experienced that the material parameters for the matrix con-
3 stituent needed to be changed going from a two-dimensional to a three-
(51)
dimensional FEM based on the overall shear response of the composite.
where σeq is the equivalent von Mises stress A similar investigation of the pure shear response is investigated here


7.5 Article P2 87

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

where small letters x α refer to deformed coordinates and capital letters

Xα are the undeformed coordinates with a reference coordinate system
placed in the centre of the geometry. b is the width of the imperfection,
β is the band inclination and ϕ0 is the largest imperfection angle and is
located in the geometry centre. The imperfection is an modified version
of the one introduced by Srensen et al. (2008) [27]. The quantities
X1,min and X1,max are calculated using
−b
X1,min = −X2 tanβ, X1,max = −X2 tanβ.
b
2 cosβ 2 cosβ (58)

The geometrical parameters used in the simulations are

L0 = 1, w0 = 0.01, b = 1, β = 0. (59)

The band inclination is set to β = 0 in the simulations which is

conditions. The width w0 = 0.01 is chosen based on a convergence study

compatible with the current implementation of the periodic boundary

where the ratio L0 / w0 was increased until the fibre bending effects were
Fig. 5. Unit cell of the 3D finite element model with fibres oriented in the x1 direction. negligible.
The FEM is composed of four parts: three fibre parts and one matrix

perfect bonding ( µ = 1) all the parts are tied together using a surface-
part. Two different constraints are used between the parts. In the case of
where a two-dimensional and three-dimensional model is compared.
based tie constraint. In the case of no bonding ( µ = 0 ) a hard-contact
The unit cell for the three-dimensional model is shown in Fig. 5. The
boundary conditions for the two models are shown in Fig. 7. The ab-
constraint is set-up between the surface normals and a tangential fric-
breviation (PBC) stands for periodic boundary conditions. The shear
state ( µ = 0.5) half of the fibre surface area is not bonded but is allowed
tionless behaviour is chosen. In the case of an intermediate bonding
stress vs shear strain response is shown in Fig. 8. Three different curves
are presented. The first and second curve show the shear response of the
to slide frictionless and the rest is perfectly bonded to the matrix. A time
two- and three-dimensional model where the same material parameters
independent kink band analysis is performed using Abaqus Standard. A
are chosen for the constituents from Section 4. The third curve show the

E m (3D)/ E m = 0.5 which is normalised with the two-dimensional mod-

nonlinear arc-length method (Static, Riks) is used to solve the problem
response of the three-dimensional FEM with a Young’s modulus
due to snap-back and snap-through behaviour in the equilibrium path.

Based on the current simulation a modified modulus E m (3D)/ E m = 0.5 is

ulus. The shear response for the first and third curve nearly coincide.
6. Results
chosen for the three-dimensional finite element model. The boundary
conditions for the FEM for kink band formation is shown in Figs. 9 and The results presented are a combination of results obtained by the
10. Focusing on Fig. 9, the left surface is simple supported and a dis- semi-analytical kink band model and the finite element model. The
placement is applied to the right surface. Periodic boundary conditions constitutive formulation presented in Section 2 is implemented in the
between the lower and upper part are utilized to imitate an infinite semi-analytical numerical scheme presented in Section 3. The boundary
band and to avoid an extensive amount of degrees of freedom. Out of conditions and the assumptions behind the finite element model was
plane deformation is constrained as shown in Fig. 10. An imperfection presented in Section 5. The behaviour and material parameters of the
is introduced as a fibre waviness where the deformed coordinates are constituents used in the two models was presented in Section 4. The
calculated based on only deviation between the material parameters is the elastic modulus

x2= 2 ϕ0 −X2 tanβ for X1 < X1,min ,

−b
of the matrix constituent. As explained in Section 5 the Young’s mod-
1
( 2 cosβ ) malized by itself and in the FEM it is set to E m (3D)/ E m = 0.5. In the semi-
ulus for the matrix constituent in the semi-analytical model is nor-

x2= 2 ϕ0 ⎡ 2 π cosβ sin + X2 tanβ ) + X1⎤ for X1,min < X1

⎣ ( ) ⎦
1 b 2 π cosβ
(X1 analytical model, fibre rotation increments ϕ ̇ are specified in the loca-

< X1,max ,
b
lized band. In the FEM displacement increments is specified on the right

x2= 2 ϕ0 −X2 tanβ for X1,max < X1,

side of the structure, Fig. 9 and a solution for the equilibrium path is
1
( b
2 cosβ ) searched for. Based on the deformed geometry it was possible to cal-
culate the fibre rotation ϕ for a given increment, which was the max-
(57) imum rotation of the fibres with respect to the x1-axis.

Fig. 6. Illustration of different mesh configurations for the cross section. The meshes are composed of 20-node quadratic brick- and 15-node quadratic triangular prism elements.


88 Chapter 7 - P2

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

Fig. 7. Boundary conditions for the two-dimensional and three-dimensional FEM used to evaluate the pure shear response.

where G f and G m are the elastic shear moduli of the fibres and matrix,

G m = 1/2/(1 + ν ) is also normalized by the elastic modulus E m . The

respectively. The shear modulus of the matrix material

same shear modulus is used for the results from the semi-analytical

initial imperfection of ϕ0 = 0.5° are presented in Fig. 11. The results

model as well as the FEM. Results obtained from the analyses with an

from the FEM are presented in Fig. 11a and from the semi-analytical

three different cohesion levels µ . The critical stresses obtained for µ = 1

model in Fig. 11b. Three curves are presented in the figures based on

(perfect bonding) and µ = 0.5 (partially bonded) are nearly identical

between the two models. For µ = 0 the FEM predicts slightly lower

quantitative agreement is seen between the models. In the case of µ = 0

stresses and no stability point is seen on the figure. However, quite good

the response of the composite is similar to a model with only matrix and

modulus E m (3D)/ E m = 0.5 is used compared to the modulus used for the
voids. This explains the lower stresses predicted by the FEM where the

semi-analytical model which is unity due to the normalisation with E m .

initial imperfection ϕ = 3°. The same conclusions can be made for the
In Fig. 12 the normalised stress versus fibre rotation is shown for an

Fig. 8. Shear stress vs shear strain for unit cells exposed to pure shear deformation. tendencies between the models as for the imperfection ϕ = 0.5°. There

µ = 1 and µ = 0.5. The shear response between the models was aligned
is a good agreement between the FEM and the semi-analytical model for

in Section 5 for the case of perfect bonding µ = 1. Since the critical kink
In Figs. 11 and 12 the normalised stress versus fibre rotation is
shown for different cohesion levels µ and initial fibre imperfections ϕ0 .
The stress σ11 is normalised with the elastic composite shear modulus G band stress correlates well between the models it is verified again that
calculated according to the kink band stress depends on the shear response of the composite.
Similar tendencies for different cohesion levels are seen between the

= f + m
1 cf cm FEM and semi-analytical model despite the substantial differences in
G G G (60) the numerical set-up.


7.5 Article P2 89

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

Based on the results, an interfacial bonding factor of µ = 0.76 was

In Section 2 results obtained by Dève (1997) [20] were presented.

computed using Eq. (39). The result can be interpreted as that 76% of
the fibre surface area was perfectly bonded in the experiment with
weakly bonded fibres. The interpretation can be made due to the strong

tween µ = 1 and µ = 0.5.

correlation between the FEM and semi-analytical model for values be-

In Fig. 13 the critical stress −σ11

cr
/ G versus initial fibre imperfection
Fig. 9. Three-dimensional visualization of the boundary condition for the finite element ϕ0 is shown for different cohesion levels µ . The semi-analytical kink
kink band model.
band model is used to generate the results shown. The FEM has not
been used to generate comparable results due to the substantial time
required per simulation. The critical stresses presented in Fig. 13 are
evaluated for the most critical kink band angle β0 which is found by

angles varying from β0 = 0° to β0 = 10°. The critical band inclination

executing 11 simulations for every imperfection ϕ0 with different β0

varies from β0 = 0° for ϕ0 ≈ 0° to β0 = 7° for ϕ0 = 5° in the case of

perfect bonding µ = 1. The constitutive model predicts that nearly the
same inclinations β0 is critical for other cohesion levels with the current
material parameters. Based on Fig. 13 it can be concluded that the
critical stress in the case of imperfect cohesion is also sensitive to fibre
imperfections. For all values of µ a steep descent is seen in the begin-
ning going from zero imperfection to increasing initial imperfections.
For descending values of µ the curves stabilize around a lower critical
value more rapidly.
The time used to solve the semi-analytical- and the finite element
simulation is noticeable different. To compare the simulation time then

Fig. 10. Boundary condition on the cross section x2−x3 plane for the kink band model.
the simulations are run on the same computer: A Lenovo T550 with a
i7-5600 Processor (2.6 GHz) with 16 GB of memory (RAM). A typical
simulation using the semi-analytical model takes approximately 20 s

Fig. 11. Non-dimensional stress versus fibre rotation ϕ , for an initial fibre imperfection of ϕ0 = 0.5° with different bonding parameters µ .

Fig. 12. Non-dimensional stress versus fibre rotation ϕ , for an initial fibre imperfection of ϕ0 = 3° with different bonding parameters µ .


90 Chapter 7 - P2

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

for the constituents behaving either elastic or elastic-plastic. A con-

stitutive equation is developed relating stress rates to strain rates in a
general framework and the laws set by objectivity are fulfilled. The
fibre and matrix bonding is controlled using a factor µ going from zero
to unity. In the two extreme cases of perfect bonding and no bonding, it
was proven that the developed constitutive equation leads to identical
equations as earlier developed expressions. A new analytical expression
is derived giving the bifurcation point of kink band formation in the
case of high fibre stiffness considering imperfect bonding.
A semi-analytical kink band analysis, with the constitutive equation
implemented, is compared with a three-dimensional micromechanical
kink band finite element model. Results obtained based on different
initial fibre imperfection is compared and the results correlates very

Fig. 13. Normalised critical kink band stress −σ11

well despite the substantial differences between the models. The in-
cr
/ G versus initial fibre imperfection ϕ0 fluence of the fibre/matrix bonding upon the critical stress versus initial
for different bonding parameters µ . The critical stresses are shown for the most critical fibre imperfection was investigated. A notable decrease in critical kink
band inclination β0 . band stress is seen when imperfect bonding is taking into account. For
all levels of bonding, it is seen that a steep descent in the critical stress is
including post processing for a simulation with 400.000 increments. A seen going from no imperfection to increasing imperfections.
typical simulation with the FEM takes about 6 h for a simulation with The comparison between the FEM and the semi-analytical model
400 pseudo time increments. The 6 h of simulation was with the mesh verifies the derived constitutive model for imperfectly bonded fibre-
presented in Fig. 6a and as mentioned earlier, periodic boundary con- reinforced composites. The constitutive model makes it possible to in-
ditions were utilized to save computation time. vestigate complex geometries and fibre layouts with information about
imperfect bonding including geometric and material nonlinearities.
7. Conclusions
Acknowledgements
A constitutive model for imperfectly bonded fibre-reinforced com-
posites has been developed. Independent constitutive laws can be used This work was supported by the Thomas B. Thriges fund.

Appendix A. Velocity gradients v cε,2

Using Eq. (12) (κ t2fα = µ t2mα ) together with Eq. (16) yields

κ C2fζ 1δ vδ,1 + κ C2fζ 2δ vδf,2 = µ C2mζ 1δ vδ,1 + µ C2mζ 2δ vδm,2, (A.1)

where the Greek letter ζ is used as the free index for t2cζ . Rearranging the latter equation and multiplying with c m κ yields

c m κ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + c m κ 2 C2fζ 2δ vδf,2 = c m κ µ C2mζ 2δ vδm,2. (A.2)

Rearranging Eq. (11) and multiplying with µ C2mζ 2δ yields

c m κ µ C2mζ 2δ vδm,2 = µ C2mζ 2δ vδ,2 + c f µ2 C2mζ 2δ vδf,2. (A.3)

The right side of Eq. (A.2) and the left side of Eq. (A.3) is identical. By combining the equations and isolating terms related to vδf,2 leads to

Mζδ vδf,2 = −c m κ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + µ C2mζ 2δ vδ,2, (A.4)

where Mαβ is given by Eq. (19). The inverse Hαβ is given by Eq. (18). The terms Mαβ and the inverse Hαβ comply to
Hαβ Mβγ = δαγ . (A.5)
Using the latter identity together with Eq. (A.4) yields the velocity gradients v∊f,2 expressed through overall homogenized velocity gradients

vεf,2 = −c m κ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + µ Hεζ C2mζ 2δ vδ,2. (A.6)
Similarly the velocity gradients v∊m,2 can be determined

vεm,2 = c f µ Hεζ (κ C2fζ 1δ−µ C2mζ 1δ ) vδ,1 + κ Hεζ C2fζ 2δ vδ,2. (A.7)

References [4] Zhou HW, Yi HY, Gui LL, Dai GM, Peng RD, Wang HW, Mishnaevsky L. Compressive
damage mechanism of GFRP composites under off-axis loading: experimental and
numerical investigations. Compos Part B: Eng 2013;55:119–27. http://dx.doi.org/
[1] Bishara M, Vogler M, Rolfes R. Revealing complex aspects of compressive failure of 10.1016/j.compositesb.2013.06.007.
polymer composites Part II: Failure interactions in multidirectional laminates and [5] Nizolek T, Begley M, McCabe R, Avallone J, Mara N, Beyerlein I, Pollock T. Strain
validation. Compos Struct 2017;169:116–28. http://dx.doi.org/10.1016/j. fields induced by kink band propagation in Cu-Nb nanolaminate composites. Acta
compstruct.2016.10.091. Mater 2017;133:303–15. http://dx.doi.org/10.1016/j.actamat.2017.04.050.
[2] Kyriakides S, Arseculeratne R, Perry EJ, Liechti KM. On the compressive failure of URLhttp://www.sciencedirect.com/science/article/pii/S1359645417303427.
fiber reinforced composites. Int J Solids Struct 1995;32(6–7):689–738. http://dx. [6] Nair SN, Dasari A, Yue CY, Narasimalu S. Failure behavior of unidirectional com-
doi.org/10.1016/0020-7683(94)00157-R. posites under compression loading: Effect of fiber waviness, Materials 10 (8).http://
[3] Wadee MA, Hunt GW, Peletier MA. Kink band instability in layered structures. J dx.doi.org/10.3390/ma10080909.
Mech Phys Solids 2004;52(5):1071–91. http://dx.doi.org/10.1016/j.jmps.2003.09. [7] Rosen BW. Mechanics of composite strengthening. In: Fiber Composite Materials,
026. American Society for Metals, Metals Park, OH (1965) 37–75.


7.5 Article P2 91

S.P.H. Skovsgaard, H.M. Jensen &RPSRVLWH6WUXFWXUHV  ²

[8] Argon AS. Fracture of composites. Treatise Mater Sci Technol 1972:79–114. including fiber bending effects. J Mech Phys Solids 2016;94:433–52. http://dx.doi.
[9] Budiansky B. Micromechanics. Comput Struct 1983;16(1–4):3–12. http://dx.doi. org/10.1016/j.jmps.2016.04.010.
org/10.1016/0045-7949(83)90141-4. [19] Skovsgaard SPH, Jensen HM. Three-dimensional constitutive model for elastic-
[10] Fleck NA, Budiansky B. Compressive failure of fibre composites due to micro- plastic behaviour of fibre-reinforced composites. Int J Solids Struct 2018:1–13.
buckling. In: Proc. 3rd Symp. on Inelastic Deformation of Composite Materials; http://dx.doi.org/10.1016/j.ijsolstr.2018.01.032.
1991. p. 235–73. [20] Dève HE. Compressive strength of continuous fiber reinforced aluminum matrix
[11] Slaughter WS, Fleck NA, Budiansky B. Compressive failure of fibre composites: the composites. Acta Mater 1997;45(12):5041–6. http://dx.doi.org/10.1016/S1359-
roles of multi-axial loading and creep. J Eng Mater Technol 1993;115(3):308–13. 6454(97)00174-2.
URLhttp://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA266398#page=265. [21] Zhang W, Zhang M, Ochiai S, Gu M. Experimental and simulation investigations of
[12] Christoffersen J, Jensen HM. Kink band analysis accounting for the microstructure tensile response of (Al2O3f/Al)w/2024Al composite. Mater Sci Eng, A
of fiber reinforced materials. Mech Mater 1996;24(4):305–15. http://dx.doi.org/ 2008;497(1–2):44–50. http://dx.doi.org/10.1016/j.msea.2008.07.048.
10.1016/S0167-6636(96)00052-X. [22] Jiang WG, Zhong RZ, Qin QH, Tong YG. Homogenized finite element analysis on
[13] Jensen HM. Models of failure in compression of layered materials. Mech Mater effective elastoplastic mechanical behaviors of composite with imperfect interfaces.
1999;31:553–64. Int J Mol Sci 2014;15(12):23389–407. http://dx.doi.org/10.3390/ijms151223389.
[14] Jensen HM, Christoffersen J. Kink band formation in fiber reinforced materials. J [23] Naya F, Herraez M, Gonzalez C, Lopes C, der Veen SV, Pons F. Computational
Mech Phys Solids 1997;45(7):1121–36. http://dx.doi.org/10.1016/S0022- micromechanics of fiber kinking in unidirectional FRP under different environ-
5096(96)00126-3. mental conditions. Compos Sci Technol 2017;144:26–35. http://dx.doi.org/10.
[15] Hsu S-Y, Vogler T, Kyriakides S. On the axial propagation of kink bands in fiber 1016/j.compscitech.2017.03.014.
composites: Part II analysis. Int J Solids Struct 1999;36:575–95. http://dx.doi.org/ [24] Rice JR. The Localization of plastic deformation. Theoret Appl Mech 1976:207–20.
10.1016/S0020-7683(98)00030-4. [25] Stören S, Rice JR. Localized necking in thin sheets. J Mech Phys Solids
[16] Wind JL, Steffensen S, Jensen HM. Comparison of a composite model and an in- 1975;23(6):421–41. http://dx.doi.org/10.1016/0022-5096(75)90004-6.
dividually fiber and matrix discretized model for kink band formation. Int J Non- [26] Hsu S-Y, Vogler TJ, Kyriakides S. Compressive strength predictions for fiber com-
Linear Mech 2014;67:319–25. http://dx.doi.org/10.1016/j.ijnonlinmec.2014.10. posites. J Appl Mech 1998;65(1):7. http://dx.doi.org/10.1115/1.2789050.
005. URLhttp://appliedmechanics.asmedigitalcollection.asme.org/article.aspx?
[17] Fleck NA, Shu JY. Microbuckle initiation in fibre composites: a finite element study. articleid=1413036.
J Mech Phys Solids 1995;43(12):1887–918. http://dx.doi.org/10.1016/0022- [27] Sørensen KD, Mikkelsen LP, Jensen HM. User subroutine for compressive failure of
5096(95)00057-P. composites. 2009 Simulia Customer Conference (1965); 2009. p. 618–32.
[18] Poulios K, Niordson CF. Homogenization of long fiber reinforced composites


Chapter 8

P3
Steady-state kink band propagation in layered materials

8.1 Motivation
Kink band formation is the dominant compressive failure mechanism for many fibre-
reinforced plastics. As explained in the Introduction, the fibre kinking strength (peak stress)
is dominated by the shear yield strength of the composite and fibre misalignments. The forma-
tion of a kink band can be divided into different stages; the linear response, incipient kinking,
peak stress, transient kinking and steady-state kink band broadening. The fibre misalignments
are difficult to estimate for a composite, thereby the exact compressive strength is unknown.
As the fibre misalignments increase in a composite, the peak strength decrease and approach
the propagation stress at kink band broadening. The stress at steady-state broadening can
thereby be seen as a lower bound for the critical compressive stress in composites.
The current publication deals with the investigation of the propagating instability known
as kink band broadening using three different methods. A case study is conducted with
material parameters from Kyriakides et al. (1995). The original source of the publication is:
(Skovsgaard and Jensen, 2018c).

8.2 Method
The phenomenon is investigated using three substantially different methods. A finite element
model (Chapter 2), a semi-analytical (Chapter 3) and an analytical model (Chapter 4) are
used.
94 Chapter 8 - P3

The FE-model consists of three layers; one matrix and two half fibre layers. Periodic
boundary conditions are applied on a skewed mesh to get a band inclination β . An arc-
length solver in Abaqus/CAE is used to solve the non-linear problem. The semi-analytical
approach is based on a kink band model developed by Jensen and Christoffersen (1997)
with a homogenised constitutive formulation from Christoffersen and Jensen (1996). The
analytical model, as described in Chapter 4, results in one transcendental equation for the
steady-state propagation state and is derived by evaluating the internal and external work at
the steady-state.

8.3 Contribution
The contributions to the publication comprise: the development of a finite element model
with periodic boundary conditions on a skewed mesh; the development of an analytical model
for kink band broadening; conducting the simulations and writing the paper.

8.4 Main findings

Initially, two main findings are the analytical model and the FE-model. The three models are
in mutually agreement. The models predict that for small band inclinations β , the predictions
for the fibre rotation φ exceeds the usual assumed linear relation at fibre lock-up φ = 2 β .
Furthermore, it was found that the stress at kink band broadening is largely influenced by the
matrix Poisson’s ratio and the uni-axial tangent modulus at large strains.
8.5 Article P3 95

Simon P. H. Skovsgaard
Department of Engineering,
Steady-State Kink Band
Aarhus University,
Inge Lehmanns Gade 10,
Aarhus C 8000, Denmark
Propagation in Layered Materials
e-mail: [email protected] Failure by steady-state kink band propagation in layered materials is analyzed using
three substantially different models. A finite element model and an analytical model are
Henrik Myhre Jensen developed and used together with a previously introduced constitutive model. A novel
Professor methodology for simulating an infinite kink band is used for the finite element model
Department of Engineering, using periodic boundary conditions on a skewed mesh. The developed analytical model
Aarhus University, results in a transcendental equation for the steady-state kink band propagation state. The
Inge Lehmanns Gade 10, three models are mutually in good agreement and results obtained using the models
Aarhus C 8000, Denmark correlate well with the previous experimental findings. [DOI: 10.1115/1.4039573]
e-mail: [email protected]

1 Introduction A study of the kink band instability in a carbon fiber composite

was conducted by Evans and Adler [6]. They analyzed different
It is well known that fiber-reinforced composites can have a
steps in the formation of kink bands including the steady-state
considerably lower compressive strength compared to the strength
propagation phenomenon. They observed the relation /  2 b,
in tension. Fleck [1] made a review of different failure modes
where / is the relative rotation of the fibers within the kink band
and stated that the compressive strength of carbon fiber-epoxy
and b is the kink boundary orientation. The relation / ¼ 2 b was
laminates are often less than 60% of their tensile strength. Long
later used by Moran et al. [7] to predict the kink band broadening
fiber-reinforced composites with a ductile and tough resin material
strength of a carbon fiber-reinforced composite. Furthermore, they
are prone to fail by plastic microbuckling which leads to the for-
assumed inextensible fibers and incompressibility of the compos-
mation of one or several kink bands. A recent investigation of the
ite. Hsu et al. [8] investigated the propagation of kink bands
compressive failure of fiber composites was conducted by Bishara
experimentally and these findings were compared with a micro-
et al. [2]. They investigated the compressive failure of a multidir-
mechanical finite element model (FE-model) in Hsu et al. [9] and
ectional fiber composite experimentally and numerically. They
reasonable correlation was seen between the studies as concluded
concluded that the failure initiated from the 0deg ply is due to the
by the authors. Kink band broadening has also been observed in
formation of a kink band. The failure was followed by delamina-
materials other than glass- and carbon fiber-reinforced compo-
tion, fiber-matrix debonding, and fiber cracking and was initiated
sites, e.g., Attwood et al. [10] observed broadening in a composite
in the vicinity of the kink band.
made of ultrahigh molecular weight polyethylene. Poulsen et al.
The critical compressive strength of fiber composites is sensi-
conducted a compression experiment on clear wood and observed
tive to initial fiber misalignments which is concluded by several
kink band broadening. Later, Byskov et al. [11] introduced a
authors, e.g., Budiansky [3] and Jensen and Christoffersen [4]. In
semi-analytical and a finite element model to investigate kink
an experimental kink band study conducted by Vogler and Kyria-
band propagation and the emphasis was given to wood and fiber
kides [5] on a AS4/PEEK composite, they showed that a kink
composites. A recent experimental investigation of the propaga-
band was formed under compressive loading in the direction
tion state was conducted by Nizolek et al. [12] on a Cu–Nb com-
of the fibers. A kink band was formed with a band orientation of
posite and they also observed the relation /  2 b.
b ¼ 15deg and an initial band width of 250 fibers (2mm). This
An investigation of steady-state kink band propagation is
initial state was followed by a state, where the band inclination
conducted in the present paper using three different models with
remained constant and the width of the band was broadened under a
substantial difference in model configuration. A finite element
constant applied stress. This is referred to as the steady-state kink
model and an analytical kink band model are proposed in the
band broadening stress or propagation stress. After the critical stress
present study and are compared with a previously developed
is reached, an unstable rotation of the fibers within a localized band
semi-analytical kink band model [13]. In contrast to the previous
will begin. Due to the kinematics of the developing kink band, a
investigations, the relation / ¼ 2 b is not used to obtain the
multi-axial strain state increases with increasing rotation. This unsta-
steady-state quantities. Instead, an approach of work balance is
ble rotation stops when it is energetically preferable to increase the
utilized for the semi-analytical and the analytical kink band
bandwidth instead of an increased rotation due to strain hardening.
model. Furthermore, the composites are assumed compressible
With increasing fiber imperfections, the compressive strength
which yields a more realistic estimate of the steady-state quanti-
of composites will tend toward the stress at steady-state kink band
ties. In contrast to the FE-model and semi-analytical model, inex-
broadening. The stress at steady-state is insensitive to fiber imper-
tensible fibers are assumed for the analytical model. In the present
fections and does only depend on the constitutive behavior of the
study, it is assumed that the matrix constituent deforms elastic/
composite. The steady-state broadening stress can thereby be seen
plastic in the whole regime, which can be unrealistic for some
as a lower bound for the critical compressive stress in composites
matrix materials due to the shear strains required for band broad-
with moderately large or unknown fiber imperfections. This lower
ening. Information about imperfect bonding and splitting between
bound of critical stress can be used to create fail safe structures and
the fiber/matrix interface can be included as in the studies con-
it is thereby valuable to predict the propagation state accurately.
ducted by Prabhakar and Waas [14] and Skovsgaard and Jensen
[15]. Information about splitting and imperfect bonding has not
Manuscript received January 22, 2018; final manuscript received March 5, 2018; been included in the current study, but good correlation is still
published online March 30, 2018. Assoc. Editor: Shaoxing Qu. seen between experiments and the results obtained as seen in

Journal of Applied Mechanics Copyright V

C 2018 by ASME JUNE 2018, Vol. 85 / 061005-1
96 Chapter 8 - P3

Sec. 5. A case study is conducted with material parameters vertical ends are far away from the deformation of interest, which
obtained by Kyriakides et al. [16] for a AS4/PEEK composite. is why the periodicity can be interpreted as shown in Fig. 1(b).
The paper is organized in six sections. The boundary conditions The study of an infinite kink band is in line with the classical
and material parameters for the proposed FE-model are presented methodology done by previous authors (Budiansky [3], Moran
in Sec. 2. Furthermore, a brief introduction into different steps in et al. [7] and Jensen [13]). The fixed geometries used in the simu-
the formation of kink bands is included in Sec. 2. A constitutive lations are
model developed by Christoffersen and Jensen [17] was used to
study kink band broadening in Jensen [13]. The latter model will L0 ¼ 1; w0 ¼ 0:01; b0 ¼ 0:5; /0 ¼ 2 deg (1)
be known as the semi-analytical model and is introduced in Sec.
3. An analytical model is introduced in Sec. 4 with the assumption where L0 and w0 are the initial length and width of the geometry,
of inextensible fibers together with simple assumptions for the respectively. The width w0 ¼ 0:01 was chosen based on a conver-
stress–strain relations. Results generated by the three models are gence study, where the width was decreased until the fiber bend-
presented in Sec. 5 and Sec. 6 concludes the paper. ing stiffness became insignificant. The length b0 controls the
extent of the imperfection. Small values of b0 lead to a localiza-
tion of the imperfection toward the geometry center and /0 is the
2 Finite Element Model largest misalignment angle in the imperfection. The geometries
A micromechanical FE-model is developed that can predict the b0, /0 , and b are used to introduce an initial fiber imperfection
steady-state kink band propagation state for a layered material or according to
fiber-reinforced composite. The FE-model can in general be used  
to investigate both the critical stress and the kink band propaga- 1 ÿb0
tion stress for arbitrary compositions of two constituents. In the x2 ¼ / 0 ÿ X2 tan b for X1 < X1;min ;
2 2 cos b
current paper, emphasis will be given to composites with rela-    
1 b0 2 p cos b
tively stiff fibers embedded in a ductile resin material with a lower x2 ¼ / 0 sin ðX1 þ X2 tan bÞ þ X1
stiffness and yield strength as in the case of glass- and carbon 2 2 p cos b b0
fiber-reinforced polymers. for X1;min < X1 < X1;max ;
In Fig. 1, an overview of the geometry and boundary conditions 1

b0

for the developed FE-model is shown. The commercial finite ele- x2 ¼ / 0 ÿ X2 tan b for X1 > X1;max (2)
ment software ABAQUS 6.14 is used for the simulations. ABAQUS/ 2 2 cos b
STANDARD is used, which is an implicit general-purpose finite ele-
ment program that tracks the equilibrium path during deformation. where xi are deformed coordinates and Xi are undeformed coordi-
A nonlinear arc-length method (Static, Riks) is used to solve the nates described in a coordinate system placed in the geometry
problem due to snap-back and snap-through behavior in the equi- center. The imperfection is a modified version of the one intro-
librium path. As seen in Fig. 1(a), the model consists of three duced by Sørensen et al. [18]. The quantities X1;min and X1;max are
layers: one matrix or resin layer and two half fiber layers. Periodic the X1-coordinates where the imperfections begin and end. Since
boundary conditions are utilized between the upper and the lower an inclination b is introduced, X1;min and X1;max are functions of
part of the geometry to simulate an infinite kink band as illustrated the X2-coordinate and are calculated according to
in Fig. 1(b). Inclined periodic boundary conditions are applied
ÿb0 b0
using a skewed mesh as shown in Fig. 2. The geometry ends are X1;min ¼ ÿ X2 tan b; X1;max ¼ ÿ X2 tan b (3)
vertical to accomplish a simple state of uniaxial loading. The 2 cos b 2 cos b

Fig. 1 (a) Geometry of the micromechanical finite element model. The darker regions represent the fiber constituent. (b)
Sketch of overall deformation.

