Fatigue Crack Propagation in

Functionally Graded Materials

A Thesis
By

Matthew T. Tilbrook

Submitted in partial fulfilment of

the requirements for
Doctor of Philosophy
School of Materials Science & Engineering, University of New South Wales

2005

Supervisor:

Mark Hoffman

Acknowledgements
A great debt of thanks is owed to all those who assisted in this project. In particular:
My supervisor, A/Prof. Mark Hoffman, for providing the opportunity to work on this
project, for being a great mentor and intellectual sparring partner, and for tolerating my
non-linear approach to both problems and schedules;
Dr Robert Moon, my co-supervisor, upon whom I could always rely for advice and
assistance, especially in laboratory and editing matters;
Lyndal Rutgers, my FGMs partner-in-crime, who paved the way in the lab and with
whom I shared many a fruitful discussion;

The technical staff who enabled things to happen in the labs: John Sharp, John Budden,
Brian Cooper, Yu Wang, Cathy Lau, Jane Gao.
The School of Materials Science and Engineering for administrative, facilities and
general support, particularly Chris, Lana, Flora, Julie, Owen and Alan.
Lab and office colleagues over the years: Pond, Pom, Richie, the Daves, Kong Meng,
Sven, Achim, Frieder, Peter, Rezan, Jacob, Singh.

My very kind hosts and collaborators in the US and Germany: Ivar Reimanis at Colorado
School of Mines, Eric Steffler at Idaho National Laboratory, Keith Rozenburg at
CSM/INL, Keith Bowman at Purdue University, Brian Cox at Rockwell Scientific,
Jurgen Rdel & group at TU Darmstadt, Achim Neubrand at Fraunhofer IWM Freiburg;
The organisers of the conferences I was fortunate enough to attend, and reviewers of
conference and journal papers for their valuable feedback, which has contributed to this
thesis;

The Australian Research Council for funding the project and providing APA scholarship;

And, with the utmost gratitude, my parents, family and friends, particularly my wonderful
girlfriend, Maarinke, for their support and encouragement.

iii

Abstract
Functionally graded materials (FGMs), those materials which exhibit a spatial variation
in composition and properties, have the potential to improve component performance in a
range of industrial, aerospace and biomedical applications. Despite the considerable
advances in understanding of functionally graded materials over the past decade, several
key areas require further work, in particular their fatigue behaviour and expected crack
propagation paths.

In the present study, propagation of cracks in FGMs under cyclic loading was
investigated via experiments and finite element (FE) analysis. Alumina-epoxy composites
with an interpenetrating-network structure and tailored spatial variation in composition
were produced via a multi-step infiltration technique. Compressed polyurethane foam
was infiltrated with alumina slip. After foam burn-out and sintering, epoxy was infiltrated
into the porous alumina body. Non-graded specimens with a range of compositions were
produced, and elastic properties and fatigue behaviour were characterised. An increase in
crack propagation resistance under cyclic loading was quantified via a novel analytical
approach. Results from homogeneous specimens were utilised for estimating spatial
property distribution and crack-extension effect in the graded specimens. Cracks were
initiated in graded specimens and propagated under cyclic four-point bend loading and
crack trajectories and growth rates were measured.

A simulation platform was developed with the commercial FE package ANSYS. Material
gradient was applied via nodal temperature definitions. Stress intensity factors were
calculated from nodal displacements near the crack-tip. Deflection criteria were
compared and the local symmetry criterion provided the most accurate and efficient
predictions. An automated mesh-redefinition algorithm enabled incremental simulation of
crack propagation. Effects of gradient and crack-geometry parameters on crack-tip
stresses were investigated, along with influences of crack-shape, crack-bridging, residual

iv

stresses and plasticity. The model provided predictions and data analysis for experimental
specimens.

Fatigue cracks in graded specimens deflected due to elastic property mismatch,

concordant with FE predictions. In other FGMs, thermal or plastic properties may
dominate deflection behaviour. Weaker step-interfaces influenced crack paths in some
specimens; otherwise effects of toughness variation and gradient steps on crack path were
negligible. Crack shape has an influence, but this is secondary to that of elastic gradient.
Cracks in FGM specimens initially experienced increase in fatigue resistance with crackextension followed by sudden decreases at step-interfaces. Bridging had a notable effect
on crack propagation resistance but not on crack path. Similarly, crack paths did not
differ between monotonic and cyclic loading, although crack-extension effects did.

Recommendations for analysis and optimisation strategies for other FGM systems are
given. Experimental characterization of FGMs is important, rather than relying on
theoretical models. Opportunities for optimization of graded structures are limited by the
properties of the constituent materials and resultant general crack deflection behaviour.

List of publications & presentations associated with this work

Journal Articles
1. Tilbrook MT, Moon RJ, Hoffman M. Crack Propagation in Graded Composites,
Compos. Sci. Technol. 2005; 65(2): 201-220.
2. Tilbrook MT, Moon RJ, Hoffman M. On the Mechanical Properties of AluminaEpoxy Composites with an Interpenetrating Network Structure. Mat. Sci. Engrg. A
2005; 393(1-2): 170-8.
3. Tilbrook MT, Moon RJ, Hoffman, M., Finite Element Analysis of Crack-Tip Stresses
and Crack Propagation in Functionally Graded Materials, Engrg. Fract. Mech. (in
publication) 2005.
4. Tilbrook MT, Reimanis IE, Hoffman M. Finite Element Simulations of Cracks near
Interfaces: Effects of Thermal, Elastic and Plastic Mismatch. J. Am. Ceram. Soc. (in
publication) 2005.
5. Tilbrook MT, Hoffman M. Approximation of Curved Cracks under Mixed-Mode
Loading. Int. J. Fract., 2005 (in review).
6. Tilbrook MT, Moon RJ, Hoffman M. Curved crack propagation in homogeneous and
graded materials. Fatigue Fract. Engrg. Mat. Struct. (in publication) 2005.
7. Tilbrook MT, Hoffman M. Crack-growth simulations with the local symmetry
criterion. Engrg. Fract. Mech., 2005 (in publication).
8. Tilbrook MT, Rutgers L, Moon RJ, Hoffman M. Fatigue crack propagation resistance
in homogeneous and graded alumina-epoxy composites. Int. J. Fatigue 2005 (in
review).
9. Tilbrook MT, Reimanis IE, Rozenburg K, M. Hoffman. Effects of plastic yielding on
crack propagation near ductile/brittle interfaces. Acta mater. (in publication) 2005.
10. Moon RJ, Tilbrook MT, Hoffman M, and Neubrand A. Al-Al2O3 Composites with
Interpenetrating Network Structures: Composite Modulus Estimation. J. Am. Ceram.
Soc., 2005; 88(3): 666-674.
11. Tilbrook MT, Rozenburg K, Steffler ED, Rutgers L, Hoffman M. Crack paths in
layered graded ceramic composites. Compos. B: Engrg. (in publication) 2005.
12. Rutgers L, Tilbrook MT, Moon RJ, Hoffman M. Effective Fracture Toughness of
Graded Alumina-Epoxy Composites. Acta materialia (in preparation) 2005.
vi

Conference Papers
13. Tilbrook MT, Moon RJ, Hoffman M. Crack Propagation in Graded AluminaPolyester Composites. Proceedings Int. Conf. Struct. Integr. Fract., Perth, 2002.
14. Tilbrook MT, Rutgers L, Moon RJ and Hoffman M. Fatigue Crack Propagation in
Graded Composites. Proc. Aust. Conf. Comp. Mat., Sydney, 2004.
15. Tilbrook MT, Rutgers L, Moon RJ and Hoffman M. Effective thermomechanical
properties of interpenetrating-structured composites. Proc. Aust. Conf. Comp. Mat.,
Sydney, 2004.
16. Tilbrook MT, Rutgers L, Moon RJ and Hoffman M. Fracture and Fatigue Crack
Propagation in Graded Composites. Proc. Multifunctional & Functionally Graded
Materials 2004, Leuven; Materials Science Forum 2005; 492-493: 573-580.
17. Tilbrook MT, Moon RJ, and Hoffman M. Propagation of curved cracks in
homogeneous and graded materials. Proc. Euro. Conf. Fract., Stockholm, 2004.
18. Tilbrook MT, Rutgers L, Moon RJ, and Hoffman M. Effective crack-propagation
resistance under monotonic and cyclic loading. Proc. Int. Conf. Struct. Integr. Fract.,
Brisbane, 2004.
19. Tilbrook MT and Hoffman M. Implementation of the local symmetry criterion for
crack-growth simulations. Proc. Int. Conf. Struct. Integr. Fract., Brisbane, 2004.
20. I. E. Reimanis, M. Tilbrook, K. Rozenburg, M. Hoffman. Effects of plasticity and
residual stress for cracks near interfaces. European Conference on Fracture, 2006.

Prize
Best student paper, Crack Propagation in Graded Alumina-Polyester Composites.
International Conference on Structural Integrity and Fracture, Perth, 2002.

Invited Presentations
Rockwell Scientific, Thousand Oaks, California, USA, June 2, 2004
School of Materials Engineering, Purdue University, Indiana, USA, June 29, 2004
School of Materials Science, Technical University of Darmstadt, Germany, July 20, 2004
Fraunhofer Institute for Material Mechanics, Freiburg, Germany, July 23, 2004

vii

List of symbols
*,

effective, auxiliary property

Paris law coefficient

crack length

ao

initial crack length

crack extension increment

ak

test-kink length

CTE

geometric length

exponential gradient parameter

compliance

compliance tensor

angular refinement width

Kronecker delta

error

Youngs modulus

corrected Youngs modulus

Etip

crack-tip Youngs modulus

E*

effective Youngs modulus

E1,E2 Youngs moduli

strain

angular function for stress

angular function for displacement

mechanical energy release rate

Gc

toughness (critical MERR)

gradient parameter

domain boundary of J-integral

height of specimen

loading angle

relative crack position

J-integral

Ko

intrinsic toughness (critical SIF)

Kbr

SIF due to bridging

Kappl

Applied SIF

stress intensity factor (SIF)

KI,KII,KIII

Kc

toughness (critical SIF)