Fig. 2 Illustration of mesh used in the finite element simulations. The coupling of the nodes locks
the deformation in an orientation b.

061005-2 / Vol. 85, JUNE 2018 Transactions of the ASME

8.5 Article P3 97

Fig. 3 Shear strain in matrix constituent, e12 , for five deformation stages during simulation for a
geometry with an initial imperfection /0 5 2 deg and a band orientation b 5 10 deg for a single strip
as shown in Fig. 1(a)

The fiber imperfection is introduced as a waviness using trigono- where e and r are the uni-axial strain and stress, respectively, ry is
metric functions, Eq. (2), where /0 is the largest angle with a reference yield stress, and n is a hardening exponent.
respect to the X1-axis. Simpler initial imperfections have been Young’s modulus of the fibers is 35 times the one for the
introduced earlier with promising results, e.g., Prabhakar and matrix. Due to this ratio, most of the bending will be carried out
Waas [14] and Davidson and Waas [19]. Simple imperfections as by the fibers and the matrix will obtain high shear straining.
the latter are not feasible in the current study. Both the band incli- Eight-noded biquadratic elements are used for the fibers with full
nation, b, and the geometry ends outside X1;min and X1;max , where integration, and four-noded bilinear elements with full integration
the fibers are aligned with the X1-axis, are of great importance to are used for the matrix. Even though the elements are incompati-
obtain the steady-state stress using the current model. ble, the errors will be small. The use of both eight- and four-
The critical stress is highly sensitive to the initial imperfection noded elements for the matrix has been compared and showed
(e.g., Wind et al. [20] and Fleck and Budiansky [21]), but the identical results. It has been possible to obtain a larger part of
steady-state broadening stress is independent of this imperfection, the post-buckling response using four-noded elements and are
and therefore, /0 and b0 can be chosen arbitrarily. The geometric therefore chosen.
parameter b is the inclination of the infinite band with respect to Five typical deformation stages are shown in Fig. 3, where the
the X2-axis. darker regions correspond to high shear straining of the matrix.
To account for an inclination b of the infinite band at the The five stages are designated in Fig. 4, where the stress ÿr11 nor-
steady-state, the coupling of the nodes is chosen as seen in Fig. 2. malized by the elastic shear modulus for the composite G is
Toward the ends of the geometry, the nodes are aligned with the shown as a function of the normalized end shortening u=L0 . Both
X2-axis, and from here, the mesh gradually becomes more skewed r11 and u are illustrated in Fig. 1(a). The response is initially lin-
until the coupled nodes are orientated with an angle of b with ear until the critical stress, rcr
11 , is reached at stage (2). The snap-
respect to the X2-axis. The displacement degrees-of-freedom are back behavior is clearly seen from stages (2) to (3), where both
coupled in pairs between the upper and the lower part using the the stress and the end displacement decrease. A large rotation of
Lagrange multiplier method in ABAQUS. This leads to identical dis- the fibers is also observed between stages (2) and (3). The fibers
placements of coupled nodes during deformation. Even though rotate further and lock-up at stage (4). From stage (4) to (5), the
that the band orientation, b, at initiation and at steady-state is in end shortening keeps increasing under steady-state conditions, at
general different, this will not influence the results at kink band constant loading and the fibers maintain the same steady-state
propagation when using a deformation theory for the constituent rotation /ss .
behavior. The same five stages are shown in Fig. 5, where the normalized
The material parameters used in the simulations are normalized stress ÿr11 =G is shown as a function of fiber rotation /. In the
values of the one obtained by Kyriakides et al. [16] and are given current figure, the band orientation is set to b ¼ 10 deg and the
by fibers lock-up in an angle approximately /ss  25 deg. The influ-
ence of multiple element layers within the matrix layer has been
cf ¼ 0:6; Ef =Em ¼ 35;  f ¼ 0:263; investigated and is shown in Fig. 6 for two simulations with 1
(4) and 4 element layers, respectively. The results are almost indistin-
vm ¼ 0:356; ry;m =Em ¼ 0:013; nm ¼ 4
guishable due to an identical initial response. Only a small part of
the post-buckling regime was obtainable with the FE-model with
where cf is the volume fraction of fibers. Superscripts ð•Þm and
multiple element layers due to lack of numerical convergence.
ð•Þf correspond to values associated with the matrix and fiber con- Apparently, the simulations tend toward the same steady-state
stituents. Em and Ef are the matrix and fiber Young’s moduli, response. One element layer is considered sufficient to obtain the
respectively.  m and  f are Poisson’s ratios for the two constitu- steady-state results presented in Sec. 5.
ents. The parameters ry;m and nm are the reference yield stress and
hardening exponent for the matrix constituent, respectively. The
nonlinear response of the matrix is described using a rate- 3 Semi-Analytical Model
independent J2-deformation theory, where the uni-axial response
is described using a Ramberg–Osgood relation according to Jensen [13] investigated kink band broadening using a constitu-
tive model developed in Christoffersen and Jensen [17]. The
model is reused in this paper for comparison with the other mod-
r 3 ry r n
 
e¼ þ (5) els. The current model will be known as the semi-analytical
E 7 E ry model. The theory behind this approach will be outlined here.

Journal of Applied Mechanics JUNE 2018, Vol. 85 / 061005-3

98 Chapter 8 - P3

Fig. 4 Applied normalized stress 2r11 /G as a function of normalized end shortening

u/L0 for simulation with a band orientation b 5 10 deg. The five deformation stages in
Fig. 3 are designated.

Fig. 5 Applied normalized stress 2r11 /G as a function of fiber rotation / for simula-
tion with a band orientation b 5 10 deg. The five deformation stages in Fig. 3 are
designated.


3.1 General Relations. The index notation is adopted, where where sij are Kirchhoff stresses and sij are components of the
Latin indices, i.e., i, j, k, take values 1, 2, 3, and Greek indices, 
Jaumann rate of Kirchhoff stresses determined by sij ¼ Lijkl e_ kl .
i.e., a, b, c, take values 1, 2. The constitutive model is described
using a time-independent constitutive law, where the constituents Combining the latter two equations, the relation between the rate
can be described independently. The strain rate and the spin are of nominal stresses and velocity gradients is
given by
t_ij ¼ Cijkl vl;k (8)
1 1
e_ ij ¼ ðvi;j þ vj;i Þ; xij ¼ ðvi;j ÿ vj;i Þ (6)
2 2 where Cijkl are components of nominal moduli
where vi;j are the velocity gradients. The rate of nominal stresses, 1 1 1 1
t_ij , is Cijkl ¼ Lijkl ÿ skj dil ÿ slj dik ÿ sil dkj þ sik dlj (9)
2 2 2 2

 and dij denotes Kronecker delta. The moduli Lijkl are to be

t_ij ¼ sij þ xjk sik þ xik skj ÿ sjk vi;k (7) described using a time-independent constitutive law, e.g., J2-

061005-4 / Vol. 85, JUNE 2018 Transactions of the ASME

8.5 Article P3 99

Fig. 6 Nondimensional stress versus normalized end displacement u/L0 and fiber rotation / for two simulations with 1
and 4 matrix element layers: (a) stress versus displacement and (b) stress versus fiber rotation

Fig. 7 Assumed deformation of the constituents based on overall deformations of the

composite. For illustrative purpose, the gray constituent reacts stiffer, as in the case of a
fiber embedded in a polymer matrix.

deformation and J2-flow theory. The moduli are to comply to the f m f m

t_21 ¼ t_21 ¼ t_21 ; t_22 ¼ t_22 ¼ t_22
symmetries (14)
f m f
t_11 ¼ cf t_11 þ cm t_11 ; t_12 ¼ cf t_12 þ cm tm
12
Lijkl ¼ Ljikl ¼ Lijlk (10)
By assuming a relationship between the rate of nominal stresses
as required by moment equilibrium sij ¼ sji and symmetry of the and velocity gradients in the form of Eq. (8) for both the constitu-
strain rates e_ ij ¼ e_ ji . The relation between Kirchhoff stresses sij ents and composite, Christoffersen and Jensen [17] showed that
and Cauchy stresses rij is the composite behavior could be described according to
sij ¼ J rij (11)
Cabcd ¼ cf Cfabcd þ cm Cm
abcd
where J is the Jacobian giving the volume ratio J ¼ V=V0 . The ÿ cf cm ðCfab2e ÿ Cm f m
ab2e ÞHef ðC2fcd ÿ C2fcd Þ (15)
relation between the rate of Cauchy stresses and the rate of nomi-
nal stresses is Mab ¼ cm Cf2a2b þ cf Cm
2a2b

r_ ij ¼ t_ij ÿ rij vk;k þ vi;k rkj (12) where Hab is the inverse of Mab complying to

Hac Mcb ¼ dab (16)

3.2 Constitutive Model by Christoffersen and Jensen. The
The velocity gradients vci;2 for the constituents can be calculated
constitutive model is based on the assumption that material lines
using the velocity gradients vi;j for the composite
parallel with the fibers are subjected to a common stretching and
rotation as shown in the first two geometries in Fig. 7. In addition,
the strains in the normal direction are averages of the constituent vfe;2 ¼ ÿcm Hef ðCf2f1d ÿ Cm m
2f1d Þ vd;1 þ Hef C2f2d vd;2
f f
(17)
strains which can be written using velocity gradients vm f m
e;2 ¼ c Hef ðC2f1d ÿ C2f1d Þ vd;1 þ Hef C2f2d vd;2

v1;1 ¼ vf1;1 ¼ vm
1;1 ; v2;1 ¼ vf2;1 ¼ vm
2;1 ; Equation (14) further entails the equalities
(13)
v1;2 ¼ cf vf1;2 þ cm vm
1;2 ; v2;2 ¼ cf vf2;2 þ cm vm
2;2
cf rf11 þ cm rm
11 ¼ r11 ; rf12 ¼ rm
12 ¼ r12 ; rf22 ¼ rm
22 ¼ r22
where cf and cm are, respectively, volume fractions of fiber and
(18)
matrix complying to cf þ cm ¼ 1. A lack of superscript refers to
overall composite quantities. It is required that the fibers are
aligned with the x1-axis prior to deformation. Additionally, the 3.3 Constituent Behavior. The behavior of the matrix con-
constitutive model is based on that the constituents transmit iden- stituent is described using a J2-deformation theory introduced by
tical traction normal to the fibers and together with overall equi- St€oren and Rice [22]. The tensor of instantaneous moduli, Lijkl, is
librium leads to given by

Journal of Applied Mechanics JUNE 2018, Vol. 85 / 061005-5

100 Chapter 8 - P3

 
ÿ
2
Lijkl ¼ Gs dik djl þ dil djk þ K ÿ Gs dij dkl
3
4
ÿ ðGs ÿ Gt Þmij mkl (19)
3

where mij are

1
mij ¼ sij (20)
2 req

Here, req is the equivalent von Mises stress and sij are components
of the deviatoric stress tensor Fig. 8 Kink band geometry with two coordinate systems, one
inside the band and one outside represented with (•)i and (•)o
pffiffiffiffiffiffiffiffi 1 rkk respectively
req ¼ 3 J2 ; J2 ¼ sij sij ; sij ¼ rij ÿ dij (21)
2 3

The bulk modulus K, the shear secant modulus Gs, and the tangent composite using the constitutive model from Christoffersen and
shear modulus Gt presented in Eq. (19) are calculated based on Jensen [17]. The semi-analytical kink band model was further
 ÿ1  ÿ1 developed in Jensen [23] to investigate steady-state kink band
3 1 ÿ 2 3 1 ÿ 2 broadening. The formulation of the kink band model will be
Gs ¼ ÿ Gt ¼ ÿ
Es E Et E repeated here for completeness.
(22)
E It is assumed that a kink band has formed where the fibers
K¼ inside the band have rotated relative to the fibers outside the band.
3ð1 ÿ 2 Þ
The material inside and outside the kink band is represented with
where  is Poisson’s ratio and E is Young’s modulus. The tangent two different coordinate systems as shown in Fig. 8, where ð•Þi
modulus Et and the secant modulus Es are calculated using the and ð•Þo represent quantities associated with the material inside
Ramberg–Osgood relation and outside the band, respectively. The stress state outside and
inside the band is updated according to
  !ÿ1
3n req nÿ1
Et ¼ E 1 þ r_ ab ¼ Cabcd vd;c ÿ rab vc;c þ va;c rcb (27)
7 ry
  !ÿ1 (23) which corresponds to Eq. (12). Here, r_ ij or vi;j or a combination of
3 req nÿ1 these can be given by external loading. The band orientation is
Es ¼ E 1 þ
7 ry updated according to

where ry is a reference yield stress. The material parameters used for b_ ¼ ÿva;b na tb (28)
the matrix constituent in the constitutive model are identical with
parameters used in the FE-model. Additionally, a restriction is set on Here, na and ta are the unit normal and unit tangent to the band,
the Ramberg–Osgood curve, so the tangent modulus does not respectively,
become smaller than a specified value Et;min . The limit stress, rL,
and strain, eL , where the minimum tangent modulus is reached, are ½no1 ; no2 Š ¼ ½cosðbo Þ; sinðbo ފ; ½to1 ; to2 Š ¼ ½ÿsinðbo Þ; cosðbo ފ;
½ni1 ; ni2 Š ¼ ½cosðbi Þ; sinðbi ފ; ½ti1 ; ti2 Š ¼ ½ÿsinðbi Þ; cosðbi ފ
rL 3 ry rL n
 
eL ¼ þ (29)
E 7 E ry
1 (24)
The band orientations bo and bi are displayed in Fig. 8. The band
   nÿ1
E 7
rL ¼ ry ÿ1
Et;min 3n orientation inside the band can also be written as bi ¼ bo ÿ /.
The fiber volume fractions are updated according to
For an equivalent uni-axial stress req < rL , the moduli given by
Eq. (23) are to be used and

;o
 cf ;i cf ;o
c_ f ;o ¼ cf ;o cm;o vf2;2 ÿ vm;o
2;2 ; nia eia ¼ noa eoa f ;o ; (30)
Af ;i A
Et ¼ Et;min ;
req where ea are components of an unit vector parallel with the fibers.
Es ¼ ; Af ;i and Af ;o are areas of fibers inside and outside the band,
eeq (25)
req ÿ rL respectively.
eeq ¼ þ eL _ is a
In the current simulation, the fiber rotation increment, /,
Et;min prescribed quantity. Continuity of tractions and displacements is
required during the increment /. _ The fibers are aligned with the
are to be used for req > rL . The fiber constituent is assumed to
behave linear elastic and can therefore be described using the gen- x1-axis initially and during rotation which leads to
eralized Hooke’s law
vo2;1 ¼ 0; vi2;1 ¼ 0 (31)
 
E 1ÿ

Lijkl ¼ dik djl þ dil djk þ dij dkl (26) Continuity of displacements across the boundary requires
1þ 2 1 ÿ 2
wia;b tia tib ¼ voa;b toa tob ; wia;b nia tib ¼ voa;b noa tob (32)
3.4 Kink Band Propagation. In Jensen and Christoffersen
[4], the authors created a semi-analytical kink band model to where wia;b is the velocity gradients seen from a system that stays
investigate the critical compressive stress of fiber-reinforced stationary during deformation

061005-6 / Vol. 85, JUNE 2018 Transactions of the ASME

8.5 Article P3 101

wia;b ¼ via;b þ Xab ; _

X21 ¼ ÿX12 ¼ /; X11 ¼ X22 ¼ 0 (33)

Continuity of traction rates across the kink band boundary

requires that

Ciabcd wid;c nia nib ¼ Coabcd vod;c noa nob ;

(34)
Ciabcd wid;c nia tib ¼ Coabcd vod;c noa tob

Equations (31) and (32) together with the latter equation yield six
boundary conditions. The last two boundary conditions used in
the simulation are

vo1;2 ¼ 0; r_ o22 ¼ 0 (35)

which implicitly lead to the boundary condition r_ o12 ¼ 0.

After the initiation of the kink band, the fibers locks-up in a
given angle and the easiest way of further deformation is by Fig. 9 External WE and internal work WI per unit volume as a
steady-state broadening of the developed kink band, which is seen function of fiber rotation / for the semi-analytical model. The
band orientation is set to b 5 20 deg in the current simulation.
in Fig. 3, stages (4) and (5). When the kink band broadens,
The lock-up condition, W I 5 W E , is marked in the figure.
the end shortening is increased as well. The work done per unit
volume by external forces during propagation is

W E ¼ Soab DEab (36)

Continuity of stresses across the band requires that
where DEab
denotes the difference in Lagrangian strain inside and riab nia nib ¼ roab noa nob ; riab nia tib ¼ roab noa tob (40)
outside the kink band evaluated in the base material. The stresses
Soab are the work conjugate of the evaluated strains. It is assumed Assuming that only ro11 is present in the base material, the latter
that the problem is primarily mechanical and thereby the steady- equations can be written explicitly as
state can be found by a balance of mechanical work. In addition,
it is assumed that the final stress state can be described using a ri11 cos ðbo ÿ /Þ2 þ 2ri12 cosðbo ÿ /Þsinðbo ÿ /Þ
deformation theory. Based on the assumptions, the work done per
unit volume in the kink band is þri22 sin ðbo ÿ /Þ2 ¼ ro11 cos ðbo Þ2
ð ei ÿri11 cosðbo ÿ /Þsinðbo ÿ /Þ þ ri12 ðcos ðbo ÿ /Þ2 ÿ sin ðbo ÿ /Þ2 Þ
WI ¼ rab deab (37)
eo þri22 cosðbo ÿ /Þsinðbo ÿ /Þ ¼ ÿro11 cosðbo Þsinðbo Þ (41)
o i
where e and e are strain states outside and inside the kink band, Using the latter equations, the stresses ri11 and ro11 can be written
respectively. The external work given by Eq. (36) and the internal in terms of the other quantities
work by Eq. (37) are evaluated for every increment /_ and when
the equality ri11 ¼ ÿri12 ½cot/ þ tanðbo ÿ /ފ ÿ ri22 cot/ tanðbo ÿ /Þ;
(42)
WE ¼ WI (38) ro11 ¼ ÿri12 ½cot/ þ tan bo Š ÿ ri22 ½cot/ tan bo ÿ 1Š

is satisfied, the propagation state is obtained. The evaluation of The work per unit volume done by external forces can thereby be
the internal and external work is shown in Fig. 9 as a function of written in terms of ri12 and ri22 as
fiber rotation, /. In the example shown, the band orientation
is fixed in an angle bo ¼ 20 deg. A steady-state lock-up angle
/ss ¼ 41 deg is achieved. The current formulation includes finite
extension of the fibers and volumetric changes of the composite,
which are in contrast with many of the previous studies of kink
band propagation. No prior restrictions are set regarding, e.g., the
relation between the fiber rotation, /, and the band orientation, b.

4 Analytical Model
An analytical model is proposed to estimate the steady-state
propagation parameters. The model is based on some of the same
assumptions made for the model presented in Sec. 3. In the current
section, the fibers are assumed to be inextensible and incompressi-
ble, which are distinct from the assumptions for the two previous
models. In Fig. 10, an illustration of a kink band that broadens
with rigid fibers is shown. During a broadening increment, a mate-
rial piece with length, l, rotates around a point, A. This leads to an
end shortening of u ¼ lð1 ÿ cos /Þ. The work done per unit vol-
ume by the external forces is
Fig. 10 Illustration of steady-state kink band propagation in
the case of inextensible fibers. The geometry with dashed lines
W E ¼ ÿro11 ð1 ÿ cos /Þ (39) represents a propagated configuration.

Journal of Applied Mechanics JUNE 2018, Vol. 85 / 061005-7

102 Chapter 8 - P3

W E ¼ ri12 ½cot/ þ tan bo Šð1 ÿ cos /Þ

þ ri22 ½cot/ tan bo ÿ 1Šð1 ÿ cos /Þ (43)

The work per unit volume done by the stresses in the kink band
is in the general case given by Eq. (37). It is assumed that only
ro11 is present in the base material, and when the fibers are rigid,
no straining and thereby no work is done by the base material.
This leads to the simplification Fig. 11 Strain state outside (state 1) and inside the kink band
ð ei (state 2). The strain state inside the kink band is rotated 60deg
I on the configuration to the right.
W ¼ rab deab (44)
0

Further, the fibers are assumed rigid within the kink band and e12 e22
does no work. The inextensible fibers lead to vf1;1 ¼ vm
1;1 ¼ 0. The
em
12 ¼ em
22 ¼ (52)
internal work can be simplified to cm cm
ðe ðe In the general case, the determination of the stress state depends
W I ¼ 2cm rm m
12 de12 þ c
m
rm m
22 de22 (45) on the total strain field and the strain history. For simplicity, it is
0 0 assumed that the shear stress only depends on the shear strain
rm m
12 ðe12 Þ and the transverse stress only depends on the transverse
Here, the superscript ð•Þi is omitted since only stresses and strains strain rm m
22 ðe22 Þ, which is motivated by the findings from the FE-
inside the band are included in the equation. When regarding rigid model. This is in general not true, but as shown in Sec. 5, this
fibers, continuity of displacements given by Eq. (32) can be sim- assumption leads to promising results. It is assumed that the trans-
plified to verse response is linear elastic, then the stress rm22 can be obtained
using Eq. (26)
wia;b tia tib ¼ 0; wia;b nia tib ¼ 0 (46)
Em ð 1 ÿ  m Þ
Expanding the repeated indices and utilizing vi2;1 ¼ 0 and vi1;1 ¼0 rm
22 ¼ em (53)
leads to ð1 þ  m Þð1 ÿ 2 m Þ 22

vi1;2 sinðbo ÿ /Þcosðbo ÿ /Þ þ vi2;2 cos ðbo ÿ /Þ2 ¼ 0; Plane strain conditions, em m
33 ¼ 0, and inextensible fibers, e11 ¼ 0,
(47) are assumed in the calculations. The shear stress, rm
vi1;2 cos ðbo ÿ /Þ2 þ vi2;2 cosðbo ÿ /Þsinðbo ÿ /Þ ¼ /_ 12 , is assumed
to follow a bilinear curve according to
and solving for vi1;2 and vi2;2 yields
m;y
rm m
12 ¼ 2 G1 e12 for em
12 < e12
_
vi1;2 ¼ /; vi2;2 ¼ /_ tanðbo ÿ /Þ (48) (54)
 
m;y G 2
rm
12 ¼ 2 G2 em
12 þ r12 1ÿ for em;y m
12 < e12
The same kinematics was obtained by Budiansky and Fleck [24] G1
when the shear strain rate in the base material is zero. The strain
rates are then where G1 and G2 are the shear moduli before and after yielding,
respectively, G ¼ E=ð2ð1 þ ÞÞ. The relation between the shear
/_ yield stress, rm;y
12 , and the uni-axial yield stress r
y;m
is
e_ i12 ¼ ; e_ i22 ¼ /_ tanðbo ÿ /Þ (49)
2 rffiffiffi
m;y 1 y;m
Integrating the latter equation with respect to /, and assuming a r12 ¼ r (55)
3
fixed band orientation bo, yields the total logarithmic strain in the
kink band where a von Mises yield criterion is assumed. The uni-axial
stress–strain relationship for the matrix material in the FE-model
cosðbo ÿ /Þ
 
/ and the semi-analytical model follow a Ramberg–Osgood relation.
ei12 ¼ ; ei22 ¼ ln (50)
2 cos bo For comparison with the other models, the first modulus is set to
E1 ¼ Em and the second modulus is E2 ¼ Em t;min . The yield stress
Figure 11 gives an illustrative example of a material point that used in the analytical model is based on
propagates from a stage outside the band and into the band. On
the configuration to the right, the stage inside the band is rotated rL ÿ E2 eL
ry;m ¼ E1 (56)
to an amount of / to show the total deformation. In the illustra- E1 ÿ E2
tion, where / ¼ 60 deg and b ¼ 20 deg, the composite will be
compressed going from stage (1) to (2). When / ¼ 2b, no volu- Here, rL and eL are calculated using Eq. (24). The quantities are
metric straining will be present. / ¼ 2b is the lock-up condition displayed in Fig. 12.
used in many of the previous studies of kink-band propagation, The traction continuity required in the constitutive model
while the present study uses W I ¼ W E . entailed Eq. (18). In line with the constitutive model, rm
22 ¼ r22
The transverse strain and the shear strain of the composite are and rm12 ¼ r12 are viable assumptions. Equations (53) and (54)
assumed to be an average of the constituents together with Eq. (43) yield

e12 ¼ cm em f f
12 þ c e12 ; e22 ¼ cm em f f
22 þ c e22 (51) W E ¼ ð2 C2 / þ C3 Þ½cot/ þ tan bo Šð1 ÿ cos /Þ
cosðbo ÿ /Þ
 
Based on the assumption of rigid fibers (i.e., ef12 ¼ ef22 ¼ 0), the þ 2 C4 ln ½cot/ tan bo ÿ 1Šð1 ÿ cos /Þ
cos b
matrix strain can be determined based on the overall strain accord-
ing to (57)

061005-8 / Vol. 85, JUNE 2018 Transactions of the ASME

8.5 Article P3 103

Fig. 12 Stress–strain relation for the semi-analytical and ana-

lytical models. The bilinear curve represents the response used
for the analytical model.

where the constants are introduced

Fig. 13 External, WE, and internal work WI done per unit
  volume as a function of the fiber rotation / for the analytical
G2 G2
C2 ¼ m ; C3 ¼ rm;y
12 1ÿ ; model. The band orientation is set to b 5 20 deg. The lock-up
2c G1 condition, W I 5 W E , is marked in the figure.
(58)
Em ð1 ÿ  m Þ
C4 ¼ m
2 c ð1 þ  m Þð1 ÿ 2 m Þ 5 Results
The three models presented in Secs. 2–4 will be used to esti-
The work done per unit volume by internal forces can be rewritten mate the steady-state kink band propagation stress rss 11 and lock-
using the stresses rm m
12 and r22 given by Eqs. (53) and (54) together up angle /ss . The band orientation b is regarded as a fixed value
with the strain increments dem m
12 and de22 given by Eq. (49) in all of the three models. In several of the previous studies, the
lock-up criteria / ¼ 2 b were used in the calculations of the
ð /y ð/   steady-state stress based on experimental observations of /  2 b.
G1 G2 G2 Vogler and Kyriakides [5] observed fiber rotations of / > 2 b
WI ¼ / d/ þ / þ ry;m
12 1ÿ d/
0 cm /ycm G1 under stress, and when the stress was removed, the fibers relaxed
ð/
Em ð1 ÿ  m Þ

cosðbo ÿ /Þ
 to /  2 b. Also, early experimental studies of kink band broad-
þ m ð1 þ  m Þð1 ÿ 2 m Þ
ln tanðbo ÿ /Þ d/ ening conducted by Evans and Adler showed / > 2 b. In Fig. 14
0 c cos b is the lock-up angle /ss shown as a function of the band orienta-
(59) tion b0.
Results obtained using the three models are displayed in the fig-
where /y is the fiber angle when matrix yielding happens and is ure, and all of the models predict a larger fiber rotation than 2b.
given by The conventional lock-up angle / ¼ 2 b is also displayed in the
figure for comparison. The predictions of the fiber angles go
2 cm ry;m ð1 þ  m Þ asymptotically toward the conventional lock-up angle for large
/y ¼ pffiffiffi (60) band orientations b. The steady-state broadening stress, ÿrss 11 =G,
Em 3 as a function of band orientation, b, shown in Fig. 15, predicts
that the propagation stress lowers with increasing band orienta-
Solving the integrals in Eq. (59) yields tions. Jensen [13] showed that the propagation stress rss 11 is
  sensitive to the minimum tangent modulus Em t;min , which was intro-
2 2
W I ¼ C1 ð/y Þ þ C2 /2 ÿ ð/y Þ þ C3 ð/ ÿ /y Þ m m
duced in Sec. 3.3. The modulus Et;min =E ¼ 1=100 is used for the
2
cosðbo ÿ /Þ
 
þ C4 ln (61)
cos b

where the constant has been introduced

G1
C1 ¼ (62)
2 cm

The two equations derived for the external and internal work per
unit volume, WE and WI, given by Eqs. (57) and (61) are analytical
expressions, where the only variable is the fiber rotation /. An
evaluation of equations is shown in Fig. 13, where a fixed band
orientation bo ¼ 20 deg is chosen. By comparing Figs. 9 and 13,
good agreement is seen in the approximation of the external work,
the internal work, and the intersection point between the curves. It
does not seem possible to solve the transcendental equation
W I ð/Þ ¼ W E ð/Þ for the steady-state lock-up angle /ss analyti-
cally based on the current formulation. The steady-state fiber Fig. 14 Steady-state lock-up angle /ss as a function of initial
angle /ss can be found either by using a nonlinear solver or by kink band orientation b0. Results are shown for the three mod-
increasing / and evaluating W I ð/Þ ÿ W E ð/Þ a sufficient number els presented. Additionally, the conventional assumption of
of times until W I ð/Þ ÿ W E ð/Þ  0. fiber lock-up / 5 2b is included for comparison.

Journal of Applied Mechanics JUNE 2018, Vol. 85 / 061005-9

104 Chapter 8 - P3

Fig. 15 Normalized steady-state kink band propagation stress Fig. 17 Normalized steady-state kink band propagation stress
2rss
11 /G as a function of initial kink band orientation b0. Results
2rss
11 /G as a function of Poisson’s ratio for the matrix constitu-
are shown for the three models presented. ent m m. The initial band orientation is b0 5 10 deg. Results are
shown for the three models presented.

semi-analytical and analytical models. Beyond that, the material

parameters given by Eq. (4) are used. Finally, the propagation stress as a function of the minimum
High correlation is seen between the three models despite tangent modulus Em m
t;min =E is illustrated in Fig. 18 for the most
the significant difference in model setup. The relatively high fiber critical band orientation, b0. When the kink band is fully formed
stiffness Ef =Em ¼ 35 is used for the FE-model, and semi- and locked-up, the shear strain within the matrix is in the order
analytical model and inextensible fibers are assumed for the ana- em
12 ¼ 20% ÿ 110% depending on the initial band orientation
lytical model. This is one of the explanations for the deviations b0. For such high strains, the Ramberg–Osgood curve predicts
seen in Fig. 15 between the models for small band orientations. unrealistic low tangent moduli. The tangent modulus is restricted
It was only possible to obtain steady-state results between b0 ¼ with the results shown in Fig. 18 following the method introduced
0 deg ÿ 15 deg using the FE-model due to convergence problems in Sec. 3.3. The Ramberg–Osgood curves obtained using Em t;min ¼
at higher kink band orientations. 0:02 ÿ 0:1 all agree with the measurements of Kyriakides et al.
Figure 16 indicates that the lock-up angle /ss is sensitive to [16] for small strains. The broadening stress is seen to be sensitive
Poisson’s ratio of the matrix material  m. The figure shows to the minimum tangent modulus, which is in line with the results
the results obtained for b0 ¼ 10 deg varying from  m ¼ 0 to shown in Jensen [13]. It was not possible to make a restriction on
 m ¼ 0:5. In the limit of  m ¼ 0:5, the matrix constituent becomes the minimum tangent modulus using the build in J2-deformation
incompressible. All the models tend toward / ¼ 2 b when the theory in ABAQUS, and results from the FE-model are, therefore,
matrix constituent becomes incompressible, which agrees with not shown in Fig. 18. A user material could have been used
assumptions made by previous authors. together with ABAQUS to restrict the tangent modulus Em t , but this
Figure 17 argues that also the propagation stress rss11 is sensitive has not been in the scope. The analytical model predicts critical
to Poisson’s ratio of the matrix. The highest stress is obtained for band orientations in the range b0 ¼ 28 deg ÿ 32 deg, where the
a nearly incompressible matrix. Previous authors have used the semi-analytical model predicts orientations in the range
assumption of an incompressible composite (i.e.,  m ¼ 0:5). b0 ¼ 20 deg ÿ 26 deg. The higher band orientations, b0, predicted
Based on the current results, the propagation stresses predicted by by the analytical model result in lower steady-state stresses which
previous authors are nonconservative for  m < 0:5. The broaden- is in line with the results shown in Fig. 15.
ing stress ÿrss m
11 =G ¼ 0:1 for  ¼ 0:35 is predicted using the FE- Evans and Adler [6] observed several kink bands with orienta-
model, which is 80% of the stress predicted using  m  0:5. tions in the range b ¼ 20 deg ÿ 25 deg and fiber rotation between
/ ¼ 45 deg and 60 deg for a carbon/carbon composite, which is in

Fig. 16 Steady-state lock-up angle /ss as a function of Pois- Fig. 18 Normalized steady-state kink band propagation stress
son’s ratio for the matrix constituent m m. The initial band orien- 2rss
11 /G as a function of normalized minimum tangent modulus
m
tation is b0 5 10 deg. Results are shown for the three models for the matrix constituent Et;min /E m . Results are shown for the
presented. semi-analytical and analytical models.

061005-10 / Vol. 85, JUNE 2018 Transactions of the ASME

8.5 Article P3 105

the range predicted by the three models. Moran et al. [7] observed References
b ¼ 20 deg ÿ 25 deg and / ¼ 40 deg ÿ 45 deg for a IM7/PEEK [1] Fleck, N., 1997, “Compressive Failure of Fibre Composites,” Adv. Appl.
composite. They observed a propagation stress in the range Mech., 33, pp. 43–117.
ÿrss11 =G ¼ 0:052 ÿ 0:086 which corresponds to a band inclination
[2] Bishara, M., Vogler, M., and Rolfes, R., 2017, “Revealing Complex Aspects of
Compressive Failure of Polymer Composites—Part II: Failure Interactions in
b ¼ 20 deg ÿ 30 deg according to the predictions in Fig. 15 using Multidirectional Laminates and Validation,” Compos. Struct., 169, pp.
the proposed models. A recent experimental study of kink band 116–128.
propagation was conducted by Nizolek et al. [12] using a Cu/Nb [3] Budiansky, B., 1983, “Micromechanics,” Comput. Struct., 16(1–4), pp. 3–12.
nanolaminate composite with a band orientation b  27 deg and a [4] Jensen, H. M., and Christoffersen, J., 1997, “Kink Band Formation in Fiber
Reinforced Materials,” J. Mech. Phys. Solids, 45(7), pp. 1121–1136.
fiber rotation of /  60 deg. The relation of / > 2b predicted by [5] Vogler, T., and Kyriakides, S., 1997, “Initiation and Axial Propagation of Kink
the three models correlates with their findings. It has not been pos- Bands in Fiber Composites,” Acta Mater., 45(6), pp. 2443–2454.
sible to compare the steady-state propagation stress with the study [6] Evans, A. G., and Adler, W. F., 1978, “Kinking as a Mode of Structural Degra-
done by Nizolek et al. [12] since the necessary material data for dation in Carbon Fiber Composites,” Acta Metall., 26(5), pp. 725–738.
[7] Moran, P. M., Liu, X. H., and Shih, C. F., 1995, “Kink Band Formation and
the current models were not present in the paper. Band Broadening in Fiber Composites Under Compressive Loading,” Acta
Metall. Mater., 43(8), pp. 2943–2958.
[8] Vogler, T. J., and Kyriakides, S., 1999, “On the Axial Propagation of Kink
6 Conclusions Bands in Fiber Composites—Part I: Experiments,” Int. J. Solids Struct., 36(4),
pp. 557–574.
Steady-state kink band propagation in layered materials or [9] Hsu, S.-Y., Vogler, T. J., and Kyriakides, S., 1999, “On the Axial Propagation
fiber-reinforced composites has been analyzed. Three substan- of Kink Bands in Fiber Composites—Part II: Analysis,” Int. J. Solids Struct.,
36(4), pp. 575–595.
tially different models have been used to investigate the phenom- [10] Attwood, J. P., Fleck, N. A., Wadley, H. N. G., and Deshpande, V. S., 2015,
enon and two of the models have been developed during this “The Compressive Response of Ultra-High Molecular Weight Polyethylene
study. A two-dimensional finite element model has been devel- Fibres and Composites,” Int. J. Solids Struct., 71, pp. 141–155.
oped that includes inclined periodic boundary conditions to inves- [11] Byskov, E., Christoffersen, J., Dencker Christensen, C., and Sand Poulsen, J.,
2002, “Kinkband Formation in Wood and Fiber Compositesmorphology and
tigate a propagating infinite kink band. Steady-state was obtained Analysis,” Int. J. Solids Struct., 39(13–14), pp. 3649–3673.
using the FE-model, when the fiber rotation inside the kink band [12] Nizolek, T., Begley, M., McCabe, R., Avallone, J., Mara, N., Beyerlein, I., and
stopped rotating and overall band broadening was observed Pollock, T., 2017, “Strain Fields Induced by Kink Band Propagation in Cu-Nb
instead. An analytical model resulting in one transcendental equa- Nanolaminate Composites,” Acta Mater., 133, pp. 303–315.
[13] Jensen, H. M., 1999, “Analysis of Compressive Failure of Layered Materials by
tion for the propagation state has been proposed. The analytical Kink Band Broadening,” Int. J. Solids Struct., 36(23), pp. 3427–3441.
model is based on the assumptions of inextensible fibers and a [14] Prabhakar, P., and Waas, A. M., 2013, “Interaction Between Kinking and Split-
simple stress–strain relation inspired by observations from the FE- ting in the Compressive Failure of Unidirectional Fiber Reinforced Laminated
model. In contrast to models proposed by several previous Composites,” Compos. Struct., 98, pp. 85–92.
[15] Skovsgaard, S. P. H., and Jensen, H. M., 2018, “Constitutive Model for
authors, no assumptions regarding the fiber lock-up kinematics Imperfectly Bonded Fibre-Reinforced Composites,” Compos. Struct., 192, pp.
have been utilized. The condition for fiber lock-up is based on a 82–92.
balance of external and internal work in line with the methodol- [16] Kyriakides, S., Arseculeratne, R., Perry, E. J., and Liechti, K. M., 1995, “On
ogy proposed by Jensen [13]. The two introduced models are used the Compressive Failure of Fiber Reinforced Composites,” Int. J. Solids Struct.,
32(6–7), pp. 689–738.
together with the constitutive model introduced by Jensen [13]. [17] Christoffersen, J., and Jensen, H. M., 1996, “Kink Band Analysis Accounting
The three models have been used to investigate the steady-state for the Microstructure of Fiber Reinforced Materials,” Mech. Mater., 24(4), pp.
lock-up angle and the propagation stress based on different param- 305–315.
eters. Initially, the influence of the band orientation, b0, has been [18] Sørensen, K. D., Mikkelsen, L. P., and Jensen, H. M., 2009, “User Subroutine
for Compressive Failure of Composites,” Simulia Customer Conference, Lon-
investigated, which is a prescribed quantity using the proposed don, May 18–21, pp. 618–632.
models. The rotation of the fibers at lock-up exceeds the linear [19] Davidson, P., and Waas, A. M., 2014, “Mechanics of Kinking in Fiber-
relation assumed by previous authors and the current findings cor- Reinforced Composites Under Compressive Loading,” Math. Mech. Solids,
relate with experimental observations. It can be concluded that 21(6), pp. 667–684.
[20] Wind, J. L., Steffensen, S., and Jensen, H. M., 2014, “Comparison of a Com-
including compressibility of the matrix constituent has a large posite Model and an Individually Fiber and Matrix Discretized Model for Kink
influence on the lock-up angle and the propagation stress. In line Band Formation,” Int. J. Non-Linear Mech., 67, pp. 319–325.
with the findings by Jensen [13], the matrix properties at large [21] Fleck, N. A., and Budiansky, B., 1991, “Compressive Failure of Fibre Compo-
strains have a large influence of the propagation stress especially sites Due to Microbuckling,” Third Symposium on Inelastic Deformation of
Composite Materials, Troy, NY, May 29–June 1, pp. 235–273.
the tangents modulus and Poisson’s ratio. [22] St€oren, S., and Rice, J. R., 1975, “Localized Necking in Thin Sheets,” J. Mech.
Phys. Solids, 23(6), pp. 421–441.
[23] Jensen, H. M., 1999, “Models of Failure in Compression of Layered Materials,”
Funding Data Mech. Mater., 31(9), pp. 553–564.
[24] Budiansky, B., and Fleck, N., 1993, “Compressive Failure of Fibre
 Thomas B. Thriges fund. Composites,” J. Mech. Phys. Solids, 41(1), pp. 183–211.

Journal of Applied Mechanics JUNE 2018, Vol. 85 / 061005-11

Chapter 9

P4
A general approach for the study of kink band broadening in
fibre composites and layered materials

9.1 Motivation
Kink band formation is the dominant compressive failure mechanism for many fibre-
reinforced plastics. The stress at steady-state kink band broadening can be seen as a lower
bound for the compressive strength in composites. Three substantially different models
were developed in Skovsgaard and Jensen (2018c) to study kink band broadening. In the
latter article, the emphasis was given to a carbon fibre-reinforced thermoplastic. Kink band
broadening is observed in different composites at different length scales e.g. Nizolek et al.
(2017) observed broadening in a Cu/Nb nanolaminate and Attwood et al. (2015) observed
kink band broadening, that was developed between plies within a composite made from
ultra-high molecular weight polyethylene.
In the current publication, the three models developed in [P3] are used within two
case studies with different composite material behaviour to study steady-state kink band
broadening and to compare with experimental results. The original source of the publication
is: (Skovsgaard and Jensen, 2019).

9.2 Method
The unit cell finite element model, the semi-analytical model and the analytical model are
used, that are described in Chapter 2, 3 and 4, respectively.
108 Chapter 9 - P4

The finite element model consists of three layers: one matrix layer and two half fibre
layers; or one elastic-plastic layer and two half elastic layers. Periodic boundary conditions
are applied on a skewed mesh to get a band inclination β . An arc-length solver is used to
solve the non-linear problem. The semi-analytical model is based on a kink band model
developed by Jensen and Christoffersen (1997) with a homogenised constitutive formulation
from Christoffersen and Jensen (1996). The analytical model, as described in Chapter 4,
results in one transcendental equation for the steady-state propagation state and is derived by
evaluating the internal and external work at the steady-state.

9.3 Contribution
The contributions to the publication comprise: the use of the three models in two case studies;
changing the constitutive relations for the semi-analytical model to be used on the ply-level;
conducting the simulations and writing the paper.

9.4 Main findings

The three models, initially introduced in [P3], were in the current publication reformulated
and presented in a format that emphasises that the models can, equivalently, be used for
layered materials and for fibre composites. Furthermore, the results indicate that the models
can be used across different length scales to study kink band broadening.
In the first case study, a carbon fibre-reinforced PEEK composite is used to study the effect
of fibre diameters upon kink band broadening using the unit cell FE-model. The predictions
for the steady-state propagation stress, using the FE-model, approach the prediction by the
semi-analytical model with decreasing fibre diameters. The semi-analytical model does not
include intrinsic size effects.
In the second case study, the semi-analytical model and the analytical model are used
to study kink band broadening for a composite made from ultra-high molecular weight
polyethylene. The two models predict steady-state quantities that are in good agreement with
the experimental observations done by Attwood et al. (2015) and Liu et al. (2014).
9.5 Article P4 109

European Journal of Mechanics / A Solids 74 (2019) 394–402

Contents lists available at ScienceDirect

European Journal of Mechanics / A Solids

journal homepage: www.elsevier.com/locate/ejmsol

A general approach for the study of kink band broadening in fibre T

composites and layered materials☆
Simon P.H. Skovsgaard∗, Henrik Myhre Jensen
Department of Engineering, Aarhus University, Inge Lehmanns Gade 10, 8000, Aarhus C, Denmark

ARTICLE INFO ABSTRACT

Keywords: The progressive non-linear mode of deformation known as steady-state kink band broadening is analysed for
Kink band broadening fibre composites and layered materials. The mode of deformation is investigated using an analytical, a semi-
Maxwell construction analytical model and a finite-element model. The semi-analytical model is based on a constitutive model, where
Propagating instability independent material behaviour can be given for two constituents. The analytical model assumes rigid fibres and
Strain localisation
results in a transcendental equation for the kink band broadening state. Both the analytical and semi-analytical
Size effects
model use a Maxwell construction to determine the steady-state, which is done by equating the internal and
external work. The influence of size-effects are explored and two case studies are performed; in the first case
study the finite element model and semi-analytical model are used upon a carbon fibre-reinforced PEEK com-
posite. The second case study is on a layered composite made from ultra high molecular-weight polyethylene
fibres with a kink band developing on ply level.

1. Introduction misalignments interacting with matrix plasticity as described in Fleck

and Budiansky (1991) and Fleck et al. (1995).
Laminates built from unidirectional fibre-reinforced composites will Steady-state kink band broadening is a propagating instability on
have an anisotropic material behaviour, where the amount of aniso- the material scale and can be put into the same category as neck pro-
tropy mainly depends on the difference in fibre/matrix material prop- pagation in polymers during stretching e.g. (1983) G'Sell (1983) and
erties and the fibre to matrix bonding. There can be a large difference in channelling cracks in layered materials e.g. Beuth (1992). The stress at
the tension, shear and transverse moduli in fibre-reinforced composites band broadening can be seen as the lower bound of the compressive
and together with the heterogeneities, this creates many possible ma- strength in fibre composites and is independent of fibre misalignments.
terial failures. Fibre-reinforced composites have high stiffness and The broadening stress can thus be used as a fail-safe compressive
strength in tension but can become unstable in compression and the strength for structural applications and it does only depend on the
compressive strength is often only 50–60% of their tensile strength. A material behaviour. Kink band broadening has been observed in dif-
frequently observed failure is kink band formation, where a band of ferent composites. In the early work of Evans and Adler (1978), they
material has rotated relative to the surrounding base material. The observed band broadening in a carbon fibre composite. Liu et al. (1996)
formation of kink bands has been observed in several experiments with observed band broadening in a IM7/PEEK carbon fibre composites,
compressive loading in the direction of the fibres including Kyriakides where the fibres stayed intact during broadening. Kink band broad-
et al. (1995) and Piggott and Harris (1980) and others. ening and transverse kink propagation was studied by Budiansky et al.
As described by Liu et al. (1996), the formation of a kink band can (1998). Later, Byskov et al., (2002) performed a compressive study on
be divided into different stages; the linear response, incipient kinking, wood (Norway Spruce) where kink band broadening was observed.
peak stress, transient kinking and steady-state kink band broadening. So Kink band broadening was observed to propagate on ply level in a
far, the main focus has been to estimate the peak compressive stress for composite made of ultra-high molecular-weight polyethylene
design purposes either by the use of experimental, numerical or ana- (UHMWPE) fibres by Attwood et al. (2015). In a recent study conducted
lytical approaches. The compressive strength in unidirectional fibre by Nizolek et al. (2017) they observed band broadening in a CueNb
composites is largely influenced by imperfections in the form of fibre nanolaminate composite.

☆
This article was presented at the IUTAM Symposium on Size-Effects in Microstructure and Damage Evolution at Technical University of Denmark (2018).
∗
Corresponding author.
E-mail addresses: [email protected] (S.P.H. Skovsgaard), [email protected] (H.M. Jensen).

https://doi.org/10.1016/j.euromechsol.2018.12.007
Received 18 September 2018; Received in revised form 18 December 2018; Accepted 21 December 2018
Available online 29 December 2018
0997-7538/ © 2018 Elsevier Masson SAS. All rights reserved.
110 Chapter 9 - P4

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

In the current study, three different models are used to investigate 2.2. Constitutive model
kink band broadening in a general framework, including a finite ele-
ment model, a semi-analytical model and an analytical model. Two case The constitutive model introduced in Christoffersen and Jensen
studies with two different material compositions are presented. In the (1996) homogenises the material properties of two individually de-
first case study, the size effects upon kink band broadening are in- scribed constituents. The model is based on the kinematic assumption
vestigated using the finite element model and the semi-analytical that material lines parallel with the x1 -direction are subjected to a
model. The analytical and semi-analytical models are used in a second common stretching and rotation. The strains in the x2 -direction are
case study using material parameters of an ultra high molecular-weight averages of the two constituent strains. The kinematics can be described
polyethylene fibre composite to predict some kink band broadening using velocity gradients
parameters. The theoretical predictions are compared with experiments
v = v a,1 = v b,1 v = c a v a,2 + c b v b,2 (7)
performed by Attwood et al. (2015) on a composite made from ,1 ,2

UHMWPE fibres. This article is contributing to the further under- where c a and c b = 1 c a are volume fractions of the constituents in the
standing of the compressive mechanics of fibre composites and layered layered material or fibre composite. Superscripts (•)a and (•)b corre-
materials. The analytical and semi-analytical model developed in sponds to values associated with the two arbitrary constituents. A lack
Jensen (1999) and Skovsgaard and Jensen (2018c) are modified to of superscript refers to overall composite quantities. Additionally, the
predict the propagation stress and fibre lock-up angle for a composite model is based on the assumption that planes with outward normal
made from UHMWPE fibres. The modifications lies in the im- parallel with the x2 -axis transmit identical tractions. This assumption
plementation of the constitutive behaviour of the UHMWPE fibres. together with overall equilibrium entails
Furthermore, the fact that band broadening was observed to develop on
ply level is taken into account. In this article it is further emphasized t2 = t2a = t2b t1 = c a t1a + c b t1b (8)
that the models developed in Jensen (1999) and Skovsgaard and Jensen Christoffersen and Jensen (1996) introduced a matrix format to
(2018c) can be used for both layered materials and fibre reinforced describe the relations for convenience. This matrix format is not re-
composites with arbitrary material constituents. peated in the current article, instead the usual index notation is kept.
Using the latter relations and under the assumption of plane strain,
2. Semi-analytical model for kink band initiation and broadening Christoffersen and Jensen (1996) showed that the composite behaviour
could be described according to
The methodology introduced in Jensen (1999) is used to investigate
kink band broadening using the homogenisation methodology devel- C = ca C a + cb C b c a c b (C a 2 C b 2 ) H (C2a C2b ) (9)
oped in Christoffersen and Jensen (1996). Originally, the model was
where,
used to study a fibre reinforced composite with two constituents, a
matrix and fibre composition. The latter model will be known as the H M =
semi-analytical model and is presented with two arbitrary constituents M = c b C2a 2 + c a C2b 2 (10)
with the superscripts (•)a and (•)b . The theory behind the approach will
be outlined here. The moduli M are initially calculated and then the inverse H is
determined so the equality H M = is satisfied. The moduli C c
2.1. General relations for the constituents are initially calculated using Eq. (3) using the
tangent moduli L . The homogenisation methodology with two
c

The strain rate and spin are given by constituents was later extended to include information about imperfect
bonding for a fibre composite in Skovsgaard and Jensen (2018b) and
1 1 was extended to a three-dimensional version in Skovsgaard and Jensen
= (vi, j + vj, i) = (vi, j vj, i )
ij
2
ij
2 (1)
(2018a). The constitutive model introduced by Jensen (1999), as-
where vi, j are the velocity gradients. Latin indices i.e. i , j, k take values suming plane strain conditions, was used in Tojaga et al. (2018) as a
1,2,3, and Greek indices i.e. , , take values 1,2. A comma denotes user subroutine UMAT in Abaqus to investigate the development of
partial derivative. The relation between the rate of nominal stresses tij kink bands in open-hole fibre composites. The deformations for the
and velocity gradients are individual constituents v c, are calculated based on the overall velocity
gradients v , in every increment. Furthermore, the stresses for the
tij = Cijkl vl, k (2) constituents are updated in every increment. Thereby, individual and
where Cijkl are components of nominal moduli independent constitutive relations can be used for the constituents
through the moduli L c with arbitrary elastic-plastic constitutive be-
1 1 1 1 haviour.
Cijkl = Lijkl +
2
kj il
2
lj ik
2
il kj
2
ik lj
(3)

and ij denote Kronecker delta and ij are the current Kirchhoff stresses. 2.3. Kink band propagation
The moduli Lijkl describe the relation between the Jaumann rate of
Kirchhoff stresses and strain rates ij = Lijkl kl and are to be described A semi-analytical kink band model was created in Jensen and
using a time-independent constitutive law e.g. J2-deformation and J2- Christoffersen (1997) using the constitutive formulation. The kink band
flow theory. Moment equilibrium ij = ji and strain symmetry ij = ji model assumes that the kink band formation and propagation can be
require the symmetries simplified into two material states; one inside the localised band and
one outside the band (base material) see Fig. 1. It has previously been
Lijkl = Ljikl = Lijlk (4) shown that the assumptions lead to good agreement with experimental
The relation between Cauchy stresses and Kirchhoff stresses are results (Jensen and Christoffersen 1997 and Skovsgaard and Jensen
(2018a)). The key equations and assumptions are introduced and for
ij =J ij (5) further details see Christoffersen and Jensen (1996) and Jensen (1999).
where J is the Jacobian giving the volume ratio J = V / V0 . The relation Fig. 1 shows the kink band geometry and the local coordinates for the
between the rate of Cauchy stresses and the rate of nominal stresses is material states. Superscripts (•)o and (•)i refer to quantities associated
with the outer base material and the inner band material, respectively.
ij = tij ij vk, k + vi, k kj (6) The angle ϕ gives the relative angle between the orientation of fibres