SIF amplitude

kI,kII

kink-tip SIFs

KR

total critical SIF

Kr

residual stress SIF contribution

relaxation volume

characteristic gradient length

specimen length

property smoothing length

Paris law exponent

shear modulus

gradient profile exponent

normal vector

number of cycles

crack angle

Poissons ratio

tip

crack-tip Poissons ratio

load

SIFs in Modes I, II and III

viii

ellipsoid polarisation tensor

p, px

bridging traction

load ratio

crack-growth resistance

RE

Youngs modulus ratio

rn

notch-root radius

rc

critical length scale

strain energy density

stress

residual stress

remanent strength

parallel stress

bridging stress

NT

notch-tip stress

specimen thickness

temperature

change in temperature

displacement

crack-sliding displacement

volume fraction

v1,v2

volume fractions of phase 1 and 2

gradient width

energy density

domain of J-integral

x1

position

geometry factor

ellipsoid shape factor

mode-mixity

Frequently-Used Abbreviations

FGM

functionally graded material

SIF

stress intensity factor

EMA

effective medium approximation

FEA

finite element analysis

MTS

maximum tangential stress

MERR

maximum energy release rate

IPN

interpenetrating network

INS

interpenetrating network structure

ix

TABLE of CONTENTS

Certificate of Originality

ii

Acknowledgements

iii

Abstract

iv

List of publications and presentations associated with this work

vi

List of symbols

viii

Table of Contents

1-1

Introduction
1.1
1.2
1.3
1.4
1.5

Functionally Graded Materials

Crack Propagation
Cracks in graded materials
Present Study
Thesis Outline

Literature Review
2.1

Graded Materials
2.1.1 Applications & Processing
2.1.2 Gradient Description
2.2
Composite Materials & Properties
2.2.1 IPN Composites
2.2.2 First- & Second-Order Bounds
2.2.3 Unit-cell Models
2.2.4 Effective Medium Approximation
2.2.5 Computational Approaches
2.2.6 Predictions
2.2.7 Heterogeneity
2.2.8 Materials containing Damage
2.3
Mechanics of Cracks
2.3.1 Stress Fields
2.3.2 Energetic Considerations
2.3.3 Fracture Mechanics Methods
2.3.4 Fracture Testing
2.3.5 Crack Deflection
2.3.6 Deviant Cracks
2.4
Crack Propagation Processes
2.4.1 Intrinsic Toughness
2.4.2 R-curve effects

1-2
1-2
1-4
1-7
1-8

2-1
2-1
2-2
2-5
2-9
2-10
2-12
2-14
2-15
2-18
2-19
2-20
2-21
2-22
2-23
2-27
2-29
2-32
2-34
2-38
2-41
2-41
2-44
x

2.4.3 Crack-Bridging
2.4.4 Fatigue
2.4.5 Bridging and fatigue
2.5
Cracks in Interfacial and Layered Materials
2.5.1 Stresses near interfaces
2.5.2 Cracks at interfaces
2.5.3 Cracks near interfaces
2.5.4 Cracks in layered materials
2.6
Cracks in Graded Materials
2.6.1 Stress singularity
2.6.2 Stress field asymmetry
2.6.3 Calculation of stress intensity factors
2.7
Crack propagation parallel to gradient
2.7.1 Stress intensity factor calculations
2.7.2 Intrinsic fracture toughness variation
2.7.3 Variation in crack-extension toughening
2.7.4 Residual stress distributions
2.7.5 Resultant fracture behaviour
2.8
Crack propagation perpendicular to gradient
2.8.1 Stress intensity factor calculations
2.8.2 Crack deflection
2.8.3 Resultant fracture behaviour
2.8.4 Crack propagation path
2.9
Fatigue in graded materials
2.10 Final remarks

Investigation
3.1
3.2
3.3
3.4

Outstanding issues
Hypotheses
Approach
Framework

Procedures
4.1

Sample processing
4.1.1 Foam preparation
4.1.2 Slip-casting & drying
4.1.3 Foam burn-out & sintering
4.1.4 Polymer infiltration
4.1.5 Grinding
4.1.6 Polishing
4.1.7 Cutting
4.1.8 Other specimens
4.2
Microstructural analysis

2-46
2-52
2-55
2-58
2-60
2-61
2-63
2-66
2-70
2-71
2-72
2-73
2-77
2-77
2-81
2-82
2-84
2-85
2-88
2-89
2-93
2-94
2-96
2-98
2-100

3-1
3-1
3-3
3-4
3-6

4-1
4-1
4-2
4-3
4-5
4-6
4-7
4-7
4-8
4-8
4-10
xi

4.3
4.4
4.5
4.6

Elastic property characterisation

Fatigue testing
Moir Interferometry
Computational Procedures

Computational Simulations: Methods & Validation

5.1
5.2

Methods
Modelling Issues
5.2.1 Application of material gradient
5.2.2 Calculation of fracture parameters
5.2.3 Simulation of crack propagation
5.3
Validation of simulation methods
5.4
Implementation of the local symmetry criterion
5.4.1 Stress intensity factors for kinked cracks
5.4.2 Finite element implementation
5.4.3 Crack growth simulations
5.4.4 Further remarks
5.5
Experimental specimens: Simulations & data analysis

Simulation Results: Effect of Elastic Gradient

6.1

6.2
6.3
6.4
6.5
7

Straight cracks: Fracture parameters

6.1.1 Effect of crack position
6.1.2 Stepped & continuous gradients
6.1.3 Effects of gradient steepness & shape
6.1.4 Effect of crack length
6.1.5 Generalised gradient effect
Propagating cracks
Toughness influences
Prior to initiation
Summary

Simulation Results: Effect of Crack Shape

7.1
Quantifying crack shape
7.2
Analytical model
7.2.1 Mechanical energy release rate
7.2.2 Deflection angle
7.3
FE Simulations
7.4
Results: Curved cracks in homogeneous materials
7.5
Results: Curved cracks in graded materials
7.6
Discussion
7.7
Summary

4-15
4-16
4-20
4-21

5-1
5-2
5-6
5-6
5-7
5-9
5-12
5-18
5-18
5-22
5-24
5-28
5-31

6-1
6-1
6-1
6-3
6-4
6-9
6-10
6-11
6-17
6-23
6-26
7-1
7-1
7-3
7-5
7-6
7-7
7-10
7-18
7-26
7-29

xii

Simulation Results: Non-linear effects

8.1
Simulation of crack-bridging
8.2
Crack-bridging & effective toughness
8.3
Crack-bridging & mode-mixity
8.3.1 Bridging tractions
8.3.2 Analytical solution
8.3.3 Computational solution
8.4
Crack-bridging & propagation path
8.5
Incorporating thermal & plastic mismatch
8.6
Results: Effect of thermal stresses
8.7
Results: Effects of plasticity
8.7.1 Moderate plastic mismatch
8.7.2 High plastic mismatch
8.7.3 Discussion
8.8
Thermal & plastic mismatch: Further remarks
8.9
Summary
Experimental Results: Homogeneous Composites
9.1
9.2
9.3
9.4
9.5
9.6

10

Experimental Results: Graded Specimens

10.1
10.2
10.3
10.4
10.5
10.6
10.7
10.8

11

Elastic properties
EMA for other IPN composites
Fatigue crack propagation
Fatigue analysis
R-curve analysis
Summary

Crack initiation
Crack propagation paths
Fatigue crack propagation
Comparison of specimens
Effective resistance to fatigue crack growth
Moir interferometry
Comparison with monotonic loading
Summary

Discussion
11.1
11.2

Hypotheses
Other FGM systems
11.2.1 Thermomechanical property gradient
11.2.2 Crack propagation resistance gradient
11.2.3 Effective property relations
11.2.4 Crack-extension effects

8-1
8-1
8-3
8-7
8-8
8-9
8-15
8-16
8-22
8-29
8-31
8-35
8-40
8-44
8-46
8-47
9-1
9-1
9-7
9-11
9-16
9-21
9-25

10-1
10-2
10-4
10-11
10-20
10-24
10-29
10-34
10-38

11-1
11-1
11-14
11-15
11-16
11-16
11-17
xiii

11.2.5 Crack path & shape

11.2.6 Interfaces
11.2.7 Monotonic & cyclic loading
11.2.8 Modelling of crack propagation
11.2.9 Analysis of FGMs
11.3 Optimisation of graded interfaces

12

Conclusions

11-17
11-18
11-18
11-19
11-20
11-21

12-1

References

R-1

Appendices:
A
EMA formulation
B
Sample ANSYS macro
C
Comparison with analytical model
D
Crack propagation in graded specimens: Additional FE results
E
Curved cracks in graded specimens: Additional FE results
F
Additional experimental results

A-1
A-6
A-25
A-26
A-33
A-35

xiv

1 Introduction
Throughout the course of history, advances in technology have been determined by the
capabilities of available materials, and constrained by their limitations. From the stone tools
of Palaeolithic man to the steel boilers of the Industrial Revolution, from the feather shafts
of medieval long-bow arrows to the advanced composite nose-cones of space-shuttles, the
engineering endeavours of humanity have utilised the materials of the time to their full
extent. All materials, however, when subjected to sufficiently extreme conditions, will fail.
Consequently, the quest for further understanding and improvement of failure resistance
has been a key focus in the science and engineering of materials.

The study of material failure has been primarily motivated by the increased technological
demands of the modern age. Investigations into fatigue, the progressive degradation of
materials under cyclic loading, developed from observations of catastrophic failures in
locomotives and factories [Whler, 1867]. The field of fracture mechanics, the analysis of
loading and response in the vicinity of a crack and the influence of these on further crack
growth, was initially motivated by failures of plane wings [Newman, 1998]. As the
methodologies for understanding failure of materials have developed, so have the materials
themselves, with failure resistance a key parameter in the design of new materials.

Structural analysis
& design
Kilometres

Metres

Graded
Materials

Composite
Materials

Millimetres

Micrometres

Material
design
Nanometres

Figure 1.1: Developing structures for improved resistance to failure involves a wide range of length
scales.

Engineering against failure has evolved into a broad field, spanning length scales from
kilometres to nanometres, as depicted in Figure 1.1, and embracing topics from mechanical
and civil engineering, to materials physics and chemistry [Berger et al., 2002]. Design at
the structural scale and choice of appropriate materials has been supplemented by advances
in structural analysis and monitoring, the development of new materials and combination of
existing materials into composites [Ashby & Brchet, 2003]. In this context, the
combination of structural design and materials design into tailored material structures,
1-1

incorporating spatial variation in material composition and properties, is a logical

progression.
1.1

Functionally Graded Materials

A key motivation for incorporating a spatial compositional variation into engineering

components is the potential reduction in interfacial stress concentration, and the
corresponding improvement in failure resistance, at joins between different materials [1].
At such an interface, mismatch of mechanical properties and the generally poor cohesive
affinity between different classes of material leads to limitations in strength and structural
reliability.

The concept of functionally graded materials (FGMs), those materials with a tailored
spatial variation in composition, originated in Japan in the 1980s [Niino et al., 1987] and
has attracted significant attention since then [Neubrand & Rdel, 1997]. A key motivation
has been the reduction in thermomechanical stresses at interfaces [Bahr et al., 2003] which
would be beneficial in applications ranging from turbine blades [Balke et al., 2001], rocketengines [Moro et al., 2002] and space-planes [Niino et al., 1987] to artificial bone implants
[Chu et al., 2001]. The application of functionally graded materials has been limited by a
lack of understanding of their resistance to failure, particularly under cyclic loading [Forth
et al., 2003]. Accordingly, further investigation of fatigue crack propagation is required [Xu
et al., 2003a]
1.2

Crack Propagation

The propagation of cracks, which can result in the gradual degradation or catastrophic
failure of components, occurs when the concentration of stresses near the crack tip
overcomes the resistance of the material in that region to failure. For a linear-elastic
material, the stresses at the crack tip may be expressed in terms of the stress intensity
factors, KI, KII and KIII, where the subscript refers to the mode of fracture: tension, shear
and torsion respectively [Williams, 1957, Barenblatt, 1962a, Erdogan, 2000].

ij =

K
2r

f ij

( )

( ) + 1i 1 j T

i,j = 1,2,3

(1.1)

1-2

where K ( = I,II,III) are the stress intensity factors (SIFs), r is the distance from the crack
tip, is the angle from the crack-tip axis in the x1 direction, is the Kronecker delta, T is
the transverse (T) stress, parallel to the crack-tip axis, ECT and CT are Youngs modulus
and Poissons ratio at the crack tip and fij()() and gi()() are angular functions [Anderson,
1995]. The stress intensity factors were related, by Irwin [1963], to the rate of strain energy
release with crack extension:
G=

1
2
2
2
(K I + K II + K III )
E'

(1.2)

For non-linear materials, modified stress intensity factors [Rice & Rosengren, 1968] or the
J-integral [Rice, 1968] may be used, depending on the extent of yielding around the crack
tip. Various theoretical and experimental methods are used to determine SIFs [Broek,
1991], including integral equation methods [Erdogan, 1983, Tada et al., 1986], finite
element method [Banks-Sills, 1991] and weight functions [Bueckner, 1970]. The
distribution of stresses around a crack is dependent on the crack and component geometry
and applied loading [Anderson, 1995]. Material inhomogeneity and residual stresses due to
thermal mismatch can also influence crack-tip stresses, so that cracks near material
interfaces exhibit different behaviour to those in homogeneous materials [Rice, 1988, He &
Hutchinson, 1989].

The critical value of SIF for crack growth to occur, KC, is known as the fracture toughness,
so the criterion for propagation for a crack loaded in tension (Mode I) is:
KI = Kc

(1.3)

Crack propagation can also occur at subcritical loadings however, particularly in reactive
environments or under cyclic loading. In the latter case, crack growth rate per cycle, da/dN,
is often related to the applied cyclic variation in stress intensity factor, K, by the Paris
law:
(da/dN) = A(K)m

(1.4)

where A and m are empirical constants. Resistance to crack growth can involve both
intrinsic and extrinsic mechanisms [Ritchie, 1999, Ritchie et al., 2000], including phasetransformations and crack-bridging [Mai & Lawn, 1987, Cox & Marshall, 1991]. These

1-3

may lead to a crack-extension-dependent toughening known as R-curve behaviour,

whereby the effective toughness, Kceff, increases with crack growth, a:
Kceff(a) = K0 + Kbr(a).

(1.5)

The contribution of crack-bridging to effective toughness, Kbr, is dictated by the relation

between traction stress, p, and crack-opening displacement, u [Cox & Marshall, 1991].

When a crack propagates, it may change direction due to asymmetry in loading at the crack
tip [Palaniswamy & Knauss, 1978]. This is referred to as mixed-mode loading and may be
quantified by the phase angle of the mode-mixity [Hayashi & Nemat-Nasser, 1981]:

K
= tan 1 II
KI

(1.6)

Numerous criteria have been used to predict crack extension direction under mixed-mode
loading [Qian & Fatemi, 1996], including the maximum tangential stress [Erdogan & Sih,
1963], maximum energy release rate [Nuismer, 1975], local symmetry [Goldstein &
Salganik, 1974] and crack-tip opening-displacement [Sutton et al., 2000] criteria. After
crack deflection, the crack shape can influence the SIFs at the new crack tip [Cotterell &
Rice, 1980]. The calculation of these for non-straight cracks can be involved [Chen, 1999],
and is generally limited to simple shapes [Lo, 1978, Noda et al., 1994]. The prediction of
propagation paths is generally achieved via finite element analysis [Bittencourt et al., 1996,
Bouchard et al., 2003].
1.3

Cracks in graded materials

Cracks in graded materials will be influenced by the spatial variation in mechanical

properties, which can be related to the compositional variation with appropriate effective
property relations [Reiter et al., 1997]. There are many approaches to the calculation of
effective properties of composites [Ashby & Jones, 1996, Hashin, 1983, Torquato et al.,
1999], from the very simple isostress & isostrain approximations, to more complicated
geometry-specific models [Tuchinskii, 1983a, Wegner & Gibson, 2000], generalised
bounds [Hashin & Shtrikmann, 1963] and the self-consistent, or effective medium,
approximation [Hill,1965, Kreher & Pompe, 1989, Eshelby, 1957].

1-4

Various theoretical studies on cracks in graded materials [Delale & Erdogan, 1983, Jin &
Noda, 1994, Eischen, 1987a, Konda & Erdogan, 1994] have shown that the crack-tip
stresses differ from those in homogeneous materials. Cracks oriented perpendicular to the
gradient experience asymmetrical loading, which would result in crack deflection [Gu &
Asaro, 1997a, 1997b], whereas those oriented parallel to the gradient experience
symmetrical loading and continue to propagate straight-ahead [Jin & Batra, 1996a]. These
findings have been verified with computational simulations [Marur & Tippur, 2000, Bleek
et al., 1998, Becker et al., 2001]. Numerical methods for analysis of cracks have been
successfully adapted to graded materials [Gu et al., 1999, Kim & Paulino, 2002, Fett &
Munz, 1997].

FGMs are characterised by a spatial variation in material composition and, consequently, in

properties. This influences the fracture behaviour in several ways: (1) crack-stress fields are
modified as compared with those in homogeneous materials, which alters stress intensity
factors and mode-mixities; (2) crack-growth resistance will vary during propagation due to
the spatial variation in material toughness and microstructure; and (3) residual stress
distributions may further modify crack-tip stresses.

The criteria for crack propagation should take into account the influences of material
property gradient [Erdogan, 1995b], along with crack geometry and applied loading, on the
applied stress intensity factor. Furthermore, it should include the effects of intrinsic
toughness variation [Jin & Batra, 1996a], crack-bridging [Cai & Bao, 1998] or other
extrinsic mechanisms, and residual stresses [Chapa-Cabrera & Reimanis, 2002a, Neubrand
et al., 2002] on the effective toughness.

Experiments on crack propagation in graded materials have tended to focus on cracks

parallel to the gradient, under monotonic [Chung et al., 2001, Moon et al., 2002] and cyclic
[Forth et al., 2003, Xu et al., 2003a] mechanical loading and cyclic thermal loading [Balke
et al., 2000, Hofinger et al., 1999]. Those experiments on cracks perpendicular to the
gradient have generally involved fast fracture [Rousseau & Tippur, 2000, Hoffman et al.,
Chapa-Cabrera et al., 2000, Chapa-Cabrera & Reimanis, 2002b] so that, although crack
1-5

deflection was observed, the variation in SIF and effective toughness across the gradient
could not be quantified. These are summarised in Figure 1.2.
(a)

Gradient
(b)

Gradient
(c)

(d)

Gradient out of page

Gradient
Figure 1.2: Crack propagation in graded materials. (a) stable propagation parallel to the gradient
under monotonic loading in a layered graded alumina-zirconia composite [Moon et al., 2002], (b)
deflection of a crack perpendicular to the gradient in step-graded CuW composite [Chapa-Cabrera
et al., 2000], (c) fatigue crack growth parallel to the gradient in SiC-alumina composite [Xu et al.,
2003a] and (d) surface cracking parallel to the gradient observed by Hofinger et al. [1999] in graded
thermal barrier coatings under cyclic thermal loading.

1-6

1.4

Present Study

The aim of this study was to investigate fatigue crack propagation in graded composite
materials, via experiments and theoretical simulation, to develop an understanding of the
process sufficient for prediction of crack growth and optimisation of structural reliability.
This was achieved through experimental investigation of crack propagation under cyclic
loading in graded alumina-epoxy composites, and finite element simulations of these
experimental specimens and other crack and gradient configurations. Particular issues
addressed include effective composite properties, spatial variation in elastic and fatigue
crack propagation resistance properties, crack-path shape, crack-bridging and thermal and
plastic mismatch effects.

This work was part of a research group project on cracks in graded interfaces, which aimed
to determine the effects of elastic property mismatch and crack-bridging on propagation
path and component reliability. The project focussed on alumina-epoxy and aluminaaluminium graded composites. Properties of these constituent materials are given in Table
1.1. The present study builds on previous work on FGMs at UNSW [Kidson, 1999,
Hoffman et al., 2001, Amundsen, 2001] and was conducted in conjunction with parallel
investigations of other aspects of crack propagation in graded composites [Leck, 2002,
Rutgers, 2004, Rutgers et al., 2004, Moon, 2004, Moon et al., 2004]. Portions of this study
have been presented in journal and conference papers, which are listed on pages (vi) and
(vii), preceding this introduction.
Table 1.1 Experimentally determined constituent phase properties.

Material
Alumina
Epoxy
Aluminium

E [GPa]
397
3.40
69

0.23
0.35
0.33

[GPa]
153
1.26
26.3

B [GPa]
283
3.78
68.6

KC [MPa m]
4
0.8
n/a

f [%]
<1
>3
~20

1-7

1.5

Thesis Outline

This thesis is set out as follows. In Chapter 2, a full review of the literature is presented and
relevant terminology and theory are introduced. This expands on the introductory overview
given in the preceding sections, with particular focus on the influence of elastic property
gradient, crack-bridging, deflection and cyclic loading on crack propagation. The
outstanding issues are summarised in Chapter 3, then the hypotheses for the present study
are presented and the investigation framework is outlined. The experimental and
computational procedures employed in the study are described in Chapter 4.

In Chapter 5, the procedures used in the finite element simulation component of the project
are detailed, and the validation of these is discussed. The results of computational
investigations are presented in Chapters 6 to 8. Chapter 6 focuses on effects of material
gradient and crack-tip location on propagation path and crack-tip stress concentration. The
effect of curvature for cracks in both homogeneous and graded materials is addressed in
Chapter 7, whilst Chapter 8 examines the influences of non-linear behaviour, including
crack-bridging and plastic yielding, on crack propagation. Mechanical and fatigue
properties measured for homogeneous composite specimens are presented and compared
with theoretical models in Chapter 9. These results provide the link between material
composition, phase morphology and properties, which underpins the behaviour of graded
specimens. In Chapter 10, the results of crack propagation experiments on graded materials
are presented and compared to finite element predictions. Findings from experiments and
computational modelling are discussed together in Chapter 11, with a focus on their
implications for design of graded material components. Finally, conclusions are presented
in Chapter 12.

1-8

2 Literature Review
In this chapter, the field of graded materials is reviewed, along with pertinent aspects of the
theory of composites, fracture and fatigue. There are two aims in mind:
(i)

to introduce the terminology, notation and theory of composites and crack

propagation utilised in the present study; and

(ii)

to present an overview of preceding work on cracks in graded materials, and thereby

identifying the motivations and objectives of the present study.