395
9.5 Article P4 111

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

Fig. 1. Kink band geometry with two coordinate systems, one inside the band and one outside represented with (•)i and (•)o respectively.

inside and outside the band. The angles o and i are the relative band of velocities Eq. (16) and equilibrium of tractions Eq. (18) give the first
orientations seen from the two systems. The outward normal n and four equations. The requirement of Eq. (14) gives further two equa-
tangent t are shown on the band boundary. tions. The last two boundary conditions used in the simulations are
The x1 -coordinate is aligned with the fibres, or layers in the case of a o o
v1,2 =0 22 =0 (19)
layered material, prior to deformation. This applies both for the state
inside and outside the band. A Jaumann frame of reference is used, so which entails no shear straining and transverse stress in the base ma-
the coordinate systems rotate with the deformation during an incre- terial. The kink band model was used to study the effect of varying
ment. The stresses in the two material states, inside and outside the kink initial imperfections 0 on the critical compressive strength 11 cr
in
band, are updated according to Jensen and Christoffersen (1997). The model can be used to study kink
band broadening in composites as well using a Maxwell condition. In
=C v , v , +v , (11)
the case of steady-state broadening, it is assumed that a kink band has
The relative band orientations are updated using formed and locked up in a given orientation ss and with a fibre rotation
ss , where the superscript (•) ss refers to a steady-state quantity. From
= v , n t (12)
this point, the easiest way of further deformation is to broaden the al-
In the case of a material composition with different bulk moduli, ready developed band. In this state the band will broaden (or propa-
then the volume fraction of the constituents are updated according to gate) under a constant applied stress i.e. the work done by the external
a b
stresses and strains on an unit volume equals the work done by an unit
c a = c a c b (v2,2 v2,2 ) (13) volume transitioning from the base state to the band state W E = W I .
Due to the choice of a Jaumann frame of reference, then the co- The external work per unit volume is
ordinate systems rotate with the fibres, leading to WE = So Eo , (20)
o i
v2,1 = 0 v2,1 =0 (14) where E o is the difference in Green-Lagrangian strain inside and
To ensure continuity of tractions and displacements the normal and outside the band expressed in the base material and S o denotes the
tangent to the band are needed and are formulated relative to the in- work conjugate Second Piola-Kirchhoff stress in the base material. The
dividual frames work done per unit volume in the band material is
i
[n1o , n 2o] = [cos( o), sin( o)] [t1o, t2o] = [ sin( o ), cos( o)] WI = d ,
o (21)
= [cos( i ), sin( i )] [t1i, t2i ] = [ sin( i ), cos( i)] (15)
where o and i are strain states outside and inside the band, re-
Continuity of displacements across the band boundary entails the spectivily. The inner and outer work are evaluated in every increment
two equations i.e. for every fibre rotation . In the simulations, the band orientation β
wi , t i t i = v o, t o t o wi , n i t i = v o, n o t o is seen as a fixed value and the fibre lock-up angle is search for where
(16)
the Maxwell condition W E = W I is satisfied. This methodology is in
where, contrast to the approach done by several previous authors, where the
kinematic relation = 2 is used, which have been observed in ex-
wi , = v i , + 21 = 12 = 11 = 22 =0 (17)
periments.
The components wi , are the velocity gradients seen from a system
that stays stationary during deformation i.e. does not follow the fibres 3. Analytical model for kink band broadening
during deformation. The relative spin is included in the spin tensor
ij . Equilibrium of tractions across the band entails In Skovsgaard and Jensen (2018c) an analytical model was devel-
Ci wi, n i n i = C o v o, n o n o C i wi , n i t i = C o v o, n o t o oped to evaluate the external and internal work per unit volume during
(18)
kink band formation. This resulted in a transcendental equation in the
The kink band model is a two dimensional model with plain strain form W E ( ) = W I ( ) , with fibre rotation ϕ being the only unknown.
condition. The deformations in every material state are fully defined by The model is inspired by the semi-analytical model regarding the two
the use of the four components of the velocity gradients. Since the kink material states and the continuity requirements. The semi-analytical
band model only includes two material states, eight components of the model is formulated in a general framework with arbitrary behaviour of
velocity gradients are to be determined in every increment. Continuity the constituents. In contrast, the analytical model makes several prior

396
112 Chapter 9 - P4

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

and transverse straining in the base material require that

wi , t i t i = 0 wi , n i t i = 0 (26)
where wi , is given by Eq. (17). The velocity gradients integrates into
the finite logarithmic strains

i i cos( o )
12 = 22 = ln
2 cos (27)
which was shown by Skovsgaard and Jensen (2018c). The same kine-
matics were obtained by Budiansky and Fleck (1993). It is assumed that
the transverse stresses can be expressed through transverse strains only,
which is a simplification compared to a general elastic-plastic con-
stitutive law. It is assumed that the transverse behaviour is purely
elastic and plane strain conditions prevail, which yields the relation
E (1 )
22 = 22
(1 + )(1 2 ) (28)
Fig. 2. Illustration of steady-state kink band propagation in the case of in- The shear stresses are assumed to dependent only on the shear
extensible fibres. The geometry with dashed lines shows a propagated config- strains, and follow a bilinear elastic-plastic relation
uration.
y
12 = 2 G1 12 for 12 < 12

assumption to the material behaviour e.g. large difference in elastic 12 = 2 G2 12 + y

12 (1 G2
G1 ) for y
12 < 12
(29)
moduli between the constituents, which is the case for many fibre re-
inforced composites. where G1 is the shear modulus prior to yielding and G2 is the shear
The model assumes that one of the constituents can be regarded as modulus beyond yielding, the quantity 12y is the shear strain where
being rigid and incompressible, which is a valid assumption for many yielding takes place.
composites when evaluating the work e.g. carbon and glass fibre By introducing the constants
composites. A case study is presented later using an ultra high mole- G1 G2 G2 E (1 )
y
cular-weight polyethylene fibre composite. It has been observed ex- C1 = C2 = C3 = 12 1 C4 =
2 2 G1 2 (1 + )(1 2 )
perimentally that these UHMWPE fibre composites have a three order
of magnitude difference in moduli in compression and shear (Liu et al. (30)
(2014)). This make them ideal for this analytical model assuming rigid then it was shown by Skovsgaard and Jensen (2018c) that the external
fibres. The kinematic constraint = 2 has not been used in the cur- and internal work done per unit volume can be expressed explicitly as
rent study as in many of the previous studies. Some interesting results
W E = (2 C2 + C3)[cot + tan o ](1 cos )
can be seen when this unnecessary kinematic constraint is relieved.
In Fig. 2 the geometry is seen that describes the analytical kink band
broadening model. At a certain point during localisation the easiest way
+ 2 C4 ln ( cos( o
cos
)
) [cot tan o 1](1 cos )
(31)
of further deformation is through kink band broadening. Assuming rigid 2
cos( o )
fibres and a locked up geometry as seen in the figure, an infinitesimal W I = C1 ( y )2 + C2 ( 2 ( y )2 + C3 y + C4 ln
cos
fibre piece of length l will rotate into the localised band. This will result
in an end shortening of = l (1 cos ) . Since mechanical work is force (32)
times displacement, the external work per unit volume is where the yield rotation,
WE = o
11 (1 cos ) (22) y
y 12
=
where 11 o
is the external applied stress. Moran et al. (1995) came to the G1 (33)
same result for the external work and was presented in a similar is introduced. The condition for steady-state broadening requires the
manner. No straining and thereby no work is done by the base material, work balance condition W E = W I to be satisfied. The two equations, Eq.
due to the assumption of rigid fibres. Furthermore, no shear and (31) and Eq. (32), giving the external and internal work do only depend
transverse stresses are applied to the base material in the current ana- on the rotation ϕ. It does not seem possible to solve the balance
lytical model. The work done per unit volume by the stresses in the kink W E = W I analytically with respect to ϕ and thereby gives a transcen-
band is dental equation for the Maxwell condition of the steady-state kink band
i i broadening.
WI = 2 12 d 12 + 22 d 22 (23)
0 0
4. Size effects upon kink band broadening
Stress equilibrium across the band requires
i i
n n =i o o
n n o i i
n t = i o o o
n t (24) 4.1. Finite element model

where again n and t are the unit normal and tangent to the band plane The influence of size effects upon kink band broadening will be
expressed in either the inner or outer material system. Using the latter investigated using a finite element model and will be compared with
equation together with Eq. (15) yields the results obtained using the semi-analytical kink band model. A case
i
= i o i o study is created for a fibre reinforced PEEK composite. The full elastic-
11 12 [cot + tan( )] 22 cot tan( )
o i i plastic behaviour of the constituents for the later composite was thor-
= 12 [cot + tan o]
22 [cot tan o 1] (25)
oughly investigated in Kyriakides et al. (1995) and is thereby chosen for
11

which express the stresses 11i

and 11o
in terms of the inner stresses 12i
the current investigation. The finite element model was initially pre-
and 22
i
. Continuity of velocities in the case of rigid fibres and no shear sented in Skovsgaard and Jensen (2018c). An illustration of the model

397
9.5 Article P4 113

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

fibre constituent is assumed to have an linear-elastic response. The

PEEK constituent behaves elastic-plastic and is fitted to a Ramberg-
Osgood relationship. The built-in plastic flow rule was chosen in
Abaqus with isotropic hardening with a von-Mises yield criterion. The
material parameters are normalized quantities of the one presented in
Kyriakides et al. (1995). The material parameters and geometries used
in the models are

c f = 0.6 E f / E m = 35 f = 0.263
m = 0.356 y, m/ E m = 0.013 nm = 4
L0 = 1 b0 = 0.5 0 = 1 = 15 (36)

where c f is the fibre volume fraction, E c and c are the elastic para-
Fig. 3. Geometry of the micromechanical finite element model with two con- meters Young's modulus and Poisson's ratio. The plastic parameters y, m
stituents. The darker regions represents the fibre constituent in the current case and nm are the reference yield stress and the hardening exponent used
study. in the Ramberg-Osgood relationship for the PEEK constituent.
The finite element model is substantially different from both the
semi-analytical model and the analytical model. Although the differ-
ences in formation, the models give remarkable similar results as con-
firmed in Skovsgaard and Jensen (2018c). There are many differences
in the numerical foundation. One of the more profound differences is
that a fibre rotation increment is applied in the semi-analytical model
Fig. 4. Illustration of mesh used in the finite element simulations. The coupling and a compressive displacement u is applied in the FE-model.
of the nodes locks the deformation in an orientation β.
4.2. Size effects and validation
geometry and boundary conditions are shown in Fig. 3. The FE-model is
a micromechanical model where the fibre and matrix constituents are The normalized fibre diameter d f / L0 is changed in the current study
modelled individually. The initial length of the geometry is L0 . To si- using the FE-model. Results showing the normalized compressive stress
mulate an infinite extent of the kink band, periodic boundary condi- 11/ G as a function of normalized end-displacement u / L0 are pre-
tions (PBC) are utilized on the top and bottom of the geometry. The sented in Fig. 5. Furthermore, deformed geometries at a normalized
PBCs make it possible to model only one full layer of a constituent and end-displacement u / L0 = 0.04 are shown in Fig. 6. Results from six si-
two half layers. The middle layer will in the current model represent the mulations are presented in the figures with different normalized fibre
PEEK constituent and the two half layers are fibres. The commercial diameters. In the two simulations with large fibre diameters
software Abaqus 6.14 was used for the simulations. An implicit finite (0.1 0.05), here the boundary condition in the left and right together
element solver was chosen together with an arc-length solver since the with a large bending stiffness prevents the geometries from reaching a
equilibrium path is non-linear. A special methodology for applying the steady-state. As the fibre diameter decreases the end effects decrease as
periodic boundary conditions is introduced as can be seen in Fig. 4. In well. As the fibre diameter decrease the compressive stress at steady-
the middle of the geometry the elements are skewed and together with state ss converges towards the same value for the simulations. Fur-
the chose of node coupling, this locks the formation and the kink band thermore, the transition in the geometry going from the base material
broadening at a given angle β. The mentioned periodic boundary con- to the kinked area sharpens with decreasing fibre diameters. A sharp
ditions, where the band is locked during deformation, resemble the transition between the base and kink band material is consistent with
boundary conditions introduced for the analytical and semi-analytical the assumptions used in the analytical and semi-analytical model. The
model. result from a simulation generated with the semi-analytical model (SA -
An imperfection in the form of a waviness was introduced in the model) is shown as a dashed line in Fig. 5. The result is presented as a
geometry initially where the deformed coordinates are given by horizontal line, since it is not possible to extract the compressive stress
as a function of end-displacement from the semi-analatical model. The
x2 =
1
2 0 ( b0
2 cos
X2 tan ) for X1 < X1, min same material behaviour is chosen for the constituents in both the FE-
model and the semi-analytical model.
1 b0 2 cos
x2 = 2 0 2 cos
sin b0
(X1 + X2 tan ) + X1 for X1, min < X1
5. Case study: kink band broadening in UHMWPE fibre composites
< X1, max
A case study is created using the analytical and semi-analytical
x2 =
1
2 0 ( b0
2 cos
X2 tan ) for X1 > X1, max
model to show the usability of the models on a layered material. The
(34) compressive response of a composite made from ultra high molecular
weight polyethylene fibres was investigated by Attwood et al. (2015).
where x i are deformed coordinates and Xi are initial coordinates ex- The composite was composed of plies with alternating 0 and 90 fibre
pressed in a coordinate system placed in the geometry centre. The angle orientation. A kink band was observed to develop during compression.
0 is the angle of the largest fibre misalignment, which is in the centre Due to a notch initially present in the geometry, the transition to a kink
of the geometry. The parameter b0 is the extent of the imperfection in band happened gradually and the deformation ended in a state with
the x1-direction. Furthermore the parameters kink band broadening. Kink bands have frequently been observed to
b0 b0 develop in a single ply on a fibre-matrix level. In the current case the
X1, min = X2 tan X1, max = X2 tan kink band developed on a ply level between the 0 - 90 layers. Pre-
2 cos 2 cos (35)
viously, the models have been used to investigate kink band formation
are introduced which give the start and end coordinate for the im- in composites on the fibre-matrix scale, but the models will in the
perfection. Simplified imperfections can also be implemented but the current case study be modified to develop on ply-level. In the case
imperfection used has shown to give consistent and stable results. The study, material parameters obtained for UHMWPE fibre composites will

398
114 Chapter 9 - P4

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

Fig. 5. Normalized compressive stress 11/ G as a function of normalized end-displacement u/ L 0 for different normalized fibre diameters.d f / L0

The bulk modulus K, the shear secant modulus Gs and the tangent
shear modulus Gt introduced in the instantaneous moduli (Eq. (37)) are
calculated based on
1 1
3 1 2 3 1 2 E
Gs = Gt = K=
Es E Et E 3(1 2 ) (40)

where ν denotes Poisson's ratio and E is Young's modulus. The tangent

modulus Et and secant modulus Es are calculated using the Ramberg-
Osgood relation
1

( )
n 1
3n eq
Et = E 1 + 7 y

( )
n 1 1
3 eq
Es = E 1 + 7 y
(41)

where y is a reference yield stress. The material parameters for the 90

plies are taken from Liu et al. (2014) and are given by

Fig. 6. Illustration of deformed geometries from the finite element simulations E 90 = 150 [MPa] 90 = 0.3 y,90 = 3.12 [MPa]
with different normalized fibre diameters d f / L0 .The band orientation is chosen n90 = 10 Et90 90
, min = E /1000 (42)
arbitrarily to be = 15 . The darker zones are areas of high shear straining.
Superscripts (•)90 corresponds to values associated with the 90 plies.
be used. The material behaviour of the constituents for the semi-ana- The parameters y,90 and n90 are the reference yield stress and hard-
lytical and analytical model will be explained in the following. ening exponent for the constituent. The reference yield stress is chosen
based on the shear yield stress y = 1.8 [MPa] measured in the ex-
5.1. Behaviour of plies for semi-analytical model periments by Russel et al. (2013) and assuming a von Mises yield cri-
terion entails the yield stress y = 3 y . The uni-axial response is de-
Different material behaviours are given to the 0 and 90 plies. The scribed using a Ramberg-Osgood relation according to
behaviour of the 90 plies are described using a J2-deformation theory 3 y n
introduced by Stören and Rice (1975). The tensor of instantaneous = +
E 7E y (43)
moduli Lijkl are given by
2 4 where σ and ε are the uni-axial stress and strain. Furthermore, a re-
Lijkl = Gs ( ik jl + il jk ) + K Gs ij kl (Gs Gt ) mij mkl , striction is set on the Ramberg-Osgood curve, so the tangent modulus
3 3 (37)
does not become smaller than a specified value Et , min . The limit stress L
where, and strain L where the minimum tangent modulus is reached are
1
mij = sij.
( )
L 3 y L
n
2 eq (38) L = E
+ 7E y

1
The stress eq is the equivalent Von Mises stress and sij are compo-
nents of the deviatoric stress tensor
L = y
( E
Et , min
1 ) 7
3n
n 1

(44)
1
= 3 J2 , J2 = sij sij, sij = kk
. These values are illustrated on Fig. 7. For an equivalent uni-axial
eq
2
ij ij
3 (39)
stress eq < L then the moduli given by Eq. (41) are to be used and

399
9.5 Article P4 115

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

( )
3 y n
L L
L = G1
+ 7 G1 y

L = y
( G1
G2
1) 7
3n
n 1
(51)
where L and L is the shear stress and strain where the tangent is equal
to G2 for a curve following a Ramberg-Osgood relation. The quantities
are shown in Fig. 7. The material parameters used for the analytical
kink band model are
E = 150 [MPa] G1 = 60 [MPa] G2 = G1/1000
Fig. 7. Shear stress-strain relation for the semi-analytical and analytical model.
The bi-linear curve represents the response used for the analytical model. = 0.3 n = 10 y = 1.8 [MPa]
(52)

Et = Et , min 5.3. Results

eq
Es =
eq
eq L
Two different models that can predict the steady-state broadening
= + state are used to estimate the response of a composite made from ultra-
eq Et , min L
(45)
high molecular-weight polyethylene fibres. The two models, a semi-
are to be used for eq > L . analytical and an analytical model, were initially presented in Jensen
The 0 plies are assumed to behave orthotropic and linear elastic and (1999) and Skovsgaard and Jensen (2018c) with normalized material
can thereby be described using Hooke's Law for an orthotropic material. values. The influence of the band orientation β upon the steady-state
The relation between stresses and strains presented in a matrix format is stress ss and fibre rotation ss are investigated for UHMWPE fibres in
given by the present case study. The results are compared and discussed with
1 23 32 21 + 31 23 31 + 21 32
0 0 0 respect to previously conducted experiments.
11
E2 E3 E2 E3 E2 E3
11
The lock-up angle ss of the plies within the kink band at steady-
12 + 13 32 1 32 + 31 12
0 0 0 state as function of initial kink band orientation β is presented in Fig. 8.
31 13
22 E3 E1 E3 E1 E3 E1 22
33 13 + 12 23 23 + 13 21 1 33 The conventionally assumed relationship between the lock-up angle
= 12 21
0 0 0
12 E1 E2 E1 E2 E1 E2 12 and the band orientation = 2 is also illustrated in the figure. Dif-
13 0 0 0 2 G12 0 0 13 ferent arguments are given in the literature on how to determine the
23 0 0 0 0 2 G13 0 23
band orientation β at steady-state for a given composite. Moran et al.
0 0 0 0 0 2 G23 (1995) argue that the energetically favourable mode sets the band or-
(46) ientation i.e. the band orientations that will result in the lowest
broadening stress ss . The two models used predict that the lock-up
where,
angle asymptotically approaches the linear relationship = 2 as the
1 12 21 23 32 31 13 2 12 23 31 angle β increases. For small values of β the two models deviate from the
=
E1 E2 E3 (47) linear relation. Similar deviation was seen by Hsu et al. (1999) for a
The moduli of the orthotropic constitutive relation comply to the carbon fibre composite where they used a micro-mechanical finite
symmetries Lijkl = Lklij , this entails the relations element model with fixed band orientations. In the experiments con-
ducted by Attwood et al. (2015), they investigated the compressive
E2 E3 E3
21 = 12 31 = 13 32 = 23 response of UHMWPE fibres. In the experiments a kink angle 36
E1 E1 E2 (48) and a lock up angle of ss 60 was observed. The experimental ob-
The material parameters for the 0 plies are again taken from Liu servation, that < 2 , disagrees with the predictions from the two
et al. (2014) and Russel et al. (2013) and are given by models, which can be due to the fact, that the models do not take de-
cohesion into account. Furthermore, it is unclear if the angles were
E1 = 68 [GPa] E2 = 150 [MPa] E3 = 150 [MPa]
measured upon loaded or unloaded specimens. There will be a change
G12 = G13 = 60 [MPa] G23 = E2/(1 + 23), of geometry going from a loaded to an unloaded specimens due to
12 0 13 0 23 0.5 (49)
The material behaviour is highly anisotropic, which can be seen in
the three order of magnitude difference between the moduli in tension
E1 and shear G12 .

5.2. Behaviour of plies for the analytical model

The original analytical model was created for kink band broadening
on the fibre/matrix scale. In the case of UHMWPE fibres, kink bands are
seen to develop on ply level. In the model presented in Skovsgaard and
Jensen (2018c) a matrix volume fraction c m was used due to the large
difference in transverse and shear properties of the constituents. In the
case of UHMWPE fibres it is assumed that the shear and transverse
properties for the two constituents (0 and 90 plies) are the same.
To make a better comparison with the semi-analytical model, then
the yield stress used is
L G2
y
12 = G1 L Fig. 8. Steady-state lock-up angle ss as a function of initial kink band or-
G1 G2 (50) ientation β. Results are shown for the two models. Additionally, the conven-
where, tional assumption of fibre lock-up = 2 is included for comparison.

400
116 Chapter 9 - P4

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

using the semi-analytical model with decreasing fibre diameters.

In the second case study the analytical and semi-analytical approach
were used to study kink band broadening in a composite made from
ultra high molecular-weight polyethylene fibres. The study was made
for UHMWPE fibre composites with a 0/90 ply lay-up. Kink band for-
mation and broadening have been observed to develop due to shearing
between different plies instead of shearing between fibres and the
surrounding matrix material. The 0 - and the 90 plies were treated as
two different constituents. The 0 plies were modelled as been linear
elastic and orthotropic in the semi-anlytical model. Furthermore, the
90 plies were given an elastic-plastic behaviour and were assumed to
behave isotropic. The analytical model assumed rigid fibres, which for
the UHMWPE fibres are a valid assumption due to the three order of
magnitude difference in the tension and shear moduli. A simple stress-
strain relationship was used in the analytical model, which results in a
transcendental equation to extract the stress and fibre lock-up angle at
Fig. 9. Steady-state kink band propagation stress 11 in [MPa] as a function of steady-state kink band broadening. The influence of the band orienta-
initial kink band orientation β. Results are shown for the two models. tion, β on the steady-state stress and lock-up angle was investigated in
the case study. The two models predict steady-state quantities that are
elasticity and relaxation/creep. in good agreement with the experimental observations done by
In the experiments performed by Liu et al. (2014) they investigated Attwood et al. (2015) and Liu et al. (2014). The results presented and
the collapse of a composite beam made from UHMWPE fibres. A tran- experimental observations argue that the kink band orientation at
sition from a shear dominated failure to a plastic hinge failure was steady-state broadening can be influenced by the boundary conditions.
observed, going from short to long beams. In the short shear dominated
beams, the failure was similar to the formation of a kink band. Even Acknowledgements
though the kink failure investigated in the present study is based on
pure compressive loading 11 in the direction if 0 plies, then interesting This work was supported by the Thomas B. Thriges fund.
similarities can be drawn. The band orientation in the shear dominated
failure was in the range = 0 5 and the fibre rotation was ap- References
proximately 20 . These angles are determined based on the figures
presented in Liu et al. (2014). The angles are in good agreement with Attwood, J.P., Fleck, N.A., Wadley, H.N.G., Deshpande, V.S., 2015. The compressive re-
sponse of ultra-high molecular weight polyethylene fibres and composites. Int. J.
the predictions from the two models presented in Fig. 8. These simi- Solid Struct. 71, 141–155. https://doi.org/10.1016/j.ijsolstr.2015.06.015. https://
larities argue that the kink band angle ss at steady-state broadening doi.org/10.1016/j.ijsolstr.2015.06.015.
can be influenced by the boundary conditions. Beuth, J.L., 1992. Cracking of thin bonded films in residual tension. Int. J. Solid Struct. 29
(13), 1657–1675. https://doi.org/10.1016/0020-7683(92)90015-L. https://doi.org/
The steady-state stress 11ss applied to the base material as a function 10.1016/0020-7683(92)90015-L.
of initial band orientation β is presented in Fig. 9. The results in the Budiansky, B., Fleck, N., 1993. Compressive failure of fibre composites. J. Mech. Phys.
figure argue that the broadening stress lowers with increasing band Solid. 41 (I), 183–211. https://doi.org/10.1016/0022-5096(93)90068-Q.
Budiansky, B., Fleck, N., Amazigo, J., 1998. On kink-band propagation in fiber compo-
orientation. In the study conducted by Jensen (1999) similar results sites. J. Mech. Phys. Solid. 46, 1637–1653. https://doi.org/10.1016/S0022-5096(97)
were predicted using material data for a carbon fibre-reinforced PEEK 00042-2.
composite. The results from the two models presented in Fig. 9 predict a Byskov, E., Christoffersen, J., Dencker Christensen, C., Sand Poulsen, J., 2002. Kinkband
formation in wood and fiber composites—morphology and analysis. Int. J. Solid
steady-state stress of 11 5 - 6 [MPa] for a kink band orientation
Struct. 39 (13–14), 3649–3673. https://doi.org/10.1016/S0020-7683(02)00174-9.
= 30 35 . This is in very good agreement with the experimental Christoffersen, J., Jensen, H.M., 1996. Kink band analysis accounting for the micro-
results presented in Fig. 10(a) in Attwood et al. (2015). The results from structure of fiber reinforced materials. Mech. Mater. 24 (4), 305–315. https://doi.
org/10.1016/S0167-6636(96)00052-X.
the analytical and semi-analytical model deviate for small band or-
Evans, A.G., Adler, W.F., 1978. Kinking as a mode of structural degradation in carbon
ientations, which also was observed in Skovsgaard and Jensen (2018c) fiber composites. Acta Metall. 26 (5), 725–738. https://doi.org/10.1016/0001-
for a carbon fibre-reinforced polymer composite. 6160(78)90023-8.
Fleck, N.A., Budiansky, B., 1991. Compressive failure of fibre composites due to micro-
buckling. Proc. 3rd Symp. on Inelastic Deformation of Composite Materials 235–273.
Fleck, N.A., Deng, L., Budiansky, B., 1995. Prediction of kink width in compressed fiber
6. Conclusions composites. J. Appl. Mech. 62, 329–337.
G’sell, C., Aly-Helal, N.A., Jonas, J.J., 1983. Effect of stress triaxiality on neck propaga-
Steady-state kink band broadening has been investigated for fibre tion during the tensile stretching of solid polymers. J. Mater. Sci. 18 (6), 1731–1742.
https://doi.org/10.1007/BF00542069.
reinforced composites and layered materials. The emphasize was given Hsu, S.-Y., Vogler, T., Kyriakides, S., 1999. On the axial propagation of kink bands in fiber
to the explanation of a semi-analytical and analytical approach to study composites : Part II analysis. Int. J. Solid Struct. 36, 575–595. https://doi.org/10.
the propagation instability. The models are formulated in a general 1016/S0020-7683(98)00030-4.
Jensen, H.M., 1999. Analysis of compressive failure of layered materials by kink band
framework and the constitutive behaviour for the constituents are
broadening. Int. J. Solid Struct. 36 (23), 3427–3441. https://doi.org/10.1016/S0020-
needed as input. Two different composites are used in two case studies. 7683(98)00158-9.
The case studies both validate and show the usability of the used Jensen, H.M., Christoffersen, J., 1997. Kink band formation in fiber reinforced materials.
J. Mech. Phys. Solid. 45 (7), 1121–1136. https://doi.org/10.1016/S0022-5096(96)
models.
00126-3.
In the first case study a carbon fibre-reinforced PEEK composite is Kyriakides, S., Arseculeratne, R., Perry, E.J., Liechti, K.M., 1995. On the compressive
investigated due to its well known constitutive behaviour of the con- failure of fiber reinforced composites. Int. J. Solid Struct. 32 (6–7), 689–738. https://
stituents. The effect of the fibre diameter was investigated upon kink doi.org/10.1016/0020-7683(94)00157-R. https://doi.org/10.1016/0020-7683(94)
00157-R.
band broadening using a micromechanical finite element model. As the Liu, X.H., Moran, P.M., Shih, C.F., 1996. The mechanics of compressive kinking in uni-
fibre diameter decreases the bending stiffness reduces, which leads to a directional fiber reinforced ductile matrix composites. Compos. B Eng. 27 (6),
sharp transition from kinked to unkinked area. This sharp transition is 553–560. https://doi.org/10.1016/S1359-8368(96)00008-X.
Liu, G., Thouless, M.D., Deshpande, V.S., Fleck, N.A., 2014. Collapse of a composite beam
consistent with the assumptions made for the two simplified models. made from ultra high molecular-weight polyethylene fibres. J. Mech. Phys. Solid. 63
The results obtained using the FE-model approach the result obtained