Firstly, the concept of functionally graded materials (FGMs) is introduced. The relationship
between composition and properties for composite materials is then discussed, with a
particular focus on composites with interpenetrating network structures. The theory of
cracks and crack propagation is outlined, with particular attention paid to the issues of
mixed-mode loading, crack path, crack-bridging, fatigue and the effects of interfaces.
Theoretical and experimental work on propagation of cracks in graded materials is
reviewed in detail. Discussion is focussed on elucidating the effect of spatial variation in
mechanical properties and crack-growth resistance on crack propagation parallel and
perpendicular to the gradient direction. The information in this chapter, while nowhere near
exhaustive, is intended to provide a sufficiently extensive background for the work in the
present study.
2.1

Functionally Graded Materials

The combination of several materials in one component offers, in many cases, significant
improvements to its functional performance. Optimally, material properties throughout a
component should be tailored to its specific application, often requiring combinations of
properties that are unattainable with a single homogeneous material. Functionally graded
materials (FGMs) offer an advantageous means of combining materials, providing a spatial
variation in composition and properties, as an alternative to homogeneous materials and
bimaterial interface structures [Neubrand & Rdel, 1997].

2-1

2.1.1 Applications and Processing

The term Functionally Graded Materials was coined in the mid 80s in Japan where they
were developed for aerospace applications [Niino et al., 1987, Hirano et al., 1988].
Generally, FGMs are composites comprised of two or more types of material, although a
property gradient may be obtained by spatially-varied treatment of a single material [Li et
al., 2000]. FGMs may be used as joints between different types of materials, graded surface
coatings or simply as graded components, as shown in Figure 2.1. Examples of natural
graded materials include seashells [Nagata, 1999], sedimentary rocks and soils [Gibson,
1967], bone, teeth and wood [Nagata, 1999]. Material gradients have been utilised for some
time, most notably in carburisation of steels to produce a hard, brittle outer layer whilst
retaining a tough ductile core, and in shot-peening of surfaces, or cold-working, which
results in a work-hardening gradient [Elber, 1974, Lacarac et al., 2000]. More recently,
graded composites have become a focus for tailoring thermal, mechanical, electrical and
other properties for enhanced performance in structural, high temperature and other
specialised applications [Ferrari, 1999]. Furthermore, progressive degradation can lead to a
graded structure that, though not being tailored for functionality, may still be analysed in
the same manner as materials which are [Sevostianov et al., 2003].

(a)

(b)

(c)

Figure 2.1 Examples of functionally graded material geometries: (a) graded coating, (b) graded
joint and (c) entire graded component.

The principal motivation for FGM development is that a spatial variation in composition
and properties at a joint between two different materials has the potential to reduce the
stress at the joint as compared to a bimaterial interface. Additionally, there is an increased
interfacial strength at the joint and thus the likelihood of debonding is reduced. This is
illustrated in Figure 2.2 in the context of a potential application of FGMs, as a thermal
barrier coating for turbine blades [Raob et al., 1997]. In this situation, a heat resistant
2-2

ceramic must be strongly bonded to the metal blade to prevent spalling or surface cracking
due to thermally-induced stresses. A compositional gradient has been shown to reduce
thermal damage and delamination at metal-ceramic interfaces [Lee & Erdogan, 1995,
Hofinger et al., 1999, Nemat-Alla, 2003] and ceramic-ceramic interfaces [Lee et al., 2001].
This is relevant not only to turbine-blades [Balke et al., 2001], but also to fusion reactors
[Chapa-Cabrera & Reimanis, 1999, Pintsuk, 2003] and aerospace applications [Niino et al.,
1987, Moro et al., 2002, Nemat-Alla, 2003]. Other applications of graded materials include
surface coatings for tools and bearings [Suresh, 2001, Lengauer & Dreyer, 2002, Schulz et
al., 2003], artificial human joint replacements [Chu et al., 2001, Vitale Brovarone et al.,
2001, Lim et al., 2002, Pompe et al., 2003], optical fibres and lenses, and thermoelectric
and piezoelectric devices [Cherradi et al., 1994, Hirano et al., 1995, Muller et al., 2003].
(a)

Bimaterial Interface

(b)

Heat-Resistant
Ceramic Coating

Heat-Resistant
Ceramic Coating
Metal
Substrate

Graded Interface

Metal
Metal
Substrate
Substrate
Stress Discontinuity

Figure 2.2 Stress distributions resulting from thermal loading for thermal barrier coatings with (a)
bimaterial and (b) graded interfaces. A bimaterial interface produces a stress discontinuity, and may
also exhibit poor adhesion between materials. In a graded interface, there is less stress
concentration, and adhesion between phases is improved, thereby reducing the probability of
delamination.

FGMs have been successfully processed via a number of methods, though they are yet to
realise their full potential in applications. This is partially attributable to the difficulties
faced in bulk-manufacture of FGMs [Mortensen & Suresh, 1995, Ilschner, 1996].
Processing techniques include: physical and chemical vapour deposition [Movchan &
Yakovchuk, 2004, Araki et al., 1994]; sputtering [Govindarajan et al., 1996]; powder
metallurgy [Chapa-Cabrera & Reimanis, 2002b, Ruys et al., 2001]; chemical solution
deposition [Lim et al., 2002]; infiltration techniques in association with slip-casting
[Cichocki et al., 1998] or electrochemical processing [Jedamzik et al., 1999, 2000];
2-3

selective irradiation [Li et al., 2000]; buoyancy-assisted casting [Parameswaran & Shukla,
2000]; electrophoretic deposition [Put et al., 2003]; diffusion [Henager et al., 1995]; lasercladding [Jiang et al., 2004]; nitridisation [Knigshofer et al., 2004] and surface heattreatment [Forth et al., 2003].

A continuously varying gradient of properties is ideal. However, this is often difficult to

produce. It has been suggested that a number of discrete steps provides a satisfactory
approximation to a continuum in terms of alleviating stress concentration [Mortensen &
Suresh, 1995]. Lee et al. [2001] demonstrated distinct differences between FGMs with 3
and 20 layers. Shimojima et al. [1999] showed that a stepped gradient with just four steps,
and an optimised composition profile, could perform better, in terms of peak von Mises
equivalent thermal stress, than gradients with 20 steps with linear or exponential variation
in properties. Even if an approximately continuous variation in composition is attained, this
often results in discrete changes in microstructure and, hence, discrete changes in
properties. Work on graded materials may be seen as one aspect of a general thrust toward
more detailed control of component manufacturing [Chen and Feng, 2004], in cohort with
rapid-prototyping [Liu et al., 2004], stereolithography [Nee et al., 2001], and advanced
computer-aided design/manufacturing (CAD/CAM) technologies [Shin et al., 2003].

FGM research has tended to focus on several, often disparate, areas: processing,
thermomechanical behaviour and properties, high temperature performance, failure
behaviour, surface coatings, and biomedical, piezoelectric and thermoelectric applications
[Neubrand & Rdel, 1997, FGM 2000, FGM 1998, MSEA 2003]. Several review articles
have been written on FGMs, covering applications [Cherradi et al., 1994], processing
[Neubrand & Rdel, 1997, Mortensen & Suresh, 1995, Kieback et al., 2003], modelling
[Markworth et al., 1995, Hirano et al., 1995], contact [Suresh, 2001] and fracture [Erdogan,
1995a, 1995b, Bahr et al., 2003]. Analytical and numerical modelling of fracture behaviour
of FGMs has been fairly prolific [Erdogan, 1995b, Jin & Batra, 1996, 1998, Gu & Asaro,
1997, Bao & Wang, 1995, Marur & Tippur, 2000], whilst experiments have been more
limited [Chung et al., 2001, Chapa-Cabrera et al., 2000, Rousseau & Tippur, 2000]. Much
of the experimental work on mechanical and fracture behaviour has been conducted on
2-4

model FGM systems providing information that may be compared to modelling results or
generalised to other types of FGMs.

2.1.2 Gradient Description

A material gradient in FGMs may be described in a number of ways [Bao & Wang, 1995].
In this review, the gradient will be described by two characteristic features, as shown in
Figure 2.3. The scale refers to the total variation of properties over the interface width, w,
whilst shape refers to the form of the variation of properties with position (ie. linear,
logarithmic or exponential). These are global descriptors for the entire graded region. The
local gradient at a particular point will depend on the global scale and shape, as well as the
position within the graded region, which shall be described as a fraction of the total width,
= x/w.

Most commonly in theoretical models, the elastic properties (rather than composition) are
assumed to vary as a function of position, typically an exponential [Erdogan, 1995b, Jin &
Batra, 1996a, Delale & Erdogan, 1983, Konda & Erdogan, 1994], power law [Kassir, 1972]
or linear [Parameswaran & Shukla, 1999, Jain et al., 2004] function. An exponential
variation may be expressed as:
( x 1 ) = 0 e x1

or

E ( x 1 ) = E 0 e x1

(2.1a,b)

where and E are the shear and Youngs moduli which vary along the x1 axis, 0 and E0
are the respective modulus values at x1 = 0 and is the inhomogeneity constant. It should
be noted that an increase in the inhomogeneity constant, , as in Equation 2.1(a) or 2.1(b),
changes both the shape and scale of the gradient. Alternatively, a power law description is
commonly assumed:

E( x 1 ) = E 0 x 1

(2.2)

where n is the exponent which dictates the shape of the property profile.

2-5

Volume Fraction of B

Graded Region

n<1

Material 1

Material 2
n=1
n>1

x
w
Figure 2.3 Material property gradient showing effect of shape, n, and scale, d, descriptors upon
gradient steepness and shape.

Gradients with exponential and logarithmic forms are comparable to the curves shown in
Figure 2.3 for n > 1 and n < 1, respectively, whilst n = 1 corresponds to a linear gradient, as
shown in Figure 2.3. Alternatively, other models have assumed a simple variation of
composition, from which a property variation was calculated [Jin & Batra, 1996, Bao &
Wang, 1995]. More recently, models have been developed for FGMs with arbitrary spatial
property variation [Wang and Gross, 2000, Wang et al., 2002, Huang et al., 2003].

Typically, FGMs have been produced with a prescribed, usually linear, variation of
composition with position [Chapa-Cabrera et al., 2000], or a variation that depends on the
production technique [Parameswaran & Shukla, 2000, Butcher et al., 1999]. In either case,
the property variation tends not to mirror that of composition. If the spatial composition
profile is known, property variation may be predicted using one of a number of theoretical
mixing laws, as discussed in Section 2.2. Alternatively, property variation may be
determined directly from experiment. For FGMs, there are three ways to do this:
(i) producing and testing individual homogeneous specimens with a range of compositions
[Chung et al., 2001, Carrillo-Heian et al., 2001, Jedamzik et al., 1999];
(ii) cutting and testing small, effectively homogeneous, specimens from a larger graded
sample [Rousseau & Tippur, 2000, Butcher et al., 1999]; or
2-6

(iii) testing the graded material in situ via indentation or ultrasonic techniques
[Parameswaran & Shukla 2000, Krumova et al., 2001, Liu et al., 2001, Nakamura et al.,
2000].
The relationship between composition and properties is generally non-linear. As different
theoretical models may predict significantly differing property values, experimental
determination of composite properties has generally been recommended.

The variation of composition and properties with position across the gradient may be
continuous or stepped, as shown in Figure 2.4 for a graded composite of alumina and
epoxy. Continuous variation has been assumed in many analytical models [Erdogan 1995b,
Gu & Asaro, 1997, Jin & Batra, 1996], though stepwise variation is more easily produced
experimentally [Neubrand & Rdel, 1997]. FGMs tend to be composite materials, for
which effective properties may only be practically considered for length scales greater than
the representative volume element, which must itself exceed the scale of the internal
composite structure [Reiter et al., 1997]. Consequently, the effective composite properties
will vary in a more continuous manner than the composition. This is discussed further in
Section 2.2.6. The validity of effective property estimates near the crack-tip in graded
composites has been investigated and verified [Dolbow & Nadeau, 2002].

2-7

Composition [%Mat 1]

400

Composition - Continuous
Composition - Stepped

350

E* - Continuous
E* - Stepped

80

300
250

60

200
40

150
100

20

50
0
0

0.2

0.4

0.6

Position,

(b)

0.8

Composite Young's Modulus, E* [GPa]

(a)100

(c)
Material
1

Material
2

Material
1

Material
2

Figure 2.4: Comparison between continuous and stepped gradients: (a) variation of composition
and composite Youngs Modulus, E*, across a graded component, where is the normalized
position, given as a fraction of the distance across the gradient. For this example, Material 1 is
epoxy (E = 3.4 GPa, = 0.35) and Material 2 is alumina (E = 390 GPa, = 0.27). Schematic
illustrations of (b) continuous and (c) stepped gradients.

It should be noted that, throughout this work, the term homogeneous is considered to
mean that a material is macroscopically homogeneous. That is, it does not display any
macroscopic variation in composition, microstructure or properties. This includes nongraded composite materials, which are microscopically heterogeneous but statistically
homogeneous.

2-8

2.2

Composite materials and properties

The quest to obtain optimal material properties for specific applications, which requires a
compromise between functional properties, weight, durability and cost, has lead to the
combination of materials into composites. These usually exhibit properties that depend on
those of the constituent phases and on the internal geometry of the composite. Traditionally
composites have been produced with particle-matrix, fibre-matrix or laminated structures,
though interpenetrating-network structured composites, with two or more continuouslyconnected phases, have attracted attention, for their improved toughness and wear
properties [Clark, 1992, Knechtel et al., 1994, Sternitzke et al., 1996].

Underpinning composite design is an understanding of the relation between properties of

the constituent materials and the effective properties of the resulting composite material.
For a particular composite, this depends on the internal geometry and may be derived
theoretically [Hashin, 1983, Sendeckyj, 1974, Stellbrink, 1996, Kreher & Pompe, 1989,
Torquato, 1998] though experimental verification of effective properties is crucial to
accurately relate material composition and properties. The prediction of effective composite
properties has received significant interest dating back to the work of Maxwell [1873] and
Einstein [1906]. Mechanical properties have constituted the prime focus of theoretical
work, in particular, Youngs modulus, Poissons ratio and the related bulk and shear
moduli. Similar approaches have been applied to other linear phenomena such as thermal
expansion, heat capacity, dielectric permittivity, and piezoelectric and ferroelectric
behaviour. There has been some success in extending such models to nonlinear or flow
phenomena, such as plasticity, acoustic impedance, thermal and electrical conductivity.
Other properties, such as fracture toughness, wear resistance, magnetic hysteresis, and
fatigue resistance, involve a range of complex mechanisms and interactions which are not
easily captured with a single model.

Effective mechanical properties have been considered extensively for different types of
composites [Butcher et al., 1999, Torquato, 1999, Torquato, 1991, 1998, Stellbrink, 1996,
Sendeckyj, 1974, Hashin, 1983, Kreher & Pompe, 1989, Christensen 1990]. For these
2-9

models, predicted properties depend significantly upon the assumed composite structure.
Models tend to come in two types: those which specify the composite internal geometry
and those which do not. Similarly, models have been developed for prediction of effective
thermal [Stellbrink, 1994, Schapery, 1968], yielding [Tamura et al., 1973, Hashin, 1983,
Carpenter et al., 1999, Ueda, 2001, Feng et al., 2004], fracture strength [Li et al., 1999] and
toughness [Jin & Batra, 1996a, Krstic, 1983] properties. It should be noted that such
models, though convenient, tend to be limited in applicability and accuracy due to the
geometric and micromechanical assumptions upon which they are based.

They have been variously applied to FGMs [Aboudi et al., 1994, Dvorak & Zuiker, 1994,
Bruck & Rabin, 1999, Lele et al., 1999, Zhang et al., 2001, Zhai et al., 1999, Dolbow &
Nadeau, 2002, Suresh & Mortensen 1997, Ueda, 2001, Cho & Ha, 2000, Neubrand et al.,
2002, Jin and Dodds, 2004], with particularly useful reviews given by Gasik [1998] and
Reiter and co-workers [1997]. Interest in graded materials has motivated the development
of a number of semi-empirical models, based on micromechanics and image analysis [Cho
& Ha, 2000, Zhai et al., 1999, Dong et al., 1999]. As FGMs often span a wide range of
compositions across the gradient, and this can be accompanied by a significant variation in
microstructural morphology, a single mixing-law may be inappropriate, with the
intermediate (percolation) region being particularly problematic [Reiter et al., 1997,
Grujicic & Zhang, 1998, Van Siclen, 2003].

In this section, various models are considered for predicting effective elastic properties of
homogeneous and graded composites exhibiting an interpenetrating-network (IPN)
structure. For a more complete treatment of mixing laws for various properties of various
types of composite structure, see the work of Torquato et al. [1999], Reiter et al. [1997],
Suresh and Mortensen [1997], Hashin [1983] or Dvorak [2000].

2.2.1 IPN Composites

Composites with an interpenetrating-network structure are of particular interest as they
have been found to demonstrate higher strength, toughness and wear resistance when
compared with other composite structures [Clark, 1992, Prielipp et al., 1995, Sternitzke et
2-10

al., 1996]. Traditionally, composite materials tended to display a fibre/matrix, layered,

laminar or particle/matrix type structure. Correspondingly, theoretical models based upon
such geometries have proliferated [Hashin, 1983, Sendeckyj, 1974, Stellbrink, 1996, Kreher
& Pompe, 1989, Torquato, 1998], whilst work on interpenetrating structures has been more
limited [Clark, 1992, Wegner & Gibson, 2000a, 2001b]. Composites with interpenetratingnetwork structures have been produced by various methods [Schwartzwalder & Sommers
1963, Clark, 1992, Lange et al, 1994, Jedamzik, 2000, Mattern et al., 2004]. They have
demonstrated superior mechanical performance [Knechtel et al., 1994, Rdel et al., 1994,
Garcia et al., 1998, Zhou et al., 1999, Kouzeli & Dunand, 2003, Neubrand, 2004] and
applicability for FGMs [Chung et al., 2001, Kieback et al., 2003]. This all suggest that
further work in the area of IPN will be useful.

Several generic models may potentially be applicable to interpenetrating network structured

composites, including the isostress- and isostrain-approximation models, Hashin-Shtrikman
bounds, and the effective medium approximation (EMA). In addition, several models have
been proposed for composites with two continuous phases with simple regular internal
geometries [Wegner & Gibson, 2000a, Tuchinskii, 1983].

The Hashin-Shtrikman bounds [Hashin & Shtrikman, 1963] were used by Torquato et al.
[1999] to successfully predict the mechanical properties of interpenetrating boron
carbide/aluminium composites. The EMA [Kreher & Pompe, 1989] was used by Hoffman
et al. [1999] to predict mechanical properties of aluminium-alumina interpenetrating
composites, although the validity of this was not confirmed experimentally. Tuchinskii
developed a simplified unit-cell model for composites with two continuous phases
[Tuchinskii, 1983]. This was observed by Jedamzik et al. [1999] and also by Schmauder et
al. [1999] to fit experimental results for interpenetrating structured tungsten/copper
composites (E1/E2 = ~3.7). There has been mixed success in using the Tuchinskii unit-cell
model for alumina-aluminium composites (E1/E2 = ~5.6) with interpenetrating structures.
Moduli were overestimated for alumina-rich composites (73 to 97 vol.% Al2O3) [Neubrand
et al., 2002], but accurately predicted by the lower bound for composites in the
~40 to 60 vol.% Al2O3 range [Daehn et al., 1996], and the 5 to 23 vol.% Al2O3, range [Peng
2-11

et al., 2001]. It is noted that though these models have been successful for co-continuous
composites, the ratio of constituent moduli in each case was relatively low, E1/E2 < 10,
compared with E1/E2 100 for alumina-epoxy composites in the present study.

2.2.2 First- and Second-Order Bounds

The simple isostress and isostrain relations are based on distributions of phases such that
each phase experiences the same stress, as in Figure 2.5(a) on page 2-12, or the same strain,
as in Figure 2.5(b) [Paul, 1960, Ashby & Jones, 1996]. These provide first-order upper and
lower bounds on effective Youngs modulus [Torquato et al., 1999], and are given as:
v1 v 2

+
E1 E 2

E * v1 E 1 + v 2 E 2

in which E is the Youngs modulus,

(2.3)

represents the effective composite property, v is

volume fraction and subscripts 1 and 2 denote each phase. Replacing Youngs modulus, E,
in these equations with bulk and shear moduli, B and , leads to the Voigt and Reuss model
expressions. These can be used to calculate Youngs modulus, E, and Poissons ratio, ,
using the following relationships:
9B* *
E = *
3B + *
*

E*
* = 2 * 1
G

(2.4, 2.5)

Similar linear and inverse-linear averaging approaches have been used to estimate effective
thermal, yielding and fracture properties [Hashin, 1983, Jin & Batra, 1996].

Without any information about the distribution of phases, the best possible bounds on
elastic properties are the second-order bounds derived by Hashin and Shtrikman [1963].
They utilised variation principles from classical mechanics to find the extremum situations
corresponding to minimum potential and complementary energies [Hashin, 1962,
Sokolnikoff, 1957]. From these, bounds for the bulk and shear modulus, denoted BHS and
HS, were obtained. For two-phase composites, these are as follows:
B

v 1 v 2 ( B 2 B1 ) 2
v1 v 2 ( B 2 B1 ) 2

B
B
~
~
HS
B + 43 1
B + 43 2

(2.6)

2-12

v1 v 2 ( 2 1 ) 2
v1 v 2 ( 2 1 ) 2
HS
~ + 9B1 + 8 1
~ + 9B 2 + 8 2

2
1

6(B1 + 2 1 )
6(B 2 + 2 2 )

(2.7)

x = v 2 x 1 + v1 x 2 .
where the following operations are used: x = v1 x 1 + v 2 x 2 and ~

(a)

(c)
Horizontally
adjacent
section
a2=za1

a2=za1

a1
(b)

Vertically
adjacent
section

a1

(e)
E*

(d)

a2=ca1
a
a2=ca1

a1

En

a1
Figure 2.5 Schematics of assumed internal geometries for mixing-law models: (a) isostress and (b)
isostrain approximations, unit-cell models of (c) Ravichandran [1995] for particle/matrix
composites and (d) Tuchinskii [1983a,b] for co-continuous composites and (e) effective medium
approximation (EMA).