401
9.5 Article P4 117

S.P.H. Skovsgaard, H.M. Jensen European Journal of Mechanics / A Solids 74 (2019) 394–402

(1), 320–335. https://doi.org/10.1016/j.jmps.2013.08.021. https://doi.org/10. materials. J. Appl. Mech. 85 (6). https://doi.org/10.1115/1.4039573. 061005.
1016/j.jmps.2013.08.021. http://appliedmechanics.asmedigitalcollection.asme.org/article.aspx?doi=10.
Moran, P.M., Liu, X.H., Shih, C.F., 1995. Kink band formation and band broadening in 1115/1.4039573.
fiber composites under compressive loading. Acta Metall. Mater. 43 (8), 2943–2958. Skovsgaard, S.P.H., Jensen, H.M., 2018b. Constitutive model for imperfectly bonded
https://doi.org/10.1016/0956-7151(95)00001-C. fibre-reinforced composites. Compos. Struct. 192, 82–92. https://doi.org/10.1016/j.
Nizolek, T., Begley, M., McCabe, R., Avallone, J., Mara, N., Beyerlein, I., Pollock, T., 2017. compstruct.2018.02.053. February. http://linkinghub.elsevier.com/retrieve/pii/
Strain fields induced by kink band propagation in Cu-Nb nanolaminate composites. S0263822317332002.
Acta Mater. 133, 303–315. https://doi.org/10.1016/j.actamat.2017.04.050. http:// Skovsgaard, S.P.H., Jensen, H.M., 2018c. International Journal of Solids and Structures
www.sciencedirect.com/science/article/pii/S1359645417303427. Three-dimensional constitutive model for elastic-plastic behaviour of fibre-reinforced
Piggott, M.R., Harris, B., 1980. Compression strength of carbon, glass and kevlar-49 fiber composites. Int. J. Solid Struct. 139–140, 150–162. https://doi.org/10.1016/j.
reinforced polyester resins. J. Mater. Sci. 15 (10), 2523–2538. https://doi.org/10. ijsolstr.2018.01.032. https://doi.org/10.1016/j.ijsolstr.2018.01.032.
1007/Bf00550757. Stören, S., Rice, J.R., 1975. Localized necking in thin sheets. J. Mech. Phys. Solid. 23 (6),
Russell, B.P., Karthikeyan, K., Deshpande, V.S., Fleck, N.A., 2013. The high strain rate 421–441. https://doi.org/10.1016/0022-5096(75)90004-6.
response of Ultra High Molecular-weight Polyethylene: from fibre to laminate. Int. J. Tojaga, V., Skovsgaard, S.P.H., Jensen, H.M., 2018. Micromechanics of kink band for-
Impact Eng. 60, 1–9. https://doi.org/10.1016/j.ijimpeng.2013.03.010. https://doi. mation in open-hole fi bre composites under compressive loading. Composites Part B
org/10.1016/j.ijimpeng.2013.03.010. 149 (March), 66–73. https://doi.org/10.1016/j.compositesb.2018.05.014. https://
Skovsgaard, S.P.H., Jensen, H.M., 2018a. Steady-state kink band propagation in layered doi.org/10.1016/j.compositesb.2018.05.014.

402
Chapter 10

P5
Load transfer at a bolted joint in a laminate made from
ultra-high molecular weight polyethylene fibres

10.1 Motivation
Composites are used in several applications for different purposes. Carbon- and glass fibre-
reinforced composites are often used to achieve structures with high strength and low weight.
Composite plates made from ultra-high molecular weight polyethylene (UHMWPE) fibres
are used in the industries of personnel and vehicle armour (See e.g. the bulletproof west
in Fig. 1.1b). UHMWPE was initially commercialised by the company DSM in the 1970s
under the trade name Dyneema®. Fibres made from UHMWPE possess an incredible
strength up to 3 GPa due to a gel-spinning/hot drawing process. Composite plates made
from UHMWPE fibre laminates are in most cases composed of an equal amount of 0◦ and
90◦ plies. Even though the fibres possess high strength, the composite shear strength is of
the order of 2 MPa i.e. three orders of magnitude difference from the fibre strength. This
mismatch in composite shear and fibre strength and stiffness make the composite plates ideal
for applications where ballistic energy absorption is important. It is challenging to measure
the mechanical properties of laminates made from UHMWPE due to this stiffness mismatch.
Furthermore, it is difficult to transfer load into the composite plates which give rise to special
geometries in experimental set-ups.
The current article regards the transfer of loading at a bolted joint in a laminate made from
UHMWPE fibres. The article is based on the work performed at the Engineering Department
120 Chapter 10 - P5

at the University of Cambridge, which is the reason for the sudden change in research field
from kink band formation to the one described in the following article.

10.2 Method
The article is mainly based on experimental work on the load transfer into plates made from
UHMWPE. Specimens including holes were machined. The specimens were clamped in the
bottom and a high steel bolt was placed through the top hole and displaced to measure and
observe the pull-through behaviour.
A displacement controlled screw-driven test machine was used. The displacement was
measured using a laser extensometer, the reaction force was measured by the load cell of
the test machine. Furthermore, the influence of pre-clamping was investigated. The amount
of transverse pre-clamping and the development of the loading was measured through an
in-house made load cell composed of two strain gauges mounted on an aluminium tube. In
the experiments, the specimen dimensions and the initial pre-clamping were varied.
Based on the experimental observations, a simple analytical model based on line-field
theory was developed that adequately catches the main mechanism of load transfer. The
analytical model gives an upper bound solution to the load transfer strength.

10.3 Contribution
The contributions to the article comprise: the development of the experimental set-up; per-
forming the experiments; creating an analytical model; creating a failure map and writing the
article.

10.4 Main findings

One of the main contributions to the field is the development of a failure map for the load
transfer mechanism that distinguish between three failure mechanisms based on the specimen
geometry. The mechanism of load transfer is determined based on the deformation pattern
of the 0◦ and 90◦ plies obtained through a combination of X-ray tomography and optical
images. It can be concluded, that the load transfer strength in UHMWPE composite plates is
dominated by the inter-laminar shear yield strength between the 0◦ and 90◦ plies. Through
the experiments and the analytical model, it can be confirmed that the inter-laminar shear
yield strength depends linearly upon the out-of-plane pressure, similarly to Coulomb’s law of
friction.
10.5 Article P5 121

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]
International Journal of Solids and Structures xxx (xxxx) xxx

Contents lists available at ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

Load transfer within the bolted joint of a laminate made from

ultra-high molecular weight polyethylene fibres
S.P.H. Skovsgaard a,b, H.M. Jensen b, N.A. Fleck a,∗
a
Department of Engineering, University of Cambridge, Trumpington Street, Cambridge, UK
b
Department of Engineering, Aarhus University, Inge Lehmanns Gade 10, Aarhus C, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: The mechanism of load transfer within the bolted joint of a laminate sheet made from ultra-high molecu-
Received 19 February 2019 lar weight polyethylene (UHMWPE) plies is investigated both experimentally and by an analytical model.
Revised 29 July 2019
The nature of load transfer and the active failure mechanisms are obtained as a function of joint ge-
Accepted 14 August 2019
ometry and of the lateral clamping force on the faces of the laminate (by pre-tensioning of the bolt). A
Available online xxx
combination of X-ray tomography and optical microscopy reveal that the dominant failure mechanism in
Keywords: the clamped joint is shear failure involving splits of the 0° plies and sliding at the interface between the
UHMWPE fibres 0° and 90° plies. A simple analytical model is developed for this shear failure mechanism and, upon not-
Joint failure ing the competing failure mechanisms of bearing failure, bolt shear and of tensile failure of the 0° plies,
Failure map a failure mechanism map is constructed in terms of the geometry of the bolted joint, for the case of
Pressure dependent shear strength no pre-tension of the bolt. The analytical model for shear failure suggests that the enhancement in joint
strength with increased pre-tensioning of bolt is due to the fact that the shear strength of the UHMWPE
increases with increasing hydrostatic pressure.
© 2019 Published by Elsevier Ltd.

1. Introduction that lead to the choice of fibres used in Dyneema®. They mea-
sured the effect of draw ratio upon the tensile modulus and
Ultra-high molecular weight polyethylene (UHMWPE) fibres strength, and concluded that an extension ratio of λ = 32 by hot
embedded in a polyurethane matrix have a high specific modu- drawing led to a fibre strength of 3.0 GPa and a Young’s mod-
lus and a high specific strength, and are commonly used for per- ulus of 90 GPa. More recently, Govaert and Lemstra (1992) and
sonnel and vehicle armour. Additionally, UHMWPE fibres are used Govaert and Peijs (1995) explored the sensitivity of the tensile re-
for ropes, nets, footwear, cut resistant gloves and for air cargo sponse of UHMWPE fibres to temperature and to strain-rate.
containers. The company DSM1 commercialized fibres made from Over the past decade, several authors have developed models
UHMWPE in the late 1970s under the trade name Dyneema®. to predict the ballistic performance of UHMWPE composite plates;
UHMWPE has extremely long molecular chains and, when fibres for example, Grujicic et al. (2009) have developed a continuum-
are produced using a gel-spinning/hot drawing process, the fibres damage based constitutive model and implemented it within a
possess a high strength on the order of 3 GPa, (Russell et al. 2013). finite element (FE) code. Iannucci and Pope (2011) have devel-
The fibres are coated in a polyurethane (PU) resin solution and oped a model for the impact response of sheets made from high
are then formed into [0/90/0/90] stacks. The PU solvent is re- performance polymer fibres. Koh et al. (2010) investigated the
moved during a drying process and the stacks are then hot behaviour of UHMWPE yarns by both quasi-static and dynamic
pressed. tests. Additionally, Karthikeyan et al. (2013) performed quasi-static
Several studies have been performed to determine the mechan- and dynamic impact tests on composite plates made from ei-
ical properties of UHMWPE fibres and composite plates. In the ther UHMWPE fibres or carbon fibre-reinforced polymers. They ob-
early work of Wilding and Ward (1978), the creep and recov- served that composites of high indentation strength in the quasi-
ery of ultra-high modulus polyethylene fibres were determined. static tests also had a high failure impulse. Russell et al. (2013)
Smith and Lemstra (1980) conducted one of the early studies created a series of test methods for the mechanical performance
of fibres, yarns and laminates made from UHMWPE. They high-
∗
lighted the need to develop new geometries for tensile testing
Corresponding author.
due to the difficulties in transferring load into laminates made
E-mail address: [email protected] (N.A. Fleck).
1
DSM, Het Overloon 1, 6411 TE Heerlen, The Netherlands. from UHMWPE. A practical means of exploring load transfer in the

https://doi.org/10.1016/j.ijsolstr.2019.08.014
0020-7683/© 2019 Published by Elsevier Ltd.

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
122 Chapter 10 - P5

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

2 S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx

(a) (b)

Fig. 1. (a). Specimen geometry. All dimensions are in mm. (b). Experimental set-up. A three-dimensional view is shown on the left. A sectional view of the top part is shown
on the right. All dimensions are in mm.

presence of confinement is to transfer load via bolted joints: this we shall explore the extent to which these failure mechanisms per-
motivates the current study. sist in UHMWPE laminates of high in-plane strength but of very
We emphasise that it is difficult to measure the mechanical low shear strength.
properties of laminates made from UHMWPE using conventional
test methods due to the very low shear strength of both fibres 2. Test method
and matrix. Consequently, indirect test methods have been de-
veloped. For example, Attwood et al. (2014) developed an out- We investigated the mechanism of load transfer into a bolted
of-plane compression test to determine the pressure sensitiv- joint comprising HB26 Dyneema® [0°/90°] plates with and without
ity of UHWMPE laminates. The inter-layer strength was mea- transverse clamping. The plates had equal volume fractions of 0°
sured by Liu et al. (2014) via tests on an end-loaded cantilever and 90° plies and a ply thickness of h = 60 μm. Specimens of over-
beam. They extracted the elastic and plastic properties by vary- all thickness t = 6.5 mm and ply layup (0°, 90°)54 were machined to
ing the load level and by suitable positioning of the loading the geometry as depicted in Fig. 1(a), with bolt diameter d = 8 mm,
pin; a FE-model was used to aid interpretation of their results. ligament width w = 6 mm and ligament height b = 2, 4, 6 or 8 mm.
The compressive response of UHWMPE fibres, and composite The lower part of each specimen was clamped between two hard-
plates made from UHMWPE, was determined experimentally by ened steel plates using eleven M4 (Grade 12.92 ) steel bolts. The
Attwood et al. (2015) and was compared with fibre-kinking the- need for a large number of small bolts to introduce load into the
ory. In a recent study, Liu et al. (2018) determined the Mode I and specimen without local joint failure at each of the small bolts
II fracture toughness of a UHMWPE laminate. They performed a is traced to the fact that UHMWPE laminates have a high ten-
penetration experiment with a sharp-tipped punch and compared sile strength but a low shear strength. Special measures must be
the measurements with a FE-model based upon a crystal plas- taken to ensure load introduction into the specimen, as discussed
ticity model for the ply behaviour. Karthikeyan et al. (2013) per- by Russell et al. (2013). The variables b and w for the ligament
formed both quasi-static loading and dynamic impact tests on end- dimensions are used in the present study in order to emphasise
clamped UHMWPE beams. They found that the method of confine- their role. They are closely related to the overall width of the joint
ment had a large influence on both the quasi-static and dynamic W = 2w + d and to the end-distance from the centre of the bolt to
behaviours. the free edge of the plate e = b + d/2.
Several investigations have been performed on the mechan- An in-plane bearing load F was imposed on each specimen via
ics of mechanically fastened joints in fibre-reinforced polymers. A a steel bolt (Grade2 12.9) of diameter d and thick washers (that
failure mechanism map for single-lap bolted joints in CFRP lami- is, clamping rings) of diameter dw = 25 mm; this was achieved by
nates was constructed by Smith et al. (1986) on the basis of a set the loading arrangement of Fig. 1(b). In addition, the effect of the
of tests on multidirectional CFRP laminates. Failure was by net- clamping pre-load T0 in the bolt upon the shear response was in-
section tension, bearing or by shear-out along splits within the vestigated by suitable torqueing of the bolt. The bolt was displaced
0° plies. A similar methodology is followed in the present study. in the x1 direction relative to the composite panel using a screw-
Camanho and Matthews (1997) distinguished five failure modes driven test machine with a displacement rate of 1 mm/min. The
of composite joints: tension, shear, cleavage, bearing and a pull- bolt displacement u was measured by a laser extensometer and
through failure mode. A decade later, Thoppul et al. (2009) made a the reaction force F was determined by the load cell of the test
thorough review of the state-of-the-art methods to study the fail-
ure of composite joints. The focus in these previous studies was on
glass and carbon fibre-reinforced polymers. In the present study, 2
ASTM F568M

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
10.5 Article P5 123

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx 3

Table 1
Dimensions, confinement and observed failure modes.

t, mm d, mm b, mm w, mm Bolt pre-load T0 , kN Failure mode

12 6 15 7 free Bolt shear

6.5 6 40 3 0 Tension
6.5 6 4 25 0 Buckling
12 8 4 6 0 Shear
12 8 4 6 free Shear
6.5 8 4 6 0 Shear
6.5 8 4 6 0.84 Shear
6.5 8 4 6 2.89 Shear
6.5 8 4 6 3.67 Shear
6.5 8 4 6 4.75 Shear
6.5 8 4 6 6.59 Shear
6.5 8 4 6 7.67 Shear
6.5 8 4 6 8.92 Shear
6.5 8 2 6 free Shear
6.5 8 4 6 free Shear
6.5 8 6 6 free Shear
6.5 8 8 6 free Shear
6.5 8 2 6 0 Shear
6.5 8 4 6 0 Shear
6.5 8 6 6 0 Shear
6.5 8 8 6 0 Shear
6.5 8 2 6 8.92 Shear
6.5 8 4 6 8.92 Shear
Fig. 2. Failure map for the choice T0 = 0. Laminate thickness t/d = 0.8 and ply thick- 6.5 8 6 6 8.92 Shear
ness h/d = 0.0075. 6.5 8 8 6 8.92 Shear

machine. The transverse clamping force T was measured via an in- (a)
house load cell consisting of two 120 Ω strain gauges mounted on
the opposing walls of an aluminium alloy tube. The strain gauges
were of dimension 3 mm x 1 mm and of gauge factor 2.15, and a
Wheatstone quarter bridge circuit was used for strain measure-
ment. The load cell was of length 20 mm, outer diameter 17 mm
and wall thickness 1.3 mm. The degree of bending of the load
cell was determined by the difference in axial strain between the
gauges and was found to be less than 20% of the mean value
throughout each test. We conclude that bending of the load cell
was negligible. The mean response of the gauges was used to cal-
(b)
culate the transverse clamping force T of the specimens, and this
transverse force was recorded by a data-logger. The clamping pre-
load T0 and the evolution of the load T during the test were mea-
sured and are reported below.

3. Observed failure modes

We begin by summarising the observed failure modes. The ac-

Fig. 3. The load transfer mechanism for shear failure of the bolted joint, with T0 ≥
tive failure mode of the bolted joint as a function of joint geom- 0. D denotes delamination between the [0°/90°] plies.
etry and of clamping pre-load was determined by a combination
of visual observation and X-ray CT microscopy.3 A failure map was
created for a bolted joint with zero clamping pre-load, T0 = 0, as CFRP laminates: they also found that shear failure dominates the
shown in Fig. 2; the location of the boundaries between failure map for low values of w/d and b/d.
mechanisms are derived in a subsequent section. The competing An additional failure mechanism was observed in the present
failure mechanisms for T0 = 0 are shear of the bolt, tensile failure study, that of a transverse plate-buckling mode for plates of large
of the 0° plies at the sides of the hole and shear failure of the ligament width w, and this is detailed below. Table 1 gives an
laminate; these are sketched in the inserts of Fig. 2. For the case overview of the dimensions, confinement and the corresponding
of pin loading with clamping washers absent, the shear mode of failure modes observed in the experiments.
failure was replaced by a bearing mode of failure. Fig. 2 is valid
for clamped specimens of low d/t ratio. The failure map was con- 3.1. Shear failure
structed for the choice d = 8 mm and t = 6.5 mm. Further work is
needed in order to determine the failure map for the clamped The shear mode of failure occurs for the clamped case, T0 ≥
specimen at large d/t to determine whether bearing failure inter- 0, in specimens of small ligament width and height. Interrupted
venes. This is beyond the scope of the present study. The failure tests and X-ray CT observations were performed in order to reveal
map in Fig. 2 is consistent with that of Smith et al. (1986) for 0/90 the deformation within the 0° plies, 90 o plies and the delamina-
tion between plies. An idealised view of this deformation mode is
sketched in Fig. 3; the accompanying interrupted test (and CT im-
3
Nikon X-Tek XT H 225ST, at an operating voltage of 50kV ages) are reported in Fig. 4 for the clamped case with T0 = 0, and

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
124 Chapter 10 - P5

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

4 S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx

(a)

(b)

(c)
Fig. 4. (a). Axial force F versus axial displacement u for the choice T0 = 0 and the unconstrained case (Free). For both tests, b = 4 mm, w = 6 mm and t = 12 mm. (b). CT
images of selected 90° or 0° plies near the mid-plane of the specimen, for the choice T0 = 0. Plan view along the x3 direction of Fig. 1. (c). Transverse views of the shear
failure of the plies above the pin for the clamped case with T0 = 0, and for the pin-loaded case, labelled free.

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
10.5 Article P5 125

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx 5

of geometry b = 4 mm, w = 6 mm and t = 12 mm. The initial unde- compression stress beneath the hole during the later stages of bolt
formed state of a representative 0° ply and 90° ply is sketched in pull-out in the shear failure mode.
Fig. 3(a) and is labelled as (1) in the CT images of Fig 4(b) and also
in the CT image of Fig. 4(c). The specimen was loaded to peak load 4. Sensitivity of joint strength to geometry and clamping
and then fully unloaded, to obtain the point (2) on the force F ver-
sus displacement u curve of the joint in Fig. 4(a); CT scans of rep- 4.1. Effect of initial clamping
resentative 0° plies and 90 o plies are given in Fig. 4(b) (again la-
belled (2)). Reloading and subsequent unloading brought the spec- The effect of initial clamping pre-load T0 upon the axial force
imen to state (3) as marked in Fig. 3, with the observed deforma- F versus bolt displacement u is given in Fig. 6(a) for the choice
tion state given in Fig. 4. The bolt washers prevent thickening of b = 4 mm, w = 6 mm and t = 6.5 mm. This choice of geometry en-
the specimen adjacent to the bolt, and shear failure of the 0° plies sures that the shear mode of failure is active. The evolution of
occurs. This is clear from the transverse section of the specimen in transverse clamping force T in each test is summarised in Fig. 6(b).
Fig. 4(c). The main features are as follows. Delamination between the 0° and
A simple kinematic representation of this failure mode is given 90° plies, as labelled zone D of Fig. 3(b), occurs at a very low load
in Fig. 3(b), and is described as follows. A portion B of 0° ply ma- of below 2 kN, and the subsequent response is linear up to the
terial translates by the same displacement u as that of the loading peak load, labelled Fm . The magnitude Fm increases (linearly) with
bolt. The remainder of the ply (labelled portion A) remains station- the pre-load T0 as shown in Fig. 6(c). This is explained by the pres-
ary, and so the collapse mechanism of the 0° ply comprises slid- sure dependence of UHMWPE fibre composites, as discussed by
ing by a displacement u along two splits (located at the boundary Attwood et al. (2014). The clamping force T also increases to a peak
between portions A and B). The deformation state of the 90° ply value of Tm with increasing bolt displacement u. The clamping
is slightly more complex. A portion C is undeformed, whereas the force resists out-of-plane swelling of the laminate, and for com-
portion E of the 90° ply remains bonded to the adjacent portion B pleteness Fig. 6(c) contains a plot of Fm versus Tm : again the rela-
in the 0° ply, and is thereby displaced by the displacement u. The tionship is linear. We note in passing that the peak load Fm occurs
portion F of the 90° ply contains the same fibres as portion E, and at the same value of bolt displacement, and the shift in clamp-
undergoes in-plane shear as sketched, and as shown in Fig. 4(b). ing force (Tm -T0 ) is constant for all specimens of a given geometry.
The portion F delaminates and slides with respect to the adjacent This is consistent with the notion that the kinematics of the shear
portion A of the 0° ply, thereby creating a delamination patch D. failure is insensitive to the level of clamping force.
Now consider the case of pin loading with the clamping wash-
ers absent, for the same geometry of b = 4 mm, w = 6 mm and
4.2. Effect of ligament height
t = 12 mm. The force F versus displacement u curve of the joint
is included in Fig. 4(a), and labelled ‘free’. A bearing failure occurs
The effect of ligament height b upon the bolt force F versus bolt
for this case of unconstrained out-of-plane expansion of the lam-
displacement u characteristic is summarised in Fig. 7(a) for 4 val-
inate, see the cross-section of state (3) in Fig. 4(c). The portion B
ues of b = 2, 4, 6 and 8 mm, w = 6 mm, t = 6.5 mm and T0 = 0. In all
of 0° plies and portion E of 90° plies (as defined in Fig. 3) un-
cases, shear failure occurred. The associated evolution of clamping
dergo out-of-plane microbuckling with intermittent delamination,
force T during each test is given in Fig. 7(b). Out-of-plane swelling
see Fig. 4(c).
of the laminate leads to an increase in clamping force with increas-
ing bolt displacement u, and a peak in the clamping force is at-
3.2. Competing collapse modes for the clamped case, T0 ≥ 0 tained at the same instant that F attains its peak value of Fm . The
peak load Fm increases with increasing b, as emphasised by the
An alternative failure mode is tensile failure of the 0° plies ad- plot of Fm versus b in Fig. 7(c). For completeness, this figure also
jacent to the loaded bolt. This has been reported previously by contains the sensitivity of Fm to b for T0 = 8.9 kN, for the case of
Attwood et al. (2014). Bolt shear intervenes for a joint of high as- a freely supported bolt (absent clamping) for which bearing fail-
pect ratio b/d and w/d. A more surprising mode that is observed ure occurs; it is clear that Fm also increases with the degree of
in the present study is out-of-plane plate-buckling for large w/d, clamping. Likewise, the peak value Tm increases with increasing
as shown in Fig. 5. The buckling failure mode dictates the peak degree of clamping, and increases in an almost linear manner with
load and occurs shortly before the load maximum, see the dashed increasing b, see Fig. 7(d).
line in Fig. 7(b). It is conjectured that this is due to the build-up
of a large tensile stress in the 90° plies above the hole leading to
5. Analytical model for shear failure

An analytical model is now proposed for the observed shear

failure of the bolted joint. The model assumes a collapse mecha-
nism and thereby gives an upper bound solution for the shear fail-
ure force. First, a relation between the axial force as function of
axial bolt displacement is derived for the case of relative sliding of
the interface between the 0° and 90° plies. Second, the additional
force due to in-plane shearing is derived, and it is shown that this
dissipation is sufficiently small for the increase in shear force to be
negligible.