2-13

2.2.3 Unit-cell Models

Ravichandran considered a three-dimensional periodic array of isolated cubic inclusions
surrounded by a continuous matrix phase, under uniaxial loading as shown in Figure 2.5(c)
[1995]. The relative side-length of the inclusion, z, is calculated from volume fraction of
inclusions, vi:
1
z =
v2

1/ 3

(2.8)

Ravichandran calculated the effective elastic properties in two different ways, and treated
the results as upper and lower bounds. In the first approach, the inclusion was initially
combined with the horizontally adjacent section of the matrix, and effective properties were
calculated using the isostrain relation. Then this was combined with the remaining matrix,
using the isostress relation. In the second approach, the inclusion was combined with the
vertically adjacent section of the matrix, using the isostress relation, and then this was
combined with the remaining matrix, using the isostrain relation. The first approach
resulted in higher predictions for Youngs modulus. Expressions were as follows:

E RU

[E E
=

E RL =

+ E1 (1 + z ) E1 (1 + z )
2

(E 2 E1 )z + E1 (1 + z )3

(zE E
2

+ E 1 (1 + z ) E1 + E 2 E1
2

(2.9)

(zE 2 + E1 )(1 + z )2

(2.10)

Ueda [2001] used a similar model to calculate elastic-plastic properties (yield stress and
tangent modulus) and thermal conductivity and expansion for FGMs with one
discontinuous and one continuous phase. The variation in composition across the gradient
was modelled by varying the size of the embedded phase.

Tuchinskii [1983a] considered a two-phase interpenetrating skeletal structure equivalent to

the unit-cell shown in Figure 2.5(d). Both phases are continuous. The relative dimension
shown in the figure, c, was calculated from the volume fraction:
v 2 = (3 2c )c 2

(2.11)

As in the Ravichandran model, two approaches were used. In the first approach, the unitcell was sectioned horizontally (relative to the vertical loading direction) and the effective
2-14

properties of each of the upper and lower sections were calculated by combining phases
with the isostrain relation. Then the upper and lower sections were combined using the
isostress relation. In the second approach, the unit-cell was sectioned vertically into four
columns and the effective properties of each of the columns sections were calculated by
combining phases with the isostress relation. Then the columns were combined using the
isostrain relation. The second approach resulted in higher predictions for Youngs modulus.
Expressions were as follows:

E2
(2 c )c
+

E1

E TU

1 c
= E1

E
1 c 2 + 2

E1

E TL

E
2 2 c(1 c )

E
E
2

= E 1 (1 c ) + 2 c 2 + 1

E
E1
c + 2 (1 c )

E1

2
c

(1 c )2

(2.12)

Predictions from this model showed fairly good agreement with experimental results from
Fe-Pb, Ti-Mg and Cu-W intermetallic composites [Tuchinskii, 1983a]. The approach was
extended to calculation of effective thermal expansion coefficients [Tuchinskii, 1983b]. A
further variation on this theme was proposed by Feng et al. [2003, 2004]. They calculated
effective elastic and plastic properties for INS composites with an arbitrary number of
phases, with possibly more than two co-continuous phases. For the case of two phases, both
co-continuous, their model reduces to that of Tuchinskii.

2.2.4 Effective Medium Approximation

For a material with a random grain structure or a matrix-inclusion topology with contacting
inclusions, the effective medium approximation method may be appropriate [Kreher &
Pompe, 1989]. This technique, based on Hills self-consistent model [Hill, 1965], treats the
composite material as being comprised of ellipsoidal inclusions of each phase, surrounded
by an effective medium with the same properties as the total composite [Budiansky, 1965,
Wu, 1966, Walpole, 1969]. This is depicted schematically in Figure 2.5(e).

2-15

The derivation utilises the (second-order) tensor formulation for stress and strain, which are
related by the (fourth-order) compliance tensor as:

= C

(2.13)

Considering a single ellipsoidal inclusion with compliance, C, in surrounding material with

effective compliance, C0, and a prescribed far-field strain, the local strain field is
determined via the results of Eshelby [1957]. This leads to a weighted mean for the
effective compliance tensor, C*, which involves the inclusion shape-dependent strain
relation tensor, T:

(C C*) = 0

(2.14)

where <> denotes an average over all phases weighted with respect to volume fraction. This
results in expressions that implicitly define the effective bulk and shear moduli:

(B B*) TB = 0

( *) T = 0

(2.15)

The strain relation tensors TB(i) and T(i) are determined from a matrix A which is defined in
terms of Eshelbys inclusion shape tensor P and shape functions f(), where = a/b is the
shape parameter describing the ellipsoid aspect ratio (prolate for < 1, spherical for = 1,
and oblate for > 1). These tensors and functions are given in Appendix A for
completeness. For spherical inclusions the strain relation tensors simplify considerably
leading to the simplified equations:
Bn B *
~ =0
Bn + B *

n *
~* = 0
n +

(2.16)

where
~
B* =

~* = 1 * 9B * +8 *

6
B * +2 *

(2.17)

By solving these self-consistently for B* and *, E* and * may be determined as in

Equations 2.4 and 2.5. In the present study, the EMA equations were solved using
MathCad. An example of the MathCad code is included in Appendix A.

The EMA may also be used to predict effective thermal strain:

2-16

C
C
*
~ * T * ~ * T = 0
C+C
C +C
From this, the effective thermal expansion coefficient may be obtained:
~
B* + B*
B
*
=
~ =0
*
B
B + B*

(2.18)

(2.19)

As well as composite thermoelastic properties, the EMA concept has been applied to
effective electrical [Stroud, 1998] and transport [Willis, 1977, Torquato & Hyun, 2001]
properties and to microcracked solids [Huang et al., 1994].

One drawback of the EMA is that, although the composite internal geometry is not
specified, phase-connectivity is implicitly assumed based on ellipsoid aspect ratio and
volume fraction. For spherical inclusions, for example, the less abundant phase (<50%
volume fraction) is implicitly assumed to be discontinuous. For spherical inclusions with E
= 0, ie. pores, the predicted composite modulus will drop to zero for 50% pore volume
fraction [Wu, 1966]. For INS composites, this assumption could lead to erroneous
predictions in some cases.

2-17

2.2.5 Computational Methods

Wegner and Gibson modelled interpenetrating network-structured two-phase composites
using finite element analysis [2000a]. They assumed one phase was comprised of slightly
overlapping spheres in a regular stacked array, whilst the other phase filled the interstitial
volume, as in Figure 2.6(a). They predicted a sharp drop-off in stiffness with increase in
volume fraction of the more compliant phase. Zhai et al. used digitised microstructural
images, as in Figure 2.6(b), as input to a finite-element model to calculate effective elastic
properties [1999]. Schmauder and co-workers have used a number of different approaches
to model heterogeneous materials [Dong et al., 1999, Schmauder et al., 1999]. They
considered the interconnectivity of phases in terms of the matricity which was determined
empirically from microstructure images, as in Figure 2.6(c) [Lele et al, 1999]. Various
others have used computational modelling of deformation mechanisms and composite
structures to obtain effective property estimates [Huang, 2001, Zhang et al., 2000].
(b)

(a)
Overlapping
stacked sphere
geometry

Triangular prism unit-cell

(c)

Sphere
Interstitial
matrix
Antisymmetry
axis
Lower half
eliminated

Upper half
modeled
Surface of
Antisymmetry

Figure 2.6 Schematics of assumed internal geometries for computational models of effective elastic
properties: (a) Wegner and Gibson [2000a], (b) Zhai et al. [1999], (c) Lele et al. [1999].

2-18

2.2.6 Predictions
Considering the various mixing laws in the context of the alumina-epoxy composite
system, the large difference in properties of the two constituents, as shown in Table 1.1 on
page 1.7, leads to widely varying predictions from different models. For instance, a
composite of 50% alumina (E = 390 GPa) and 50% epoxy (E = 4 GPa), the predictions
from Equations (2.3) and (2.4) are 202 GPa and 7.9 GPa respectively. These are clearly
extremely disparate and a more precise prediction is required. For a 50/50 composite of
alumina and aluminium (E = 70 GPa), the predicted values, 240 GPa and 132 GPa
respectively, differ less drastically though still significantly. An advantage of investigating
an alumina-epoxy composite system is that the increased variation, between different model
predictions, enables improved identification of the appropriate choice of model for
composites with interpenetrating network structure. Effective property predictions from the
models discussed in Sections 2.2.1 to 2.2.4 are shown in Table 2.1.

10% Epoxy
40% Epoxy
70% Epoxy

31.5
8.4
4.84

351
235
119

HashinShtrikman

Ravichandran

Tuchinskii

Upper Lower Upper Lower Upper Lower

320
170
72.2

56.9
13.5
6.36

346
210
93.4

307
181
85.5

343
193
75.4

258
130
55.3

EMA

Isostrain

Composition

Isostress

Table 2.1: Effective Youngs modulus values (in GPa) for alumina-epoxy composites predicted
with a range of mixing laws. Properties of alumina and epoxy are given in Table 1.1 on page 1.7. A
shape factor of 1 was used for the EMA predictions.

313
92.3
8.43

When the effective composite properties are calculated, the composite structure must be
averaged over a finite-sized region so that the fraction of each phase in the representative
volume element is the same as in the overall composite [Hashin, 1983]. Indeed the concept
of effective composite properties is only valid for length scales significantly greater than
the scale of the composite microstructure [Ostoja-Starzewski, 1998, Dolbow & Nadeau,
2002]. In an FGM comprised of several compositional steps, assuming the composite
structure is continuous across the steps, the variation in effective properties will in fact be
smoothed slightly, due to the finite size of the (hypothetical) representative volume element
2-19

used to calculate these effective properties. This is illustrated in Figure 2.7, which is a
modification of Figure 2.4(a). The sharp steps in the compositional variation are smoothed
due to the continuity of phases in the interface region.

Composition [%Mat 1]

400

Composition - Continuous
Composition - Stepped

350

E* - Continuous
E* - Stepped
E* - Stepped-averaged

80

300
250

60

200
40

150
100

20

50
0
0

0.2

0.4

0.6

Position,

0.8

Composite Young's Modulus, E* [GPa]

100

Figure 2.7: Example of composition and property distribution across a continuous-gradient and a
stepped-gradient FGM. A slight smoothing of properties is observed for the stepped gradient, due to
the finite size of the region over which the composite structure must be averaged to obtain an
effective homogenised property.

2.2.7 Heterogeneity
All the models described thus far, with the exception of certain computational models,
involve an averaging, or homogenisation. There is an underlying assumption that the real
heterogeneous material will behave in the same way as a hypothetical homogeneous
material with the same properties as the effective properties of the composite. This is
generally valid for elastic properties, in which the total strain is an average of the strains
across the composite. For fracture strength, however, the heterogeneous property
distribution may lead to areas of weakness which will limit the strength of the entire
composite [Fitzpatrick et al., 1999]. Accordingly, heterogeneity should be taken into
consideration when estimating such properties [Pindera et al., 2004]. Various approaches
2-20

have been used to achieve this [Ostoja-Starzewski, 1998, Mishnaevsky et al., 2000], and
some very interesting studies have related distribution of phases to stress distribution and
failure susceptibility for heterogeneous materials [Pyrz & Bochonek, 1998, Mishnaevsky,
2004]. Most of this is beyond the scope of the present study; it suffices to note that
heterogeneity can have a distinct influence on mechanical properties [Dvorak, 2000].

2.2.8 Materials containing Damage

When a material sustains internal damage, for instance, microcracks or internal
delaminations, the transfer of stresses through the material under mechanical loading will
be interrupted by the presence of the damage. This results in a degradation of the
mechanical properties [Budiansky, 1976]. The relation between the internal damage state,
and its development, and the effective mechanical properties has motivated the
development of the field of continuum damage mechanics [Ibijola, 2002, Krajcinovic,
2000]. Although initially inspired by creep-related damage in metals [Kachanov, 1957],
damage mechanics has been most successful in application to composite materials,
particularly under impact or cyclic loading [Wnuk, 1998, Lamon, 2001, Chaboche &
Maire, 2001, 2002, Le Pen et al., 2002]. The damaged material is treated as a continuum
with effective properties calculated through a homogenisation scheme [Fish & Yu, 2002,
Souchet, 2003], similar to the calculation of effective composite properties as discussed
above. The damage-state is treated as an internal thermodynamic variable and constitutive
relations are used to relate it to mechanical properties, and to predict its development. The
shortage of direct experimental verification represents a drawback for continuum damage
mechanics, though this will undoubtedly be addressed in future, as more advanced
techniques for internal imaging, micro- and nanomechanical testing and computational
analysis are developed.

2-21

2.3

Mechanics of cracks

Before discussing a suitable characterisation scheme for failure in graded materials, a brief
summary will first be given of fracture theory for homogeneous materials, and for
interfacial and layered materials. This section, and the following, introduce concepts of
fracture and fatigue in homogeneous materials. The concepts introduced here may be found
in more detail in a number of review papers [Erdogan, 1983, 2000, Newman, 1998,
Cotterell, 2002] and text-books [Atkins & Mai, 1985, Broek, 1991, Lawn, 1993, Anderson,
1995].

All materials, as a result of production processes or applied loads, contain microscopic

flaws from which fracture initiates.

When a load is applied, the flaws act as stress

concentrators, with sharper flaws producing a greater concentration of stress. This stress
leads to opening and growth of the flaws. For a particular applied stress, there is a critical
flaw size for which growth occurs. Once the flaw begins propagating, the stress
concentration, which is the driving force for crack growth, tends to increase under constant
applied load, thereby driving further crack propagation. The resistance of a material to
failure is the major limiting factor in terms of its load performance and reliability. This
varies significantly between different materials, depending on the amount of plastic
deformation that occurs during fracture.

This section focuses on the concentration of stresses around a crack under various
conditions, and the methods used to describe, calculate and measure this concentration;
essentially the geometrical, mechanical aspects of crack propagation. Section 2.4 addresses
the material aspects of crack propagation: the effects of the crack-tip stress concentration
on the material in that region, and the relation to crack-growth resistance. Separating the
issues of crack propagation in this manner reflects the view taken through much of the
development of fracture mechanics, and indeed in this current study, though naturally there
is some overlap.

2-22

2.3.1 Stress fields

Applying a load to an uncracked material sample, the stress field will be defined by the
loading configuration, the component shape and the local response of the material to stress.
The last of these may be broadly defined as linear or nonlinear, as illustrated in Figure 2.8.
Linear elastic deformation has received considerable attention [Love, 1927, Sokolnikoff,

Applied Stress,

1957, Timoshenko & Goodier, 1970, Boresi & Chong, 1987].

Linear (elastic)

Non-Linear (elastic-plastic)

Deformation strain,
Figure 2.8: Examples of linear and non-linear deformation of material in response to applied stress.

The presence of a crack represents a modification of the specimen geometry, so that the
transfer of stresses across the plane of the crack is greatly, if not completely, impaired.
Stresses must redirect around the crack, leading to a concentration of stresses at the crack
tip. This is depicted schematically in Figure 2.9 for an internal crack and an edge crack.
The exact form of the concentrated stress field at the crack tip depends on the stress-strain
response of the material there.

2-23

(a)

(b)

2a

Figure 2.9: Schematic of a) an internal crack and b) an edge crack. Stress lines are included for
illustrative purposes.

A solution for the stress field around the crack tip in a linear elastic material was presented
by Williams [1957] and others [Irwin, 1957, Barenblatt, 1962a]:
ij =

K
2r

f ij

( )

() + 1i 1 j T + A

2r w ij

( )

()

(2.20)

where i,j represent any of x,y and z, K is a coefficient with units Pam1/2, fij() encapsulates
the angular variation in stress, is the Kronecker delta, T is the transverse (T) stress,
parallel to the crack-tip axis, and r, are polar coordinates as defined in Figure 2.10. The
subscript refers to the mode, as shown in Figure 2.11 and discussed below.

Crack-opening displacements
r
2v(x)
2u(x)

Crack

Crack-tip stresses
rr
r

T
Parallel T-stress

Figure 2.10: Diagram of crack tip showing stresses and crack-opening displacements.

Close to the crack tip, all the terms except the leading term become negligible and it
becomes sufficient to describe the stress field as:

2-24

 ij =

K
2r

f ij ()

(2.21)

Approaching the crack tip, ie. r 0, the stress theoretically becomes infinite. K is a
measure of the magnitude of the singularity and was termed the stress intensity factor (SIF)
by Irwin [1957]. It is understood that in brittle materials crack propagation occurs when the
SIF becomes equal to a critical value, K = KC. The critical stress intensity factor, Kc, often
referred to as the fracture toughness, is material-dependent, and is discussed in detail in
Section 2.4. Various solutions have been given for SIFs for different configurations
[Erdogan, 1983, Tada et al., 1986, Broek, 1991].

As shown in Figure 2.11, a crack may be loaded in three different modes: tension, shear
(in-plane) or torsion (antiplane-shear); these have corresponding SIFs: KI, KII, KIII. Mode I
is the simplest to analyse and reproduce experimentally, hence the vast majority of work in
fracture mechanics is based on mode I loading. Furthermore, cracks generally tend to
deflect to propagate in Mode I.
(a)

(b)

(c)

Figure 2.11: Crack loading modes: I tension (opening), II shear (in-plane sliding), III torsion (outof-plane sliding).

Generally, the SIF is related to crack length and applied stress as:
K = Y a

(2.22)

where Y is a geometrical factor related to the crack and component shape and the load
conditions, is the applied stress, is included explicitly for historical reasons, and a is the
size of the flaw, as shown in Figure 2.9. When K is equal to Kc, fast fracture occurs. This
analysis assumes the material behaves elastically up to fracture, and is known as Linear
Elastic Fracture Mechanics (LEFM).

This is applicable for brittle materials such as

2-25

alumina and epoxy. Residual stresses, if they are present, will also contribute to the cracktip stress field, and can have a notable effect on structural reliability [Hill & Pantonin,
1998].

If the material is not perfectly elastic, non-linear deformation, or yielding, occurs at the
crack tip. This will change the stress distribution around crack tip, so that SIFs may become
invalid. The modified stress-field is dependent on the form of the non-linear stress-strain
relation. Solutions were presented by Hutchinson [1968] and by Rice and Rosengren [1968]
for power-law hardening materials. The so-called Hutchinson-Rice-Rosengren (HRR)
solution is applicable only for the case of small-scale yielding. This is referred to as ElasticPlastic Fracture Mechanics (EPFM). Crack-tip loading is described in terms of the Jintegral [Rice, 1968] which summarises the loading in all modes. From hereon, discussion
is focussed predominantly on linear-elastic fracture mechanics, though many of the
concepts introduced are also pertinent to elastic-plastic fracture.

Material displacements in the region of the crack tip are expressed in Cartesian coordinates
as:

ui

K (2 + 2) 8r ( )
g i ()
E

{,i} = {I,2},{II,1},{III,3}

(2.23)

where K ( = I,II) are the stress intensity factors (SIFs), r is the distance from the crack
tip, is the angle from the crack-tip axis in the x1 direction, E and are Youngs modulus
and Poissons ratio and gi()() are angular functions given, inter alia, by Anderson [1995].
The crack profile, or crack opening displacements (CODs), v(x) and u(x), are obtained from
the displacement field at = 0. Near the crack tip, the crack opening profile is parabolic, as
calculated by Irwin [1957]:

v( x ) =

KI
E'

8x

u(x) =

K II
E'

8x

(2.24)

where E is equal to E for plane stress loading, and equal to E/(1-2) for plane strain
conditions. These are included in Figure 2.10.

2-26

The parallel stress term, T, in Equation 2.20 can also be important [ODowd and Shih,
1991]. This is referred to as the T-stress and is the component of stress parallel to the
original crack tip. The T-stress, being the non-singular parallel term in the Williams cracktip stress-field expansion, can also influence critical load in brittle materials [Ayatollahi,
2002] and, in materials with yielding, it has an effect on plastic constraint [Larsson and
Carlsson, 1973, Rice, 1974]

2.3.2 Energetic considerations

As an alternative to considering that fracture occurs when the stresses near the crack tip
become sufficiently large to break the material, the process may be considered in terms of
energies. Crack advance causes a release of elastic potential energy, which must be equal to
or greater than the energy input required to break the bonds and form new fracture surfaces.
This approach was used by Griffith [1920]; using the results of Inglis [1913] for the stress
field around an elliptical hole he calculated the amount of energy released when the crack
extended, then assumed this must be equal to the surface energy of the new surfaces. The
rate of release of mechanical energy, U, with increase in crack area, A, is known as G, and
has units of J/m2:
G=

dU
= G c = 2
dA

(2.25)

where Gc is the critical energy release rate (also referred to as R), equivalent to twice the
surface energy, . It has since been established that much more energy is absorbed in
deforming the material close to the crack tip than creating the new surfaces, however the
underlying principles of Griffiths analysis stand.

A simplified approach has been presented by Ashby and Jones [1996] inter alia, based on
the assumption that stresses are relaxed within a certain volume around the crack, as shown
in Figure 2.12. For a section of material containing a crack under tensile loading, the
approach is based on the assumption, which is proportional to the square of the crack
length, a. The released elastic energy associated with the stress relaxation is proportional to
the volume and the square of the applied stress, .

2-27

ka
Relaxed
material

2a

Figure 2.12: Straight crack showing assumed relaxation volumes before and after crack extension
[Ashby and Jones, 1996].

This leads to an expression for the energy released, Urel:

U rel

k 2 a 2
=
2E '

(2.26)

where E is the effective Youngs modulus for plane stress (E = E) or plane strain (E =
E/(1-2)), and k describes the aspect ratio of the relaxation volume. The rate of energy
release, G, associated with a crack extension, dA = 2dl, may thus be determined as:
G=

dU rel
dU rel
=
dA
da

dA

da

k 2 a
2E '

(2.27)

The value for k, which dictates the magnitude of G, is determined from the shape of the
relaxation volume. It can be shown that k is 2, for a through-thickness straight crack in an
infinite medium. Although this is a very simplistic approach, the form of the expression for
mechanical energy release rate is correct. The reason for this is that the stress distribution,
though it is actually more complicated than assumed, still scales with the square of length.