5.1. Plastic dissipation by inter-laminar shearing

The portion F delaminates and slides with respect to the por-

tion A of the 0° ply, thereby creating a delamination patch D, see
Fig. 5. Optical image of specimen of width w = 25 mm, b = 4 mm, t = 6.5 mm, show- Fig. 3(b). The delamination patch D is divided into two zones D1
ing failure by buckling of plate beneath the pin. and D2, see Fig. 8. The axial bolt displacement u is related to the

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
126 Chapter 10 - P5

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

6 S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx

(a) (b)

(c) 18

17

16

15
T = T0
Fm 14
T = Tm
(kN) 13

12

11

10
0 2 4 6 8 10 12

T (kN)
Fig. 6. (a). Axial force F versus axial displacement u. (b). Transverse force T versus axial displacement u. (c). Maximum axial force Fm versus initial transverse force T0 and
maximum value Tm . Dashed lines are best fits to the data. Throughout, specimen ligament height b = 4 mm, width w = 6 mm and laminate thickness t = 6.5 mm.

rotation φ of the 90° plies by where t is the specimen thickness and h is the ply thickness. Now,
the principle of virtual work requires that
u = w sin φ (1)
 
δu δu
and the fibres are taken to be inextensible. The delamination F δ u = 2 n I τy A1 + A2 (4)
patches D1 and D2 are of area A1 and A2 , respectively, where
2 3
 
d where δ u is a virtual displacement. Here, a simple rigid-plastic
A1 = w cos φ b+ −u constitutive relation is assumed such that τ y is the inter-laminar
2 (2)
1 shear yield strength. The factors of 1/2 and 1/3 arise from the fact
A2 = w cos φ u that the assumed displacement field varies across the width of the
2
delamination patch. Upon making use of Eqs. (1) and (2), the vir-
The total number of ply interfaces nI , over which inter-laminar tual work statement (4) reduces to
sliding occurs, is
 
d 2
nI =
t
−1 (3) F = nI τy w cos φ b+ − w sin φ (5)
h 2 3

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
10.5 Article P5 127

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx 7

Fig. 7. (a). Axial force F versus axial displacement u for the choice T0 = 0. (b). Transverse force T versus axial displacement u for the choice T0 = 0. (c). Maximum axial force
Fm /w versus ligament height b. Solid lines show predictions by the analytical model. (d). Maximum transverse force Tm /w versus ligament height b. Dashed lines show best
fit to the data. Throughout, specimen w = 6 mm, thickness t = 6.5 mm besides an experiment showing buckling with w = 25 mm (dashed line).

5.2. Additional plastic dissipation due to in-plane shearing

The additional force F due to in-plane shearing is now de-

duced. The portion F in Fig. 3 undergoes in-plane shearing. The
portion F has an in-plane area of
 
d
AF = 2 b+ w cos φ (7)
2
Fig. 8. Sketch of the geometry for analytical model. The virtual displacement δ u is related to the virtual rotation δφ
by
with peak value Fpc for the plastic collapse mode achieved at φ =
0, such that δ u = w cos φ δφ (8)
 
d Using the displacement-rotation relation, Eq. (1). During a vir-
Fpc = nI τy w b+ (6)
2 tual displacement, the area AF shears an amount δφ . Then, the

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
128 Chapter 10 - P5

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

8 S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx

additional term in the principle of virtual work for in-plane shear- 7. Failure mechanism map
ing is of the form
The background to the construction of the failure map of
F δ u = AF h δφ τy n90 (9) Fig. 2 is now given. We consider each mechanism in turn. Intro-
duce the non-dimensional geometric parameters
where n90 is the number of 90° plies. Again, a simple rigid-
plastic constitutive relation is used, where the in-plane shear yield t b w
t̄ = b̄ = w̄ = (14)
strength τ y is identical to the inter-laminar yield strength. The d d d
relation (9) is simplified via Eqs. (7) and (8) to read
along with the non-dimensional force on the bolt
 
d F
F = 2 b+ τy n90 h (10) F̄ = (15)
2 d t τy

Upon normalising the additional axial force F due to in-plane where d is the bolt diameter, t the plate thickness and τ y is the
shearing by the peak load Eq. (6) due to inter-laminar shearing we inter-laminar shear yield strength.
obtain
 d

F 2 b+ n h 7.1. Shear failure
=  2
 90 (11)
Fpc b+ d nI w
2 The load maximum for the shear failure is given by Eq. (6) us-
ing the simple analytical model. The non-dimensional force at
Now, the number of interfaces nI is twice the number of 90°
shear failure (plastic collapse) is
plies n90 . Consequently, the additional force arriving from in-plane
shearing scales as Fpc
 1
1
F̄pc = = nI w̄ b̄ + (16)
d t τy 2 t̄
F h
= (12)
Fpc w
7.2. Bolt shear
Recall that the ply thickness h equals 60 μm and the ligament
width w equals 6 mm for the majority of the test specimens. Thus, The bolt carries a transverse shear force V = F/2 at two loca-
in-plane shearing will increase the axial force by only 1% and this tions. Assume that the bolt shears plastically when the shear stress
is deemed negligible. on the section attains the shear strength τ bf . High strength bolts
are almost elastic, ideally plastic in their response, with a tensile
√
strength of σ bf = 1200 MPa, and a shear strength τ bf = 1200/ 3
6. Comparison of shear failure prediction with observation MPa = 693 MPa by the von Mises yield criterion. Consequently, bolt
shear occurs at a load
The analytical model of shear failure is now compared with π
the experimental results by treating the inter-laminar shear Fb f = d2 τb f (17)
2
strength τ y as a free parameter that depends upon the degree
of clamping. The predictions of Eq. (6) are compared with the and, upon introducing the non-dimensionalisation we obtain
measured values of maximum force Fm in Fig. 7(c), assuming
Fb f π τb f 1
that τ y = 0.95 MPa for clamping-free, τ y = 2.2 MPa for T0 = 0 and F̄b f = = (18)
τ y = 2.5 MPa for T0 = 8.9 kN. Recall that the measured inter-laminar d t τy 2 τy t̄
shear strength is τ y = 2 MPa as obtained by Liu et al. (2014) and
Attwood et al. (2014); they used a double-notch shear test. The
7.3. Tensile failure of the laminate
agreement is satisfactory.
The out-of-plane clamping pressure increases the inter-laminar
Tensile failure of the fibres within the 0° plies occurs at an axial
yield strength as follows. Attwood et al. (2014) studied the out-of-
stress of σ f = 30 0 0 MPa within the fibres. Recall that the strength
plane compressive response of UHMWPE laminates. They observed
of the composite normal to the fibre direction is three orders of
a pressure sensitivity of the form
magnitude lower than in the direction of the fibres and is thereby
negligible. The strength of the composite in tension is
τy = τ0 + μ p (13)
1
where τ y is the shear yield strength, τ 0 is the strength in the σt = σ f c f (19)
2
absence of pressure, p is the pressure and μ is a non-negative
pressure sensitivity coefficient. Attwood et al. (2014) found that where cf = 0.83 is the volume fraction of fibres (Liu et al. (2014)).
a coefficient μ = 0.05 gave good agreement with experimental A factor of 1/2 is introduced due to the equal volume fraction of
results. The increased inter-laminar yield strength as observed 0° and 90° plies, and the fact that the 90° plies provide a negligi-
in the current study can be explained by the pressure sensitiv- ble contribution to the strength. The smallest cross sectional area
ity of UHMWPE. To illustrate this, consider a specimen of liga- normal to the force F is 2 w t, see Fig. 1. Consequently, the force at
ment width w = 6 mm, ligament height b = 8 mm and bolt diame- tensile failure is
ter d = 8 mm. An initial pre-load of T0 = 8.9 kN results in an average
pressure p= 43 MPa beneath the clamping ring. Upon substituting Ft = 2 w t σt (20)
an initial yield strength, τ 0 = 0.95 MPa, a pressure sensitivity coef-
and so the non-dimensional force F̄0 at tension failure is
ficient μ = 0.05 and p= 43 MPa into Eq. (13) the predicted yield
strength is τ y = 3.1 MPa. This is in reasonable agreement with the Ft σt
inferred value of τ y = 2.5 MPa. F̄t = =2 w̄ (21)
d t τy τy
Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
10.5 Article P5 129

JID: SAS
ARTICLE IN PRESS [m5G;August 29, 2019;10:6]

S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck / International Journal of Solids and Structures xxx (xxxx) xxx 9

7.4. Construction of failure map nancial support for this work in the form of an ERC MULTILAT
grant 669764, and to DSM for providing the Dyneema® compos-
A failure map for T0 = 0 is constructed with geometric axes b/d ites used in this study.
and w/d in order to identify regimes of dominance of the compet-
ing failure modes, see Fig. 2. The active mode has the lowest fail- References
ure load from the relations Eqs. (16), (18) and (21). The boundaries
Attwood, J.P., Fleck, N.A., Wadley, H.N.G., Deshpande, V.S., 2015. The compressive
are located by equating the failure load of competing mechanisms. response of ultra-high molecular weight polyethylene fibres and composites. Int.
The precise boundary between shear failure and out-of-plane plate J. Solids Struct. 71, 141–155. doi:10.1016/j.ijsolstr.2015.06.015.
buckling is unknown; consequently, the boundary is not drawn in Attwood, J.P., Khaderi, S.N., Karthikeyan, K., Fleck, N.A., Omasta, M.R., Wadley, H.N.G.,
Deshpande, V.S., 2014. The out-of-plane compressive response ofDyneema
the failure map. ®composites. J. Mech. Phys. Solids 70, 200–226. doi:10.1016/j.jmps.2014.05.017.
Camanho, P., Matthews, F., 1997. Stress analysis and strength prediction of mechan-
ically fastened joints in FRP: a review. Compos. Part A Appl. Sci. Manuf. 28,
8. Concluding remarks 529–547. doi:10.1016/S1359-835X(97)0 0 0 04-3.
Govaert, L.E., Lemstra, P.J., 1992. Deformation behavior of oriented UHMWPE fibers.
The present study highlights the dominance of the shear mode Colloid Polym. Sci. 270, 455–464. doi:10.10 07/BF0 0665989.
Govaert, L.E., Peijs, T., 1995. Tensile strength and work of fracture of oriented
of joint failure in a bolted joint made from a UHMWPE laminate, polyethylene fibre. Polymer (Guildf) 36, 4425–4431. doi:10.1016/0032-3861(95)
and subjected to out-of-plane clamping by the bolt. The 0° plies 96848-3.
split such that the central portion of 0° plies (adjacent to the bolt) Grujicic, M., Arakere, G., He, T., Bell, W.C., Glomski, P.S., Cheeseman, B.A., 2009.
Multi-scale ballistic material modeling of cross-plied compliant composites.
is sheared-out from the joint by movement of the bolt. The 90° Compos. Part B Eng. 40, 468–482. doi:10.1016/j.compositesb.20 09.02.0 02.
plies that are stacked with the central portion of 0° plies are also Iannucci, L., Pope, D., 2011. High velocity impact and armour design. Express Polym.
dragged-out of the joint by the bolt displacement. This leads to Lett. 5, 262–272. doi:10.3144/expresspolymlett.2011.26.
Karthikeyan, K., Russell, B.P., Fleck, N.A., O’Masta, M., Wadley, H.N.G., Desh-
tensile pull-through of the 90° plies and to delamination of the 0° pande, V.S., 2013. The soft impact response of composite laminate beams. Int. J.
plies. Impact Eng. 60, 24–36. doi:10.1016/j.ijimpeng.2013.04.002.
The strength at shear failure is increased substantially by in- Koh, A.C.P., Shim, V.P.W., Tan, V.B.C., 2010. Dynamic behaviour of UHMWPE yarns
and addressing impedance mismatch effects of specimen clamps. Int. J. Impact
creasing clamping force, and this is explained in terms of the pres-
Eng. 37, 324–332. doi:10.1016/j.ijimpeng.2009.10.008.
sure sensitivity of the shear strength of the UHMWPE compos- Liu, B.G., Kandan, K., Wadley, H.N.G., Deshpande, V.S., 2018. Deep penetration of
ite. A simple analytical model highlights the importance of slip ultra-high molecular weight polyethylene composites by a sharp-tipped punch.
J. Mech. Phys. Solids doi:10.1016/j.jmps.2018.06.001.
between the plies in providing the resistance to a shear failure.
Liu, G., Thouless, M.D., Deshpande, V.S., Fleck, N.A., 2014. Collapse of a composite
Also, a failure map is constructed and provides useful guidelines beam made from ultra high molecular-weight polyethylene fibres. J. Mech. Phys.
for joint strength as a function of geometry. In order to predict Solids 63, 320–335. doi:10.1016/j.jmps.2013.08.021.
the plate buckling mode at large ligament width w, a 3D finite el- Russell, B.P., Karthikeyan, K., Deshpande, V.S., Fleck, N.A., 2013. The high strain rate
response of ultra high molecular-weight polyethylene: from fibre to laminate.
ement model would be required and this is beyond the scope of Int. J. Impact Eng. 60, 1–9. doi:10.1016/j.ijimpeng.2013.03.010.
the present study. Smith, P., Lemstra, P.J., 1980. Ultra-high strength polyethylene filaments by solution
spinning/drawing. 3. Influence of drawing temperature. Polym. (U. K.) 21, 1341–
1343. doi:10.1016/0032- 3861(80)90205- 0.
Acknowledgements Smith, P.A., Pascoe, K.J., Polak, C., Stroud, D.O., 1986. The behaviour of single-
lap bolted joints in CFRP laminates. Compos. Struct. 6, 41–55. doi:10.1016/
0263- 8223(86)90067- X.
The research work was supported by the Thomas B. Thriges Thoppul, S.D., Finegan, J., Gibson, R.F., 2009. Mechanics of mechanically fastened
Fond; Aage og Johanne Louis-Hansens Fond; Knud Højgaards Fond; joints in polymer-matrix composite structures - a review. Compos. Sci. Technol.
69, 301–329. doi:10.1016/j.compscitech.2008.09.037.
Augustinus Fonden; Reinholdt W Jorck og Hustrus Fond; Marie
Wilding, M.A., Ward, I.M., 1978. Tensile creep and recovery in ultra-high modulus
& M. B. Richters Fond; and the Graduate School of Science and linear polyethylenes. Polymer (Guildf) 19, 969–976. doi:10.1016/0032-3861(78)
Techonology (Aarhus University). The authors are grateful for fi- 90208-2.

Please cite this article as: S.P.H. Skovsgaard, H.M. Jensen and N.A. Fleck, Load transfer within the bolted joint of a laminate made from
ultra-high molecular weight polyethylene fibres, International Journal of Solids and Structures, https://doi.org/10.1016/j.ijsolstr.2019.08.
014
Chapter 11

C1
Micromechanics of kink band formation in open-hole fibre
composites under compressive loading

11.1 Motivation
One of the main methodologies to transfer load between two structures is through bolted
joints. Structural applications made in fibre-reinforced composites may include open-holes
for different reasons. Experiments suggest that kink band formation is the dominant failure
mechanism for open hole fibre composites with a soft and ductile resin material. It has been
observed that fibre kinking is initiated from the hole edge and is followed by other failure
mechanisms e.g. delamination and fibre breakage.
The current publication deals with the investigation of compressive failure of open-hole
fibre composites by kink band formation. Despite the many contributions to the problem,
different aspects of the phenomenon still remain unknown. The original source of the
publication is: (Tojaga et al., 2018).

11.2 Method
The investigation of kink band formation is challenging since the simulations must include
large deformations and rotations together with non-linear elastic-plastic material behaviour.
Furthermore, often arc-length solvers are required since the equilibrium path includes snap-
back and snap-through behaviour.
132 Chapter 11 - C1

The finite element software Abaqus/CAE 6.14 together with a user subroutine (UMAT)
written in Fortran 77 are used to solve the problem. The constitutive model by Christoffersen
and Jensen (1996) is implemented in the UMAT. The model by Christoffersen and Jensen
(1996) is described in Section 3.2. Two versions of the constitutive model are implemented:
a version with perfect bonding and one with no bonding between the fibre and matrix con-
stituent. Material parameters from Kyriakides et al. (1995) are used in the study.

11.3 Contribution
The contributions to the publication comprise: changes made on a user subroutine initially
created by Sørensen et al. (2009) to have better convergence. Further contributions include
participations in supervision of Vedad Tojaga, discussion of the article content and proofread-
ing.

11.4 Main findings

Illustrations of the different fields (e.g. stress and strain fields) prior to the peak stress are
presented. These graphical illustrations give valuable inside to the initiation of kink bands.
The influence of fibre orientations relative to the loading direction is investigated for an
open-hole unidirectional fibre composite. Furthermore, the influence of perfect fibre/matrix
bonding and no bonding are investigated where the realistic compressive strength lies between
the two boundaries.
One of the main strengths behind the current publication, compared to previous studies on
open-hole compression of composites, is the constitutive relation that includes information
about the current stress and strain state for the constituents during loading.
11.5 Article C1 133

Composites Part B 149 (2018) 66–73

Contents lists available at ScienceDirect

Composites Part B
journal homepage: www.elsevier.com/locate/compositesb

Micromechanics of kink band formation in open-hole fibre composites under T

compressive loading
Vedad Tojaga∗, Simon P.H. Skovsgaard, Henrik Myhre Jensen
Department of Engineering, Aarhus University, Inge Lehmanns Gade 10, 8000 Aarhus C, Denmark

A R T I C LE I N FO A B S T R A C T

Keywords: The micromechanics of kink band formation in open-hole fibre composites under compressive loading is de-
Laminates scribed. The objective being the development of a methodology for designing of structural components with
Microstructures open-holes. Our results explain why failure by kink band formation propagates from the edges of an open-hole in
Plastic deformation a direction almost perpendicular to the loading direction and why the 0 plies govern the compressive failure of
Micro-mechanics
an open-hole laminate. The proposed design methodology accounts for the microstructure, including the fibre/
matrix bonding, and the nonlinear behaviour of the constituents, enabling it to prevent local failure at the hole
edges, or global failure, by kink banding of a laminate containing stress concentrations.

1. Introduction plastic behaviour of the constituents, making it suitable for analysing

plastic microbuckling. Sørensen et al. [4] implemented the model [3] as
The compressive strength of a unidirectional fibre composite is often a user subroutine (UMAT) into the finite element program ABAQUS.
less than 60 % of its tensile strength; hence, in many cases, the com- Jensen [5] extended the latter formulation to account for full decohe-
pressive strength dictates the design. An important mechanism of sion between the matrix and the fibres. The models in Refs. [3] and [5]
compressive failure in fibre composites is a localized imperfection- do not take the fibre bending rigidity into account; however, the fibre
sensitive material instability known as plastic microbuckling, leading to bending rigidity decreases with the fibre diameter and in most real-life
the formation of one or more kink bands by localization of plastic de- applications, fibre composites contain many thin fibres embedded in a
formation. Both terminologies, i.e. plastic microbuckling and kink band matrix. A constitutive model for imperfectly bonded fibre-reinforced
formation, are adopted in this paper. Manufacturing and curing of a polymer composites by Skovsgaard and Jensen [6] describes the tran-
composite inevitably introduce fibre misalignments. The matrix pro- sition between these two models. Recently, Skovsgaard and Jensen [7]
vides lateral stability to the fibres and prevents, or postpones, the oc- developed a three-dimensional constitutive model for elastic-plastic
currence of plastic microbuckling, inducing a complex strain field in the behaviour of fibre-reinforced composites inspired by the two-dimen-
matrix. After a small amount of fibre rotation under a remote com- sional model [3]. Depending on the given structural component and
pressive stress, geometric softening associated with the fibre rotation how well controlled the fibre misalignments are, due to manufacturing,
outweighs the plastic strain hardening of the matrix, and a microbuckle the critical stress may not be suitable for engineering applications.
nucleates and then propagates through the composite. The remote Beyond the critical stress, in the post-critical region, the kink band may
compressive stress, at which the composite fails by the formation of propagate (under steady-state conditions) through the material at a
kink bands, i.e. the critical stress, is sensitive to fibre misalignments constant stress known as the steady-state kinking stress. At steady-state,
since the load carrying capacity of the material is lost locally. The latter the fibres in the kink band stop rotating, i.e. lock-up, forcing the kink
description emphasizes the importance of accounting for the nonlinear band to spread into the base material. A semi-analytical model by
behaviour of the constituents and the microstructure of a composite Jensen [8] predicts the steady-state kinking stress. Recently, Skovs-
when investigating plastic microbuckling. In the early work on kink gaard and Jensen [9] developed an analytical model for steady-state
band formation, Argon [1] and Budiansky [2] identified that fibre kinking. Liu et al. [10] suggest that, for unidirectional fibre composites
misalignments and plastic shear deformation in the matrix govern the containing stress concentrations, e.g. holes and notches, a lower bound
critical stress. Christoffersen and Jensen [3] developed a homogenized for the critical stress can be estimated by the steady-state kinking stress.
constitutive model accounting for the microstructure of a unidirectional Structural applications of fibre composites may include components
fibre composite and allowing all cases of time-independent elastic- with open-holes. The ability to predict compressive failure of open-hole

∗
Corresponding author.
E-mail addresses: [email protected] (V. Tojaga), [email protected] (S.P.H. Skovsgaard), [email protected] (H.M. Jensen).

https://doi.org/10.1016/j.compositesb.2018.05.014
Received 14 March 2018; Received in revised form 1 May 2018; Accepted 9 May 2018
Available online 12 May 2018
1359-8368/ © 2018 Elsevier Ltd. All rights reserved.
134 Chapter 11 - C1

V. Tojaga et al. Composites Part B 149 (2018) 66–73

fibre composites is a problem of considerable importance to the Aero- t˙ij = Cijkl vl, k (1)
space Engineering community. It provides useful information about
how a composite performs in an open-hole application and how re- relates the nominal stress rates ṫij to the velocity gradients vi, j through
sistant it is to defects from fabrication or in-service. Open-hole com- the nominal moduli Cijkl . This paper adopts the index notation, the
pression tests on fibre composites suggest that the dominant mechanism summation convention and Cartesian coordinates. Latin indices, e.g.
of compressive failure is plastic microbuckling [11–20], provided that i, j, k , l , take values 1,2,3 and Greek indices, e.g. α, β , η , γ , take values

vk, k = 0 , the relation between the nominal moduli Cijkl and the in-
the matrix displays ductile material behaviour. Plastic microbuckling 1,2 . Neglecting the incremental volume change of a material, i.e.
starts progressively from the edges of a hole and propagates in a di-
rection normal to the loading direction. Modelling of compressive stantaneous moduli Lijkl is

Cijkl = Lijkl + 2 σik δjl − 2 σil δjk − 2 σlj δik − 2 σkj δil
failure interactions in an open-hole T300/914 laminate using a homo- 1 1 1 1
genized description of the material separated by interfaces reveals, in (2)
accordance with the latter experimental observations, that plastic mi-
where the instantaneous moduli Lijkl relate the Jaumann rate of the
crobuckling starts progressively from the edges of a hole [22]. Experi-
Cauchy stresses σ̂ij to the strain rates ε̇ij , i.e.

σˆij = Lijkl ε˙kl , ε˙ij = 2 (vi, j + vj, i )

mental studies by Soutis and Fleck [13], Soutis et al. [14–16] and Ahn
and Waas [28] suggest that plastic microbuckling, in the 0° plies, 1
(3)
governs the compressive failure of anopen-hole carbon/epoxy laminate.
Whitney and Nuismer [23] developed an average stress and a point and consequently, they satisfy the minor symmetries, i.e.

Lijkl = Lijlk = Ljikl

stress failure criterion for notched laminates. The point stress failure
criterion predicts failure when the stress at a distance d 0 away from the (4)
hole reaches the unnotched strength. The average stress failure cri-
terion predicts failure when the average stress over a distance a 0 The relation between the Jaumann rate of the Cauchy stresses σ̂ij
reaches the unnotched strength. The distance parameters a 0 and d 0 vary and the nominal stress rates ṫij is

σˆij = t˙ij − ωjk σik − ωik σkj − σij vk, k + σjk vi, k , ωij = 2 (vi, j − vj, i )
with hole size and stacking sequence among other factors [24]. After 1
observing that plastic microbuckling, in the 0° plies, governs the com- (5)
pressive failure of an open-hole laminate, Soutis et al. [15] developed a where ωij denote the spin.
cohesive zone model for predicting the open-hole compression strength.
In the cohesive zone model, plastic microbuckling initiates when a re-
mote compressive stress multiplied by a stress concentration factor 3. Constitutive models
reaches the unnotched strength. A crack replaces the microbuckled
zone and when the crack obtains a critical length such that the remote For convenience, the notation in Section 2 is adopted in the matrix
compressive stress attains a maximum value, catastrophic failure takes form

t˙αc = Cαβ
place. For the simplicity of application, many textbooks adopt the latter c
v ,cβ (6)
failure criteria; however, these criteria depend on an accurate mea-
surement of unnotched strength and compressive fracture toughness, equivalent to Eq. (1), where

v c = ⎡ 1c ⎤ , t˙1c = ⎡ 11
t˙ ⎤
t˙2c = ⎡ 21
t˙ ⎤
which highly depend on the microstructure. Ahn and Waas [17] defined

⎢ v2 ⎥ ⎢t˙12
c ⎥, ⎢t˙22
c ⎥
c c c
a global-local approach in which a micro-region around a hole is ana- v
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
lysed. Boundary conditions to be applied on the micro-region are ob-
(7)
tained from a global homogenous model. Extensions of this method are
validated and used by airframe manufacturers to predict the open-hole from which it follows that

= ⎡ 1111 ⎤ = ⎡ 1121 ⎤
compression strength of laminates [25-26]. These studies can capture

⎢C1211 c ⎥ , ⎢C1221 c ⎥,
Cc c
C1112 Cc c
C1122
⎣ ⎦ ⎣ ⎦
kink band formation, which is the dominant failure mechanism in c c
C11 c C12 c
compression, and damage in the off-axis plies within a laminate. C1212 C1222

= ⎡ 2111 ⎤ = ⎡ 2121 ⎤
Despite the many contributions, the micromechanics associated

⎢C2211 c ⎥ , ⎢C2221 c ⎥
Cc c
C2112 Cc c
C2122
⎣ ⎦ ⎣ ⎦
with kink band formation in open-hole fibre composites under com- c c
C21 c C22 c
pressive loading remains less understood. This paper aims to describe C2212 C2222 (8)
the micromechanics with the objective of developing a design metho-
dology, accounting for the nonlinear behaviour of the constituents and Superscript c refers to one of the constituents, i.e. fibre f or matrix
the microstructure of the composite, for designing of structural com- m . The x1- and x2 - axes are assumed to be parallel with and normal to
ponents with open-holes. It is inspired by [27]. To accomplish the latter the fibres prior to deformation, respectively. This section presents the
objective, the constitutive models in Refs. [3] and [5] are applied using constitutive models by Christoffersen and Jensen [3] and Jensen [5],
finite element analysis. The semi-analytical model in Ref. [8] enables a assuming perfect fibre/matrix bonding and no fibre/matrix bonding,
verification of the findings in Ref. [10]. respectively, in the form

t˙α = Cαβ v , β (9)

2. Constitutive relations
equivalent to Eq. (6). When superscripts are omitted, the quantities
The constitutive relation refer to the composite.

Fig. 1. Incremental deformation in a RVE of a uni-

directional fibre composite with perfect fibre/matrix
bonding.

67
11.5 Article C1 135

V. Tojaga et al. Composites Part B 149 (2018) 66–73

3.1. Constitutive model with perfect fibre/matrix bonding

Assuming displacement continuity in a representative volume ele-

ment (RVE) of a unidirectional fibre composite (Fig. 1), i.e.

v ,1f = v ,1m = v ,1 (10)

c f v ,2f + c mv ,2m = v ,2 (11)

c f + c m = 1, and overall equilibrium, i.e.

where the sum of the fibre and matrix volume fractions is unity, i.e.

c f t˙1f + c mt˙1m = t˙1 (12)

t˙2f = t˙2m = t˙2 (13)

By combination of Eqs. (6), (10)–(13), one obtains

v ,2f = (c mC22f + c f C22 − C21f ) v ,1 + C22

m −1 m m
) (c m (C21 v ,2) (14)

v ,2m = (c mC22f + c f C22 ) (c f (C21f − C21 ) v ,1 + C22f v ,2)

−1

plane strain and ratio R/ W = 0,1.

m m Fig. 3. Finite element mesh, with 1600 fully integrated biquadratic elements in
(15)

and the composite moduli

Cαβ = c f Cαβ + c mCαβ − c f c m (Cαf2 − Cαβ )(c mC22f + c f C22 ) (C2fβ − C2mβ )
f m m m −1

(16)

3.2. Constitutive model with no fibre/matrix bonding

In addition to Eqs. (10) and (12), assuming displacement continuity

between the matrix and the composite (Fig. 2), i.e. Fig. 4. Initial fibre orientation α .

v ,2m = v ,2 (17)
Table 1
and that the tractions normal to the fibres are not transmitted to the AS4/PEEK material parameters.
fibres, i.e. σ ym [MPa] Em [MPa] νm nm
σyf [MPa] E f [MPa] νf nf cf

ṫ2f = 0 (18) 4.140 214.000 0,263 1,9 0,6 82,1 6.140 0,356 4

such that the overall tractions ṫ2 , acting on the composite, are an
average of the local tractions in the matrix ṫ2m , i.e.

t˙2 = c mt˙2m (19)

we obtain

v ,2f = −(C22f ) C21f v ,1

−1
(20)

and the composite moduli

Cαβ = c f Cαβ + c mCαβ − c f Cαf2 (C22f ) C2fβ

−1
Fig. 5. Kink band geometry, with an inclination angle β . The fibres in the kink
f m
(21) band are rotated at an angle ϕ .

1) Load c f , σijc , Lijkl

c c
, Cαβ , Cαβ, σαβ from previous time step
4. Implementation 2) Calculate vα, β from

Sørensen et al. [4] implemented the constitutive model in vα, β = (c˙ αγ c˙ ζη Fγη
1
− Fαζ
0
) Fζβ0 −1 (22)
Subsection 3.1 as a user subroutine (UMAT) into ABAQUS. The current 1 0
implementation of the constitutive models in Section 3, in which a where Fαβ and Fαβ denote the deformation gradients at the end and in
the beginning of the increment, respectively, and ċαβ denotes the in-
superscript refers to the constituents and no superscript refers to the
cremental directional cosines. A rigid body rotation contributes to the
composite, contains the following steps:

Fig. 2. Incremental deformation in a RVE of a uni-

directional fibre composite with no fibre/matrix
bonding.

68
136 Chapter 11 - C1

V. Tojaga et al. Composites Part B 149 (2018) 66–73

m
Fig. 6. (a) Matrix shear stress σ12 . (b) Local fibre orientations. For comparison, the initial fibre orientation α is 0°.

1 0
components of the velocity gradient v1,1 and v2,2 unless Fαβ and Fαβ refer 5. Finite element analysis
to the same frame of reference; hence, the first expression in the par-
enthesis refers to the frame of reference in the beginning of the incre- The finite element model (Fig. 3) of an open-hole plate, which is
ment. simply supported along its left edge and subjected to a compressive
displacement d , incrementally, along its right edge, is implemented into
3) Calculate vαc, β from Eqs. (10), (14) and (15) or (10), (17), (20) ABAQUS. α (Fig. 4) denotes the initial fibre orientation. Plane strain
4) Update σαβ from Eq. (5) conditions apply; hence (Fig. 3), resembles a ply of an open-hole la-
5) Update σijc from Eq. (3) minate under compressive loading. The user subroutine (UMAT) loads
6) Update c f from Ref. [8] the AS4/PEEK material parameters (Table 1) from Kyriakides et al.

c˙ f = c f c m (v2,2 − v2,2
[28].
f m
) (23)
6. Localization of deformation
c
7) Update from the J2 flow theory of plasticity with isotropic
Lijkl
hardening, i.e. The semi-analytical model by Jensen [8] considers the kink band

rotation ϕ0 = 0∘ . In the procedure, we then prescribe a fibre rotation

geometry (Fig. 5), with an initial band inclination β0 and an initial fibre
= ⎡ 1 (δik δjl + δil δjk ) + − β c 2 E c / E c − (1t − 2ν c ) / 3 c 2 ⎤
Ec / E c − 1

⎣2 (σe ) ⎦
Ec νc sijc skl
c

1 + νc 1 − 2ν c ij kl
c 3
Lijkl δ δ
t increment ϕ̇ such that the fibres in the kink band rotate relative to the
fibres in the base material, i.e.

v2,1 = 0 , ′ = ϕ˙
(24)

where E c denotes the Young's modulus, ν c denotes the Poisson's ratio, v2,1

where superscript ′ refers to kink band coordinates.

(28)
Etc denotes the tangent modulus, sijc denotes the deviatoric part of the
Cauchy stress σijc , σec denotes the effective von Mises stress, i.e.
Continuity of velocities across the boundary between the base ma-

σec =
vα′, β qβ′ qα′ = vα, β qβ qα , vα′, β qβ′ nα′ = vα, β qβ nα
3 c c terial and the kink band requires that
s s
2 ij ij (25)
(29)
δij denotes the Kronecker delta and β c contains the conditions for plastic
band boundary in the base material coordinates, respectively, and nα′
where nα and qα denote the unit normal and the unit tangent to the
flow

1 for σec = (σec )max and σ˙ ec ≥ 0 and qα′ in the kink band coordinates, i.e.
βc = ⎧
⎨ (n1, n2) = (cos β , sin β ) , (q1, q2) = (−sin β , cos β ),
⎩ 0 for σe < (σe )max or σ˙ e < 0
(n1′, n2′) = (cos β′, sin β′) , (q1′, q2′) = (−sin β′, cos β′)
c c c
(26)

where β and β′ denote the inclination angle of the kink band relative to
(30)
The Ramberg-Osgood relation yields the tangent modulus

Etc =
Ec
nc − 1
the base material and the kink band, respectively, from which it follows
3 c ⎛ σec ⎞
+1
⎝ ⎠
n ⎜ c⎟

β˙ ′ = −vα′, β qβ′ nα′

that the band inclination is updated according to

β˙ = −vα, β qβ nα ,
7 σy (27)
(31)
where σ yc and nc denote the yield stress and the hardening index, re-
spectively. Continuity of traction rates across the boundary between the base

nα′ Cαβγη
′ vη′, γ nβ′ = nα Cαβγη vη, γ nβ , nα′ Cαβγη
′ vη′, γ qβ′ = nα Cαβγη vη, γ qβ
material and the kink band requires that
c
8) Update Cαβ from Eq. (2) followed by (8)
Update Cαβ from Eq. (16) or (21) (32)
Continuity of unit area A = A f / c f across the boundary between the
9)
10) Update Lαβηγ from Eq. (8) followed by (2)
11) Return Lαβηγ , σαβ and the state variables in step 1 to ABAQUS
two regions requires that

= nα′ eα′
′

steps 2 , 6 − 10 are new or modified compared to Sørensen et al. [4].

cf cf
′
The implementation of the constitutive model in Subsection 3.2 and nα eα
Af Af (33)

69
11.5 Article C1 137

V. Tojaga et al. Composites Part B 149 (2018) 66–73

m
Fig. 7. (a) Matrix shear stress σ12 . (b) Effective matrix von Mises stress σem . For comparison, the matrix yield stress σ ym is 82,1 MPa . (c) Tangent matrix modulus Etm . For
comparison, the Young's modulus E m is 6.140 MPa . (d) Local fibre orientations. For comparison, the initial fibre orientation α is 5°. (e) Longitudinal fibre stress σ11f .

where eα and eα′ denote a unit vector parallel with the fibres in the base underestimates the fibre rotation ϕ . In the present study, at steady-state,
material and the kink band, respectively. c f is the fibre volume fraction the work done per unit volume by the stresses in the kink band WI
in the base material from Eq. (23). Due to different stress states inside equals the work done per unit volume by the external loads WE , i.e.

′ WI = WE
and outside the kink band, Eq. (33) yields the fibre volume fraction in

the lock-up condition ϕ = 2β0 , neglecting the volumetric straining of

the kink band, denoted c f . At steady-state, many previous studies use (34)

the matrix between rigid fibres. Experimental studies by Poulsen et al. where
[29] and Vogler and Kyriakides [30] suggest that this lock-up condition

70
138 Chapter 11 - C1

V. Tojaga et al. Composites Part B 149 (2018) 66–73

WI = ∫ σαβ
ε′
′ dεαβ
′ , WE = Sαβ ΔEαβ

ε′ and ε denote the strain states inside and outside the kink band, re-
ε (35)

spectively, ΔEαβ denote the difference in Lagrangian strains inside and

outside the kink band expressed in the base material and Sαβ denote the

when WI = WE , the applied stress, denoted − σ (Fig. 5), is the steady-

work conjugated second Piola-Kirchhoff stresses. At steady-state, i.e.

state kinking stress, denoted − σs . Equations (28)–(35), together with

the constitutive model in Subsection 3.1 and the constitutive relations
in Sections 2 and 4, comprise an incremental procedure for determining
Fig. 8. Normalized applied stress − σ / G as a function of normalized applied
the steady-state kinking stress of the AS4/PEEK composite (Table 1).

elastic shear modulus of the composite G normalizes the applied stress − σ and
displacement d/ W from initial to critical stress for α = 5°, 10°, 20°, 40°. The
7. Results and discussion
the initial width of the plate W normalizes the applied displacement d .
Fibre misalignments together with plastic shear deformation in the
matrix govern the critical stress of a unidirectional fibre composite;
consequently, in Subsection 7.1, these factors are subsequently ana-
lysed at the critical stress. In Subsection 7.2, a suitable design metho-
dology of structural components with open-holes is investigated. Unless
otherwise stated, the constitutive model in Subsection 3.1 applies.

7.1. Micromechanics

For α=0°, the matrix shear stress σ12 m

(Fig. 6a) and the local fibre
orientations (Fig. 6b) resemble the numerical results by Fleck et al.
[20], where they used a constitutive model taking the fibre bending

Fig. 9. Normalized (converged) critical stress − σc / G as a function of normal-

rigidity into account. In the present case, the fibre bending rigidity is
negligible if the fibre diameter is much smaller than the size of the hole.
ized hole radius R/ W associated with perfect fibre/matrix bonding and full The carbon fibres used in Ref. [28] were 7 micrometres in diameter. For

the critical stress − σc and the initial width of the plate W normalizes the hole
decohesion for α = 5°. The elastic shear modulus of the composite G normalizes comparison, a typical clearance hole for a M10 bolt is 11 millimetres in
diameter.
radius R . m
For α = 5°, the matrix shear stress σ12 (Fig. 7a) is no longer sym-
metrical around the hole. The effective matrix von Mises stress σem
(Fig. 7b) indicates that the matrix yields at the locations where the
m
matrix shear stress σ12 is highest (Fig. 7a). Since the uniaxial stress-
strain relation follows the Ramberg-Osgood relation, i.e. Eq. (27), a
significant reduction in the tangent matrix modulus Etm (Fig. 7c) is seen
at the locations where the effective matrix von Mises stress σem is highest
(Fig. 7b). The local fibre orientations (Fig. 7d) show the formation of a
kink band due to an increase in matrix flexibility (Fig. 7c), suggesting
that plastic microbuckling propagates in a direction almost normal to
the loading direction corresponding to the experimental observations
[11–21]. In addition (Fig. 7d), shows elastic microbuckling in the re-
gions near the hole where the matrix does not yield (Fig. 7c). Wind
et al. [31] showed that the fibres do not rotate significantly from the
Fig. 10. Fibre rotation ϕ , at steady-state, as a function of initial band inclination initial to the critical stress when considering a unidirectional fibre
β0 . composite with fibre misalignments. This finding leads to the conclu-
sion that the deformation of the open-hole causes the fibres to rotate
severely near the hole and in the regions where they are less supported
by the matrix, making the open-hole a severe imperfection. The long-
itudinal fibre stress σ11f (Fig. 7e) indicates that the highest compressive
stresses are at the top and bottom edges of the hole where the fibres
rotate the most (Fig. 7d), suggesting that plastic microbuckling propa-
gates from the top and bottom edges of the hole corresponding to the
experimental observations [11–21].
In a laminate under compressive loading, the 0° plies are stiffer than
the off-axis plies and carry most of the load. Due to the stiffness pro-

prior to failure, as (Fig. 8), showing the applied stress − σ as a function

vided by the fibres, no severe deformation of these plies takes place

Fig. 11. Normalized steady-state kinking stress − σs / G as a function of initial

of the applied displacement d for various values of α , indicates. In the
off-axis plies, the fibres are more prone to rotate, causing the matrix to

the steady-state kinking stress − σs .

band inclination β0 . The elastic shear modulus of the composite G normalizes
shear in between the fibres, which (Fig. 8) also indicates. (Fig. 8)
suggests, in accordance with the experimental observations by Soutis
et al. [13–17], that plastic microbuckling, in the 0° plies, governs the
compressive failure. This observation leads to the conclusion that one

71
11.5 Article C1 139

V. Tojaga et al. Composites Part B 149 (2018) 66–73

Fig. 12. Cut-out open-hole composite, with radius R and

outer dimensions W .

should monitor the applied stresses acting on these plies to prevent an open-hole laminate. These findings lead to the conclusion that the
laminate failure, provided that the number of 0° plies is high in com- constitutive models (Section 3) are applicable to describe the micro-
parison with the number of off-axis plies. mechanics using a homogenized description of the composite. This
The constitutive model in Subsection 3.2 verifies these findings; homogenization takes the microstructure, including the fibre/matrix
however, the fibres are not supported laterally by the matrix in this bonding, and the nonlinear behaviour of the constituents into account.
model; hence, it provides a lower bound for the critical stress, with the A non-empirical approach is introduced for designing of structural

critical stress, denoted − σc , as a function of the hole radius R , illus-

situation in practice somewhere in between, as (Fig. 9), showing the components with open-holes. Using the mentioned homogenization to
describe the behaviour of the composite, the semi-analytical model
trates. The applied theory of plasticity ignores the dilatational stress, (Section 6) predicts, in accordance with experimental observations, that
which makes it suitable for fibres but less suitable for polymeric ma- the kink band lock-up condition at steady-state, used by many previous
terials; hence, it may overpredict matrix failure. The lower bound studies, underestimates the rotation of the kink band relative to the
(Fig. 9) compensates for this uncertainty. In this regard, it is empha- base material. Our results indicate that the steady-state kinking stress
sized that the constitutive models in Section 3 allow all cases of time- can be estimated as a lower bound, provided that the hole is small in
independent elastic-plastic behaviour of the constituents. comparison to the laminate. This design approach is suitable to prevent
global failure by kink banding of a laminate containing stress con-
centrations; however, to prevent local failure at the hole edges, the
7.2. Design
constitutive models in Section 3 predict the local compressive strength
using finite element analysis. The present work suggests a convenient
At steady state, (Fig. 10), showing the fibre rotation ϕ as a function
and useful predictive tool for design engineers.
experimental studies [29–30], that ϕ > 2β0 .
of the initial band inclination β0 , indicates, in accordance with the

(Fig. 11) shows the steady-state kinking stress − σs as a function of

Acknowledgements

state kinking stress − σs, min/ G = 0,059; where G denotes the elastic shear
the initial band inclination β0 , with the minimum normalized steady-
Thomas B. Thriges fund supported this work.
modulus of the composite. A comparison with (Fig. 9) indicates that a
References
lower bound for the critical stress of a unidirectional open-hole fibre
composite can be estimated by the minimum steady-state kinking
[1] Argon AS. Fracture of composites. Treatise Mater Sci Technol 1972;1:79–114.
stress, provided that the hole is small in proportion to the composite, in [2] Budiansky B. Micromech Comp Struct 1983;16:3–12.
which case one avoids global failure. Still, local failure, at the hole [3] Christoffersen J, Jensen HM. Kink band analysis accounting for the microstructure
edges, may take place. To avoid failure at the hole edges, the applied of fiber reinforced materials. Mech Mater 1996;24:305–15.
[4] Sørensen KD, Mikkelsen LP, Jensen HM. User subroutine for compressive failure of
stress, acting on a cut-out of the composite surrounding the hole composites. SIMULIA customer conference. London, May. 2009. p. 618–32.
(Fig. 12), must be within the range (Fig. 9). In this regard, the ratio R [5] Jensen HM. Models of failure in compression of layered materials. Mech Mater
W
must remain low, desirably close to 0,1; otherwise, the free edges, i.e. 1999;31:553–64.
[6] Skovsgaard SPH, Jensen HM. Three-dimensional constitutive model for elastic-
the top and bottom edges (Fig. 3), will interact to a higher degree with plastic behaviour of fibre-reinforced composites. Int J Solid Struct 2018:1–13.
the open-hole, in which case (Fig. 9) underestimates the local load [7] Skovsgaard SPH, Jensen HM. Constitutive model for imperfectly bonded fibre-re-
bearing capacity. If the number of 0° plies, in a laminate, is high in inforced composites. Compos Struct 2018;192:82–92.
[8] Jensen HM. Analysis of compressive failure of layered materials by kink band
comparison with the number of off-axis plies, these plies govern the
broadening. Int J Solid Struct 1999;36:3427–41.
compressive failure of the laminate (Subsection 7.1); hence, the latter [9] Skovsgaard, S.P.H., Jensen, H.M., Steady state kink band propagation in layered
design approaches also apply for laminates. materials, J Appl Mech 82, 061005.
(Fig. 10) and (Fig. 11) correspond to [8]. [10] Liu XH, Moran PM, Shih CF. The mechanics of compressive kinking in unidirec-
tional fiber reinforced ductile matrix composites. ComposPart B Eng
1996;27:553–60.
[11] Waas A, Babcock C. Observation of the initiation and progression of damage in
8. Conclusion compressively loaded composite plates containing a cutout NASA Progress Report,
Grant NSG-1483 1986
A non-empirical approach is introduced for determining the mi- [12] Guynn EG, Bradley WL, Elber W. Micromechanics of compression failure in open
hole composite laminates. STP1012. Compos Mater Fatig Fract 1989;2:118–36.
cromechanics of kink band formation in open-hole fibre composites [13] Soutis C, Fleck NA. Static compression failure of carbon fibre T800/924C composite
under compressive loading. Our results explain and predict, in ac- plate with a single hole. J Compos Mater 1990;24:536–58.
cordance with experimental observations, that a kink band propagates [14] Soutis C, Fleck NA, Curtis PT. Hole-hole interaction in carbon fibre/epoxy laminates
under uniaxial compression. Composites 1991;22:31–8.
from the edges of an open-hole in a direction almost perpendicular to [15] Soutis C, Fleck NA, Smith PA. Failure prediction technique for compression loaded
the loading direction and the 0° plies govern the compressive failure of

72
140 Chapter 11 - C1

V. Tojaga et al. Composites Part B 149 (2018) 66–73

carbon fibre-epoxy laminate with an open hole. J Compos Mater 1991;25:1476–98. and compression after impact strength on carbon fiber/epoxy laminates for the ares
[16] Soutis C, Curtis PT, Fleck NA. Compressive failure of notched carbon fibre com- 1 composite interstage. NASA/TP-2011-216460. 2011.
posites. Proc Math Phys Sci 1993;440:241–56. [25] Davidson P, Pineda EJ, Heinrich C, Waas AM. A unified model for predicting the
[17] Ahn J, Waas AM. Prediction of compressive failure in laminated composites at room open-hole tensile and compressive strengths of composite laminates for aerospace
and elevated temperature. AIAA J 2002;40:346–58. applications. 54th AIAA/ASME/ASCE/ AHS/ASC Structures, Structural Dynamics,
[18] Khamseh AR, Waas AM. Failure mechanisms of uniply composite plates with a and Materials Conferences. Boston, Massachusetts. April, 2013.
circular hole under static compressive loading. J Eng Mater Technol [26] Joseph APK, Waas AM, Wooseok J, Pineda EJ, Liguore S, Wanthal S. Progressive
1992;114:304–10. damage and failure prediction of open hole tension and open hole compression
[19] Waas AM, Junghyun A, Khamseh AR. Compressive failure of notched uniply com- specimens. 56th AIAA/ASME/ASCE/ AHS/ASC Structures, Structural Dynamics,
posite laminates. ComposPart B Eng 1998;29:75–80. and Materials Conferences. Kissimmee, Florida. January, 2015.
[20] Fleck NA, Liu D, Shu JY. Microbuckle initiation from a hole and from the free edge [27] Tojaga V. Methodology for modelling failure of open-hole fiber composites in
of a fibre composite. Int J Solid Struct 2000;37:2757–75. compression Master’s thesis Aarhus University; 2018.
[21] Suemasu H, Takahashi H, Ishikawa T. On failure mechanisms of composite lami- [28] Kyriakides S, Arseculeratne R, Perry EJ. On the compressive failure of fiber re-
nates with an open hole subjected to compressive load. Compos Sci Technol inforced composites. Int J Solid Struct 1995;32:689–738.
2006;66:634–41. [29] Poulsen JS, Moran PM, Shih CF, Byskov E. Kink band initiation and band width
[22] Allix O, Feld N, Baranger E, Guimard JM, Ha-Minh C. The compressive behavior of broadening in clear wood compressive loading. Mech Mater 1997;25:67–77.
composites including fiber kinking: modelling across the scales. Meccanica [30] Vogler TJ, Kyriakides S. Initiation and axial propagation of kink bands in fiber
2014;49:2571–86. composites. Acta Mater 1997;45:2443–54.
[23] Whitney JM, Nuismer RJ. Stress fracture criteria for laminated composites con- [31] Wind JL, Steffensen S, Jensen HM. Comparison of composite model and an in-
taining stress concentrations. J Compos Mater 1974;8:253–65. dividually fiber and matrix discretized model for kink band formation. Int J Non Lin
[24] Hodge AJ, Nettles AT, Jackson JR. Comparison of open-hole compression strength Mech 2014;67:319–25.

73
Appendix A

Jaumann rate of Cauchy stresses

From the definition of Cartesian tensors, a second order tensor Ai j transforms between
arbitrary coordinate system by
A′i j = Rik R jl Akl (A.1)

Where Ri j are components of a proper orthogonal rotation tensor. An infinitesimal line

element between to neighbouring points P and Q is denoted dx j and dx′j , when they are
observed from the two coordinate systems xi and xi′ , see Fig. A.1. The two systems coincide
at time t, so dx j and dx′j are identical at this instant. During a rigid body rotation, at instant
t = t + dt, the coordinate system xi stays stationary and the system xi′ takes part of the rotation.
The differnce dx j − dx′j observed from the two systems is equal to

dx j − dx′j = dv j dt (A.2)

where dv j denotes the relative velocity of point Q with respect to P. The relative velocity is

x2 , x2′ x2 Q x2 , x2′
Q x′ 2 t + dt dx j dv j dt
t
t dx′j
P P ′
x1
x1 , x1′ x1 x1 , x1′

Figure. A.1 Illustration of the concept of rotating coordinate systems.

142 Appendix A - Jaumann rate of Cauchy stresses

given by
∂ vi
dvi = dx j (A.3)
∂xj
The velocity gradient tensor ∂ vi /∂ x j can be divided into a symmetric and anty-symmetric
part
 
1 ∂ vi ∂ v j
ε̇i j = +
2 ∂ x j ∂ xi
  (A.4)
1 ∂ vi ∂ v j
ωi j = −
2 ∂ x j ∂ xi

Where ε̇i j is the strain rate tensor and ωi j is the spin tensor. When considering a rigid body
rotation the relative velocity Eq. (A.3) can be written as

dv j = ω ji dxi′ (A.5)

Combining Eqs. (A.2) and (A.5) yields

dx j − dx′j = ω ji dxi′ dt (A.6)

Which can be rewritten

dx j = (δi j + ω ji dt) dxi′ (A.7)

Since ωi j is an anti-symmetric tensor ω ji = −ωi j which leads to

dx j = (δi j − ωi j dt) dxi′ (A.8)

The expression in the parenthesis can be compared to the directional cosines given by
Eq. (A.1). Let σ̇i j and σ̊i j be the stress rates referred to the fixed xi and the rotating xi′
coordinate system respectively. Using Eq. (A.8) the stress components at the time t + dt of
the rotating coordinates becomes


σi j + σ̊i j dt = (δik − ωik dt) δ jl − ω jl dt (σkl + σ̇kl dt) (A.9)

By neglecting terms containing squares and cubes of dt, the equation can be expanded to

σi j + σ̊i j dt = δik δ jl σkl + δik δ jl σ̇kl dt − δik ω jl σkl dt − δ jl ωik σkl dt (A.10)
 143

Using the property, that Kronecker delta substitutes index, yields

σi j + σ̊i j dt = σi j + σ̇i j dt − ω jl σil dt − ωik σk j dt (A.11)

The equation can be simplified to give the stress rate in the rotating coordinate system

σ̊i j = σ̇i j − ω jk σik − ωik σk j (A.12)

In some literatures the relation is written in the equivalent form

σ̊i j = σ̇i j − ωik σk j + σik ωk j (A.13)

This stress rate is known as the co-rotational Jaumann-Zaremba stress rate and is often used
in constitutive relations since the stress rate is objective.
Appendix B

Nominal stresses

In this chapter the relation between the rate of nominal and Kirchhoff stresses is set-up.
The relative motion between two neighboring points can be described through the de-
formation gradient Fi j = ∂ xi /∂ X j where xi are components of the current coordinates of
material points and Xi are the reference coordinates. The inverse of the deformation gradient
Fi−1
j = ∂ Xi /∂ x j relates material lines of the reference configuration as a function of material
lines in the current configuration

∂ Xi
dXi = dx j = Fi−1
j dx j (B.1)
∂xj

Due to conservation af mass, the product between the density and the volume remains
constant
ρ ds j dx j = ρ0 dSi dXi (B.2)

where ρ and ρ0 are densities of material points in the current and reference configuration
respectively. dsi and dSi are infinitesimal surfaces in the current and reference configuration.
Combining Eqs. re f de fg rad and (B.2) yields

∂ Xi
ρ ds j dx j = ρ0 dSi dx j (B.3)
∂xj

Which can be rewritten  

ρ0 ∂ Xi
ds j − dSi dx j = 0 (B.4)
ρ ∂xj
146 Appendix B - Nominal stresses

Since dx j in general is non-zero, the equation embraced by the parenthesis must be zero,
which gives the Nanson’s formula

ρ0 ∂ Xi
ds j = dSi (B.5)
ρ ∂xj

Which relates surface vectors between the current and reference configuration.
To give a relation between Cauchy stresses σi j (true stress) and the nominal stresses
ti j an infinitesimal force d f j is introduced. The following relation between the stresses is
introduced
d f j = ti j dSi = σk j dsk (B.6)

Substitution of dsk using Eq. (B.5) yields

ρ0 ∂ Xi
ti j dSi = σk j dSi (B.7)
ρ ∂ xk

Since dSi is a common factor the relation between the nominal and Cauchy stresses is

ρ0 ∂ Xi
ti j = σk j = J Fik−1 σk j (B.8)
ρ ∂ xk

The relation between Cauchy and Kirchhoff stress is given by

τi j = J σi j (B.9)

Where J is the Jacobian, which represent the density ratio (J = ρ0 /ρ) from the initial to
the current configuration. The relationship between nominal stress and Kirchhoff stress is
thereby
∂ Xi
ti j = τk j = Fik−1 τk j (B.10)
∂ xk
To calculate the nominal stress rate the material derivative is utilised. The operator represent-
ing the material derivative of a spatial field is introduced

d(•) ∂ (•) ∂ (•)

= + vm = (•),t + vm (•),m (B.11)
dt ∂t ∂ xm

The first term in Eq. (B.11) denotes the spatial time derivative of the spatial field, while the
second term is the convective rate of change of the spatial field. By using the product rule of
 147

differentiation the material time derivate of Eq. (B.8) can be written as

 
d(ti j ) ∂ Xi d(τk j ) d ∂ Xi
= + τk j (B.12)
dt ∂ xk dt dt ∂ xk

Now focus is given to the material time derivative of the inverse of the deformation gradient.
By using the operator Eq. (B.11) the derivative can be written as
 
d ∂ Xi
= Xi,kt + vm Xi,km (B.13)
dt ∂ xk

A term vm,k Xi,m is added and subtracted from the expression

 
d ∂ Xi
= Xi,kt + vm Xi,km + vm,k Xi,m − vm,k Xi,m (B.14)
dt ∂ xk

Using the product rule of differentiation the equation can be rewritten

 
d ∂ Xi
= (Xi,t + vm Xi,m ),k − vm,k Xi,m (B.15)
dt ∂ xk

The term in the parenthesis is the material time derivative of the initial coordinates dXi /dt.
Since the initial coordinates do not change with time dXi /dt = 0. The equation can be
simplified into  
d ∂ Xi
= −vm,k Xi,m (B.16)
dt ∂ xk
By using Eq. (B.16) the stress rate can be written as

t˙i j = Fik−1 τ̇k j − vm,k Fim

−1
τk j (B.17)

This can be rewritten, using the fact that the repeated indices can be renamed, and that the
Kirchoff stress tensor is symmetric

−1


t˙i j = Fim τ̇m j − τ jk vm,k (B.18)

If the initial state coincide with the current state then ∂ Xi /∂ xm = δim .


t˙i j = δim τ̇m j − τ jk vm,k (B.19)

Kronecker delta renames index

t˙i j = τ̇i j − τ jk vi,k (B.20)
148 Appendix B - Nominal stresses

The Jaumann rate of Kirchhoff stresses is given by

τ̊i j = τ̇i j − ω jk τik − ωik τk j (B.21)

Combining Eqs. (B.20) and (B.21) yields

t˙i j = τ̊i j + ω jk τik + ωik τk j − τ jk vi,k (B.22)

Replacing the spin tensor ωi j using Eq. (3.3) yields

1
1

t˙i j = τ̊i j + v j,k − vk, j τik + vi,k − vk,i τk j − τ jk vi,k (B.23)
2 2
Expanding the parenthesis

1 1 1 1
t˙i j = τ̊i j + v j,k τik − vk, j τik + vi,k τk j − vk,i τk j − τ jk vi,k (B.24)
2 2 2 2
The last and the fourth term on the right have the same components since the Cauchy stress
tensor is symmetric the terms cancel out.

1 1 1 1
t˙i j = τ̊i j + v j,k τik − vk, j τik − vi,k τk j − vk,i τk j (B.25)
2 2 2 2
Reapeted index can be renamed

1 1 1 1
t˙i j = τ̊i j + v j,k τik − vl, j τil − vi,k τk j − vl,i τl j (B.26)
2 2 2 2
Kronecker Delta is added and indices are substituted
1 1 1 1
t˙i j = τ̊i j + vl,k τik δ jl − vl,k τil δ jk − vl,k τk j δil − vl,k τl j δik (B.27)
2 2 2 2
The Jaumann stress rate can be calculated using an incremental stiffness tensor

τ̊i j = Li jkl ε̇kl (B.28)

The incremental stiffness tensor must satify the symmetries

Li jkl = L jikl = Li jlk (B.29)

 149

Using the symmetries of Li jkl in Eq. (B.28) leads to

1
1

τ̊i j = Li jkl ε̇kl = Li jkl vk,l + vl,k = Li jkl vk,l + Li jkl vl,k
2 2 (B.30)
1

= Li jlk vl,k + Li jkl vl,k = Li jkl vl,k
2
Combining Eq. (B.30) and (B.27) yields

1 1 1 1
t˙i j = Li jkl vl,k + vl,k τik δ jl − vl,k τil δ jk − vl,k τk j δil − vl,k τl j δik (B.31)
2 2 2 2
Isolating vl,k yields
 
1 1 1 1
t˙i j = Li jkl + τik δ jl − τil δ jk − τk j δil − τl j δik vl,k (B.32)
2 2 2 2

By moving around terms, the fourth order tensor in the parenthesis can written as

1 1 1 1
Ci jkl = Li jkl − δil τk j − δik τl j − τil δk j + τik δl j (B.33)
2 2 2 2
Which is the tensor of nominal moduli relating nominal stress increments to velocity gradients
through the Kirchhoff stress tensor.
Appendix C

Constitutive formulation

In this appendix the derivations behind the constitutive model developed by Christoffersen
and Jensen (1996) is explained.
It is convenient to decompose the constitutive equations into a matrix representation for
simplicity in the derivation, where the rate of nominal stresses are collected in the vectors
! !
t˙11 t˙21
ṫ1 = ṫ2 = (C.1)
t˙12 t˙22

The velocity components are collected in the vector

!
v1
v= (C.2)
v2

The constitutive equation given by Eq. (3.26) can be written in the form

ṫα = Cαβ v,β α, β ⊂ {1, 2} (C.3)

The matrix system written in the full form

    
t˙11 C1111 C1112 C1121 C1122 v1,1
t˙  C C1212 C1221 C1222  v 
 12   1211   2,1 
 = (C.4)
t˙21  C2111
 
C2112 C2121 C2122  v1,2 
t˙22 C2211 C2212 C2221 C2222 v2,2
152 Appendix C - Constitutive formulation

The components of Cαβ represents the matrices

!
C1111 C1112
C11 =
C1211 C1212
!
C1121 C1122
C12 =
C1221 C1222
! (C.5)
C2111 C2112
C21 =
C2211 C2212
!
C2121 C2122
C22 =
C2221 C2222

Eq. (3.22) can be rewritten using the introduced matrix notation

f m
ṫ2 = ṫ2 = ṫ2 (C.6)

c
ṫ2 for the constituents can be calculated using Eq. (C.3)

c
ṫ2 = Cc2β vc,β = Cc21 vc,1 + Cc22 vc,2 (C.7)

Using the first assumption given by Eq. (3.17) vc,1 = v,1 leads to

c
ṫ2 = Cc21 v,1 + Cc22 vc,2 (C.8)

Using the latter equation with Eq. (C.6) yields the equality

f f
C21 v,1 + C22 v,2f = Cm m m
21 v,1 + C22 v,2 (C.9)

To find a relation between the velocity gradients of the constituents, Eq. (3.19) is used. The
equation (3.19) can be written using the matrix notation

c f v,2f + cm vm
,2 = v,2 (C.10)

Isolating v,2f yields

1
v,2f = v,2 − cm vm


f ,2 (C.11)
c
Multiplying by Cm
22 on both sides yields

f 1
Cm Cm m m m


22 v,2 = 22 v,2 − c C22 v,2 (C.12)
cf
 153

Cm m
22 v,2 can be isolated from Eq. (C.9), and combined with the latter equation, which yields

f 1  m 
f f

Cm
22 v,2 = f
C22 v,2 − cm C21 v,1 + C22 v,2f − Cm
21 v,1 (C.13)
c

Moving c f to the left and expanding the parenthesis yields

f m f m f f
c f Cm m m m
22 v,2 = C22 v,2 − c C21 v,1 − c C22 v,2 + c C21 v,1 (C.14)

Moving terms related to v,2f to the left and collecting terms related to v,β yields
   
f f f
cm C22 + c f Cm
22 v ,2 = cm
C m
21 − C m
21 v,1 + C22 v,2 (C.15)

Introducing the matrix C∗22

f
C∗22 = cm C22 + c f Cm
22 (C.16)

the velocity gradients v,2f for the fibre can be isolated using the latter term
   
v,2f = C∗−1
22 c m
C m
21 − C f
21 v,1 + C m
22 v,2 (C.17)

The equation can also be written in the form

 
v,2f = cm C∗−1
22 C m
21 − C f ∗−1 m
21 v,1 + C22 C22 v,2 (C.18)

In the same manner, the velocity gradients vm

,2 for the matrix can be determined as

 
f ∗−1 f ∗−1 f
vm
,2 = c C 22 C 21 − C m
21 v,1 + C22 C22 v,2 (C.19)

The terms in front of v,2 can be replaced. It is convenient to write the velocity gradients in
the equivalent form
   
v,2f = −cm C∗−122
f
C21 − Cm m ∗−1
21 v,1 − c C22
f
C22 − Cm22 v,2 + v,2
    (C.20)
f ∗−1 f f ∗−1 f
vm
,2 = c C 22 C 21 − C m
21 v,1 + c C 22 C 22 − C m
22 v,2 + v,2

In the latter equations the following identities were utilized

 
f
C∗−1 m m ∗−1
22 C22 v,2 = −c C22 C22 − Cm22 v,2 + v,2
  (C.21)
f f
C∗−1
22 C 22 v,2 = c f ∗−1
C 22 C 22 − C m
22 v,2 + v,2
154 Appendix C - Constitutive formulation

The first identity will be proven in the following. The terms with coefficients of v,2 can be
isolated in the right part of the equality
   
f
C∗−1
22 C m
22 v,2 = −cm ∗−1
C 22 C 22 − C m
22 + I v,2 (C.22)

Where I is the identity matrix. The identity matrix can be replaced by I = C∗−1 ∗
22 C22
   
f
C∗−1
22 C m
22 v ,2 = −c m ∗−1
C 22 C 22 − C m
22 + C ∗−1 ∗
22 C 22 v,2 (C.23)

The matrix C∗−1 ∗

22 can be isolated and put outside the parenthesis, and C22 can be expanded
using Eq. (C.16)
   
f m f
C∗−1
22 C m
v
22 ,2 = C ∗−1
22 −c m
C 22 − C m
22 + c f m
C 22 + c C 22 v,2 (C.24)

The inner parenthesis is expanded

 
f m f
C∗−1 m ∗−1
22 C22 v,2 = C22 −cm C22 + cm Cm f m
22 + c C22 + c C22 v,2 (C.25)

f
The two terms (cm C22 ) cancel out

C∗−1 m ∗−1
cm Cm f m


22 C22 v,2 = C22 22 + c C22 v,2 (C.26)

Cm
22 is isolated
C∗−1 m ∗−1
c m + c f Cm


22 C22 v,2 = C22 22 v,2 (C.27)

using the property that the sum of the volume fractions is unity c f + cm = 1 yields

C∗−1 m ∗−1 m
22 C22 v,2 = C22 C22 v,2 (C.28)

Similarly, the second identity in Eq. (C.21) can be proven. The nominal stress rate for the
c
components in the vector ṫ1 can be calculated using Eq. (C.3)

c
ṫ1 = Cc1β vc,β = Cc11 v,1 + Cc12 vc,2 (C.29)

f m
Using Eqs. (C.20) the nominal stress rate vectors ṫ1 and ṫ1 can be determined
     
f f f f f
ṫ1 =C11 v,1 + C12 −cm C∗−1
22 C 21 − C m
21 v,1 − cm ∗−1
C 22 C 22 − C m
22 v,2 + v,2
      (C.30)
m f ∗−1 f f ∗−1 f
ṫ1 =C11 v,1 + C12 c C22 C21 − C21 v,1 + c C22 C22 − Cm
m m m
22 v,2 + v,2
 155

Collecting terms related to v,β

     
f f f f f m f f
ṫ1 = C11 − cm C12 C∗−1
22 C 21 − C m
21 v ,1 + C 12 − c C C ∗−1
12 22 C 22 − C m
22 v,2
     
m f m ∗−1 f f m ∗−1 f
ṫ1 = Cm11 + c C12 C22 C21 − Cm
21 v,1 + Cm 12 + c C12 C22 C22 − Cm
22 v,2
(C.31)

For overall equilibrium of nominal stress rates, Eq. (3.25) was introduced which can be
rewritten using the matrix format

f m
ṫ1 = c f ṫ1 + cm ṫ1 (C.32)

Combining this equation with Eq. (C.31) yields

      
f f f f m f f
ṫ1 =c f C11 − cm C12 C∗−1
22 C 21 − C m
21 v,1 + C 12 − c C C ∗−1
12 22 C 22 − C m
22 v,2 +
      
f m ∗−1 f f m ∗−1 f
cm Cm11 + c C12 C22 C21 − Cm21 v,1 + Cm 12 + c C12 C22 C22 − Cm
22 v,2
(C.33)

Moving around terms, the equation can be simplified into

    
f f ∗−1 f
ṫ1 = c f C11 + cm Cm f m
11 − c c C12 − Cm
12 C22 C21 − Cm
21 v,1 +
     (C.34)
f f ∗−1 f
c f C12 + cm Cm f m
12 − c c C12 − Cm
12 C22 C22 − Cm
22 v,2

c
Next the focus is on the nominal stress rate vector ṫ2 for the constituents. Again the relation
given by Eq. (C.3) is used
c
ṫ2 = Cc21 v,1 + Cc22 vc,2 (C.35)

Inserting v,2f and vm

,2 from Eq. (C.20) yields the nominal stress rates

     
f f f f f
ṫ2 =C21 v,1 + C22 −cm C∗−1
22 C 21 − C m
21 v,1 − cm ∗−1
C 22 C 22 − C m
22 v,2 + v,2
      (C.36)
m f ∗−1 f f ∗−1 f
ṫ2 =C21 v,1 + C22 c C22 C21 − C21 v,1 + c C22 C22 − Cm
m m m
22 v,2 + v,2
156 Appendix C - Constitutive formulation

Collecting terms related to v,β entail the two alternative expressions for ṫ2 according to the
second assumption Eq. (C.6)
     
f f f f m f f
ṫ2 = C21 − cm C22 C∗−1
22 C 21 − C m
21 v,1 + C 22 − c C C ∗−1
22 22 C 22 − C m
22 v,2
     
f m ∗−1 f f m ∗−1 f
ṫ2 = Cm21 + c C22 C22 C21 − Cm21 v,1 + Cm 22 + c C22 C22 C22 − Cm22 v,2
(C.37)

The two expressions of ṫ2 give the same result based on the second assumption. It is of
interest to collect the two terms into one expression using the volume fractions, who obey
the property c f + cm = 1.
      
f f f f m f f
ṫ2 =c f C21 − cm C22 C∗−1
22 C 21 − C m
21 v,1 + C 22 − c C C ∗−1
22 22 C 22 − C m
22 v,2 +
      
f m ∗−1 f f m ∗−1 f
cm Cm21 + c C22 C22 C21 − Cm21 v,1 + Cm 22 + c C22 C22 C22 − Cm
22 v,2
(C.38)

Moving around terms, the equation can be simplified into

    
f f ∗−1 f
ṫ2 = c f C21 + c m Cm f m
21 − c c C22 − Cm
22 C22 C21 − Cm
21 v,1 +
     (C.39)
f f ∗−1 f
c f C22 + c m Cm f m
22 − c c C22 − Cm
22 C22 C22 − Cm
22 v,2

The matrix system may be written in the form given by Eq. (C.3), where the components of
Cαβ is
   
f f ∗−1 f
Cαβ = c f Cαβ + c m Cm
αβ − c f m
c C α2 − C m
α2 C 22 C 2β − C m
2β (C.40)

Based on the composite constitutive relation the nominal stress increments can be determined.
By returning to the original index notation used in the dissertation, the moduli given by
Eq. (C.40) can be written in the form introduced in Eq. (3.27). Going from the matrix notation
to the index notation, the components of the matrix C∗22 are equal to the components given
by Mαβ . Similarly, the components of C∗−1 22 are equal to the components of Hαβ .

Volume fraction increment

The fibre c f and matrix cm volume factions will change during deformation, due to the
difference in the constituent properties. It is chosen to determine the fibre volume fraction
increment ċ f . Based on this volume fraction increment, both volume fractions can be
 157

determined. The volume fractions for the constituents are based on

Vf Vm
cf = cm = (C.41)
V V
Where V is the total volume of a unit cell and V c are volumes of the constituents in the unit
cell. The volume increment for the constituents can be calculated using the velocity gradients

V̇ c = V c vck,k (C.42)

Which can be expanded using Eq. (C.41)

V̇ c = V cc vck,k (C.43)

The volume increment of a unit cell is a combination of the increments of both constituents
 
f
V̇ = V c f vk,k + cm vm
k,k (C.44)

The total fibre volume can be written as

V f = cf V (C.45)

The fibre volume increment can be determined using the product rule of differentiation

V̇ f = ċ f V + c f V̇ (C.46)

Isolating the term ċ f V yields

ċ f V = V̇ f − c f V̇ (C.47)

V̇ f and V̇ can be replaced using Eqs. (C.43) and (C.44)

 
f f
ċ f V = V c f vk,k − c f V c f vk,k + cm vm
k,k (C.48)

The volume V is a common factor and can be omitted

 
f f
ċ f = c f vk,k − c f c f vk,k + cm vm
k,k (C.49)

Multiplying the first part on the right by (cm + c f ) and expanding the parenthesis yields

f f f
ċ f = c f c f vk,k + c f cm vk,k − c f c f vk,k − c f cm vm
k,k (C.50)
158 Appendix C - Constitutive formulation

This can be simplified into  

f
ċ f = c f cm vk,k − vm
k,k (C.51)

Expanding the repeated indices yields

 
f f f
ċ f = c f cm v1,1 + v2,2 + v3,3 − vm m m
1,1 − v2,2 − v3,3 (C.52)

The constitutive model is two-dimensional and assumes plane strain conditions, thereby
vc3,3 = 0. The constitutive model is based on the assumption that material lines parallel with
f
the fibres (parallel with x1 axis) are subject to a common stretch, thereby v1,1 = vm
1,1 . The
latter equation can thereby be simplified into
 
f
ċ f = c f cm v2,2 − vm
2,2 (C.53)
Bibliography

Argon, A. S. (1972). Fracture of composites, volume 1. ACADEMIC PRESS, INC.

Ashby, M., Evans, T., Fleck, N., Hutchinson, J., Wadley, H., and Gibson, L. (2000). Metal
Foams: A Design Guide. Elsevier Science.
Attwood, J. P., Fleck, N. A., Wadley, H. N. G., and Deshpande, V. S. (2015). The compressive
response of ultra-high molecular weight polyethylene fibres and composites. International
Journal of Solids and Structures, 71:141–155.
Beuth, J. L. (1992). Cracking of thin bonded films in residual tension. International Journal
of Solids and Structures, 29(13):1657–1675.
Budiansky, B. (1983). Micromechanics. Computers & Structures, 16(1-4):3–12.
Budiansky, B. and Fleck, N. (1993). Compressive failure of fibre composites. Journal of the
Mechanics and Physics of Solids, 41(I):183–211.
Chater, E. and Hutchinson, J. W. (1984). On the Propagation of Bulges and Buckles. Journal
of Applied Mechanics.
Christoffersen, J. and Jensen, H. M. (1996). Kink band analysis accounting for the mi-
crostructure of fiber reinforced materials. Mechanics of Materials, 24(4):305–315.
Corneliussen, A. H. and Shield, R. T. (1961). Finite deformation of elastic membranes
with application to the stability of an inflated and extended tube. Archive for Rational
Mechanics and Analysis, 7(1):273–304.
Ericksen, J. L. (1975). Equilibrium of bars. Journal of Elasticity, 5(3):191–201.
Evans, A. G. and Adler, W. F. (1978). Kinking as a mode of structural degradation in carbon
fiber composites. Acta Metallurgica, 26(5):725–738.
Fleck, N. (1997). Compressive failure of fibre composites.
Fleck, N. A. and Budiansky, B. (1991). Compressive failure of fibre composites due to
microbuckling. Proc. 3rd Symp. on Inelastic Deformation of Composite Materials, pages
235–273.
Fleck, N. A. and Shu, J. Y. (1995). Microbuckle initiation in fibre composites : A finite
element study. Journal of the Mechanics and Physics of Solids, 43(12):1887–1918.
160 Bibliography

Guggenheim, E. A. (1959). Thermodynamics, Classical and Statistical, pages 1–118.

Springer Berlin Heidelberg, Berlin, Heidelberg.
Hsu, S.-Y., Vogler, T., and Kyriakides, S. (1999). On the axial propagation of kink bands
in fiber composites : Part II analysis. International Journal of Solids and Structures,
36:575–595.
Hsu, S.-Y., Vogler, T. J., and Kyriakides, S. (1998). Compressive Strength Predictions for
Fiber Composites. Journal of Applied Mechanics, 65(1):7.
Huang, Z.-M. (2018). On micromechanics approach to stiffness and strength of unidirectional
composites. Journal of Reinforced Plastics and Composites, page 073168441881193.
Hutchinson, J. W. and Koiter, W. T. (1970). Postbuckling theory. Applied mechanics reviews,
23(12):1353–1366.
Hutchinson, J. W. and Neale, K. W. (1978). Sheet Necking-II. Time-Independent Behavior.
Hutchinson, J. W. and Neale, K. W. (1983). Neck propagation. Journal of the Mechanics
and Physics of Solids, 31(5):405–426.
Hutchinson, J. W. and Suo, Z. (1991). Mixed Mode Cracking in Layered Materials. Advances
in Applied Mechanics.
Jäger, J., Sause, M. G., Burkert, F., Moosburger-Will, J., Greisel, M., and Horn, S. (2015).
Influence of plastic deformation on single-fiber push-out tests of carbon fiber reinforced
epoxy resin. Composites Part A: Applied Science and Manufacturing, 71:157–167.
Jensen, H. M. (1988). Collapse of hydrostatically loaded cylindrical shells. International
Journal of Solids and Structures, 24(1):51 – 64.
Jensen, H. M. (1999a). Analysis of compressive failure of layered materials by kink band
broadening. International Journal of Solids and Structures, 36(23):3427–3441.
Jensen, H. M. (1999b). Models of failure in compression of layered materials. Mechanics of
Materials, 31:553–564.
Jensen, H. M. and Christoffersen, J. (1997). Kink band formation in fiber reinforced materials.
Journal of the Mechanics and Physics of Solids, 45(7):1121–1136.
Jensen, H. M. and Thouless, M. D. (1995). Buckling Instability of Straight Edge Cracks.
Journal of Applied Mechanics, 62(3):620–625.
Ji, W., Waas, A. M., and Bazant, Z. P. (2013). On the Importance of Work-Conjugacy and
Objective Stress Rates in Finite Deformation Incremental Finite Element Analysis. Journal
of Applied Mechanics, 80(4):041024.
Jones, R. (1975). Mechanics of Composite Materials. International student edition. Scripta
Book Company.
Kyriakides, S. (1993). Propagating Instabilities in Structures. Advances in Applied Mechanics,
30:67 – 189.
Kyriakides, S. (1998). Propagating Instabilities in Solids. APS March Meeting Abstracts.
Bibliography 161

Kyriakides, S., Arseculeratne, R., Perry, E. J., and Liechti, K. M. (1995). On the compressive
failure of fiber reinforced composites. International Journal of Solids and Structures,
32(6-7):689–738.
Kyriakides, S. and Yu-Chung, C. (1991). The initiation and propagation of a localized
instability in an inflated elastic tube. International Journal of Solids and Structures,
27(9):1085 – 1111.
Lager, J. R. and June, R. R. (1969). Compressive Strength of Boron-Epoxy Composites.
Journal of Composite Materials, 6:48–56.
Liu, G., Thouless, M. D., Deshpande, V. S., and Fleck, N. A. (2014). Collapse of a composite
beam made from ultra high molecular-weight polyethylene fibres. Journal of the Mechanics
and Physics of Solids, 63(1):320–335.
Liu, X. H., Moran, P. M., and Shih, C. F. (1996). The mechanics of compressive kinking in
unidirectional fiber reinforced ductile matrix composites. Composites Part B: Engineering,
27(6 SPEC. ISS.):553–560.
Mesloh, R. E., Sorenson, J. E., and Atterbury, T. J. (1973). Buckling–and offshore pipelines.
Gas, 49(7):40 – 43.
Moran, P. M., Liu, X. H., and Shih, C. F. (1995). Kink band formation and band broadening in
fiber composites under compressive loading. Acta Metallurgica Et Materialia, 43(8):2943–
2958.
Naya, F., Herraez, M., Gonzalez, C., Lopes, C., der Veen, S. V., and Pons, F. (2017).
Computational micromechanics of fiber kinking in unidirectional FRP under different
environmental conditions. Composites Science and Technology, 144:26–35.
Nizolek, T., Begley, M., McCabe, R., Avallone, J., Mara, N., Beyerlein, I., and Pollock, T.
(2017). Strain fields induced by kink band propagation in Cu-Nb nanolaminate composites.
Acta Materialia, 133:303–315.
Palmer, A. C. and Martin, J. H. (1972). Buckle propagation in submarine pipelines. Nature,
254:46 – 48.
Paterson, M. S. and Weiss, L. E. (1966). Experimental deformation and folding in phyllite.
Bulletin of the Geological Society of America, 77(4):343–374.
Poulios, K. and Niordson, C. F. (2016). Homogenization of long fiber reinforced composites
including fiber bending effects. Journal of the Mechanics and Physics of Solids, 94:433–
452.
Prabhakar, P. and Waas, A. M. (2013). Interaction between kinking and splitting in the
compressive failure of unidirectional fiber reinforced laminated composites. Composite
Structures, 98:85–92.
Rice, J. R. (1976). The Localization of plastic deformation. Theoretical and Applied
Mechanics, pages 207–220.
Rosen, B. W. (1965). Mechanics of composite strengthening. Fiber Composite Seminar of
the American Society for Metals, pages 37–75.
162 Bibliography

Sammis, C. G. and Ashby, M. F. (1986). The failure of brittle porous solids under compressive
stress states. Acta Metallurgica, 34(3):511–526.
Schreyer, S. L. and Masur, E. F. (1966). Buckling of shallow arches. Journal of the
Engineering Mechanics Division, 92:1 – 20.
Shaffer, G. D. (2006). An Archaeomagnetic Study of a Wattle and Daub Building Collapse.
Journal of Field Archaeology.
Skovsgaard, S. P. H. and Jensen, H. M. (2018a). Constitutive model for imperfectly bonded
fibre-reinforced composites. Composite Structures, 192(February):82–92.
Skovsgaard, S. P. H. and Jensen, H. M. (2018b). International Journal of Solids and Structures
Three-dimensional constitutive model for elastic-plastic behaviour of fibre-reinforced
composites. International Journal of Solids and Structures, 139-140:150–162.
Skovsgaard, S. P. H. and Jensen, H. M. (2018c). Steady-State Kink Band Propagation in
Layered Materials. Journal of Applied Mechanics, 85(6):061005.
Skovsgaard, S. P. H. and Jensen, H. M. (2019). A general approach for the study of kink band
broadening in fibre composites and layered materials. European Journal of Mechanics -
A/Solids, 74(September 2018):394–402.
Slaughter, W. S., Fleck, N. A., and Budiansky, B. (1993). Compressive failure of fibre
composites: the roles of multi-axial loading and creep. Journal of Engineering Materials
and Technology, 115(3):308 – 313.
Sørensen, K. D., Mikkelsen, L. P., and Jensen, H. M. (2009). User subroutine for compressive
failure of composites. 2009 Simulia Customer Conference, (1965):618–632.
Sutcliffe, M. and Fleck, N. (1997). Microbuckle propagation in fibre composites. Acta
Materialia, 45(3):921–932.
Timoshenko, S. and Young, D. (1962). Elements of Strength of Materials. Fourth Edition. D.
Van Nostrand Company.
Tojaga, V., Skovsgaard, S. P. H., and Jensen, H. M. (2018). Micromechanics of kink band
formation in open-hole fi bre composites under compressive loading. Composites Part B,
149(March):66–73.
Vogler, T. and Kyriakides, S. (1997). Initiation and axial propagation of kink bands in fiber
composites. Acta Materialia, 45(6):2443–2454.
Wind, J. L., Steffensen, S., and Jensen, H. M. (2014). Comparison of a composite model and
an individually fiber and matrix discretized model for kink band formation. International
Journal of Non-Linear Mechanics, 67:319–325.
 PhD Dissertation
Author: Simon Peter Hald Skovsgård
Supervisor: Henrik Myhre Jensen
Department of Engineering
Aarhus University, Denmark