The mechanical energy release rate, G, was related by Irwin to the stress intensity factors
[1957]:
G=

1
1+
2
2
2
(K I + K II ) +
K III
E'
E

(2.28)
2-28

where E is equal to E for plane stress, and to E/(1-2) for plane strain conditions. G has
units of Pam, or J/m2. From a mechanics point of view, SIF is related to crack opening
force whereas G is related to energy released in fracture. Fracture occurs when G is equal
to a critical value Gc, equal to Kc2/E.
An equivalent energy release rate for ductile materials is determined via the J-integral. This
represents an integration over all the work done in the sample when the crack extends, and
may be calculated by integrating around the crack tip, outside the region of non-linear (nonconservative) stress-strain behaviour [Rice, 1968]. The J-integral is comprised of terms for
internal strain energy and surface traction energy, and is evaluated along a path, o, around
the crack tip:
J = ( W1i ij
O

du j
dx 1

)n i d

(2.29)

where W is the strain energy density and ni is the outward vector normal to the crack path.
The J-integral may also be evaluated for linear-elastic fracture in which case it becomes
equal to G.

2.3.3 Fracture Mechanics Methods

There are various methods for calculating stress intensity factors using analytical and
computational models, and experimentally quantifying crack-tip stress fields and
determining critical stress intensity factors.

Typically the crack problem is described in terms of a system of integral equations with
appropriate boundary conditions. Stresses are related to a potential field [Westergaard,
1939, Muskhelishvili, 1953]. Integral equations are solved using transforms and numerical
methods. Stress intensity factor values are extracted by collocation from displacement or
stress values near the crack tip. For a discussion of this area, see the work of Erdogan and
coworkers [Erdogan, 1978, 1983, Erdogan et al., 1973].

The weight-function technique for estimating SIF was developed by Bueckner [1970].
Weight functions, which are determined from the CODs of a reference loading case, relate
2-29

applied stress distribution to SIF for particular configurations of load, specimen and crack
geometry. Petroski and Achenbach developed a useful method for calculating the weight
function without a particular known solution [1978]. The weight function, h, relates the rate
of crack-opening with crack extension at a particular point, x = x, and the SIF, at a
particular crack length, a = a:
h(x' , a ' ) =

E u
2K (a ' ) a

x=x'
a =a '

(2.30)

It may then be used to calculate SIF from the applied traction distribution, p(x):
a

K = p( x )h ( x , a )dx

(2.31)

A large amount of work has been done in this area by Fett and co-workers [Fett & Munz,
1992, 1995, Fett 1992, 1998, Fett et al., 1995, 2000c].

The finite element method (FEM) has been used extensively in fracture modeling [BanksSills, 1991, Zienkiewicz, Taylor, 2000]. The boundary element method [Aliabadi, 1997],
element-free Galerkin method [Belytschko et al., 1994] and other computational
formulations, though less common than FEM, are also used frequently for analysing
fracture problems. In finite element analyses of cracks, the stress-field at the crack tip must
be modeled appropriately. While specific crack-tip elements have been used [Barsoum,
1975], it has been shown that these are unnecessary [Henshell & Shaw, 1975]. By using
triangular elements at the crack tip with midside nodes offset to a quarter of the way from
the crack tip, the idealised 1/r singularity will be simulated correctly.

The fracture parameters (SIFs and G) may be calculated for homogeneous materials via
several different approaches [Kim & Paulino, 2002a]. These include correlation of crack-tip
displacement or stress field values with theoretical expressions, calculation, via the crack
closure integral or load-point compliance, of energy release associated with crack extension
and evaluation of the path-independent J-integral, or equivalent domain integral, around the
crack tip.

2-30

Stress intensity factors may be calculated from the crack-opening displacement profile near
the crack tip [Kim & Paulino, 2002a]:

uiE
(2 + 2)g i

()

()

8r

{,i} = {I,2},{II,1},{III,3}

(2.32)

Referred to as the displacement correlation technique (DCT), this is a straight-forward

method for calculating stress intensity factors, KI, KII and KIII. A similar approach is used to
calculate SIFs by correlation of nodal stress values near the crack tip. In displacement or
stress correlation methods, correlation points are required to be close to the crack tip, so
that the results are not influenced by higher-order terms, as in Equation 2.20. A possible
drawback is that high stress and strain gradients at the crack tip may not be accurately
simulated between nodal points, which may compromise results [Williamson et al., 1995].

The mechanical energy release rate may be calculated more directly from loads and
displacements near the crack tip, or at the applied loading points. In the first case, the crack
closure integral represents the work involved in closing the crack [Rybicki & Kanninen,
1977]. This has been modified for mixed-mode loading [Raju, 1987]. In the second case,
the value of G may be related to the change in compliance with increase in crack length:
G=

P 2 dC
2t da

(2.33)

The use of this method in a finite-element model can become complicated as the
application of point loads can lead to unrealistic deformations near the load points, which
would prevent accurate calculation of compliance. This may be avoided by applying loads
via contact. This is more physically realistic but can be computationally demanding.

The J-integral may be calculated from finite-element data either directly, or by converting it
to a domain integral, which can be more convenient for FE analysis [Raju & Shivakumar,
1990]:

du j
dW
d
J=

( ij
) d
O dx
dx 1
1 dx i

(2.34)

where o is the domain contained within path o. The path and domain are schematically
depicted in Figure 2.13.
2-31

Displacement
correlation points

J-Integral
Path
J-Integral
Domain

Crack Tip

Figure 2.13: J-integral path and domain, showing displacement correlation points also.

Various techniques have been used to extract T-stress values from analytical and
computational models. These include the use of interaction integrals based on Bettis
reciprocal theorem [Nakamura & Parks, 1992, Sladek & Sladek, 1997, Chen et al., 2001] or
on the Eshelby energy-momentum tensor [Kfouri, 1986], relation to J-integral [Moon &
Earmme, 1998], weight functions [Sham, 1991, Fett, 1998], and direct extraction from
nodal stresses [Chen, 2000, Yang & Ravi-Chandar, 1999] or displacements [Tan & Wang,
2003], sometimes involving multiple analyses [Larsson & Carlsson, 1973]. Useful
overviews of these methods have been provided in the literature [Ayatollahi et al., 1998,
Aliabadi, 1997, Yang & Ravi-Chandar, 1999].

2.3.4 Fracture Testing

Fracture testing methods vary depending on the type of material and anticipated fracture
behaviour. Discussion here will focus primarily on fracture testing of brittle materials
[Anderson, 1992], although more general discussion may be found in Anderson [1995] or
Broek [1991]. Typically, specimens are made with a pre-notch to control the position of
crack growth, and are subjected to a tensile load. Some configurations, such as Compact
Tension (CT) and Double Cantilever Beam (DCB) configurations, involve the direct
application of tensile stress. Often tensile testing of ceramics is difficult as the increased
2-32

stress at application points can damage the specimen. Hence specimens are often loaded in
flexure, in 3-pt or 4-pt bending [Quinn & Morrell, 1991]. Specimens usually comply with
standard dimensions, for which SIF expressions are given in textbooks [Tada et al., 1986,
Broek, 1991, Anderson 1995] and testing standards [eg. ASTM Standard E1820-01, E39990, DIN 51109-98]. Applied load is measured with a load cell and displacement of various
points on the specimen is measured with transducers or microscopes.

P/2

P/2

Figure 2.14: Schematic diagram of 4-point bending configuration.

The creation of the notch is usually done by indenting the surface with a hardness indenter,
as in a Vickers hardness test, or by sawing a notch. Often this is followed with precracking under a low cyclic load. The size of the notch and pre-crack, and the method by
which they are produced, are critical in determining the residual stress-field and crack-wake
zone and, hence, are important considerations [Damani et al., 1996, Stech & Rdel, 1996,
Steinbrech et al., 1990]. Testing conditions, loading rate and ambient environment are
usually consistent with the standards. Crack propagation is observed and measured via
optical microscopy and by calculation of specimen compliance. The stress or strain fields
are often measured using strain gauges, Raman spectroscopy [Pezzotti, 1999], Moire
interferometry [Post et al., 1994, Han et al., 2001], coherent gradient sensing [Xu et al.,
1999] or digital image correlation [Lambros & Abanto-Bueno, 2002]. Fractography, the
observation and measurement of the post-fracture surface, often reveals much about the
fracture micromechanisms. These may also be observed during the fracture process with
electron microscopy [Rdel et al., 1990].

2-33

2.3.5 Crack deflection

There are various situations in which the loading at the crack tip is not just Mode I, but a
combination of modes, known as mixed-mode loading. Depending on the situation, the
crack may deflect so that it propagates under Mode I conditions, or it may continue to
propagate under mixed-mode loading conditions [Magill & Zwerneman, 1995]. In the case
of mixed-mode loading, the use of G is more appropriate than K as a determinant fracture
parameter as it incorporates loading from all modes. Discussion will focus here on mixed
mode I and II loading conditions. There are various methods for obtaining mixed-mode
loading in experiments [Qian and Fatemi, 1996, Kah Soh & Bian, 2001, Richard, 2001],
several of which have been analysed in depth [He et al., 1990, Fett et al., 1995].
For cracks loaded under mixed-mode conditions, a loading phase angle, , is used to
describe the mode-mixity [Erdogan, 2000]:

tan =

K II
KI

(2.35)

(a)

KII

KI
r
2v(x)

Crack

Kink ak

2u(x)
KII

(b)

KI

kI

Crack

kII

kII
kI

Figure 2.15: Diagram of crack under mixed-mode loading then deflected crack.

Mixed-mode loading tends to lead to crack deflection. The magnitude of this deflection is
predicted by a number of models [Cotterell & Rice, 1980, Qian & Fatemi, 1996]. The
various criteria that have been used to predict crack deflection angles in analytical and
computational models may be divided into two types: those methods which enable
calculation directly from the existing crack; and those which require consideration of a
virtual crack extension.
2-34

The most commonly used is the Maximum Tangential (or Circumferential) Stress criterion
(MTS- or MCS-criterion) proposed by Erdogan & Sih [1963]. This predicts that crack
extension will occur in the direction in which the tangential stress () at the crack tip is
maximised:
ie.

d
=0
d

d 2
<0
d 2

(2.36)

This has been experimentally verified by Williams and Ewing [1972] who examined
fracture paths in PMMA with initial cracks at various angles to the applied stress. As the
crack-tip stress distribution may be expressed in terms of the stress intensity factors, KI and
KII, the propagation direction may be directly determined for a given crack:
2
2 2 1 + 8 tan
tan =
8 tan
2

(2.37)

A crack-tip opening displacement criterion (CTOD-criterion) has been developed by Sutton

et al. [2000].

They defined a critical CTOD for fracture, for both mode-I and -II

propagation, at a material-dependent characteristic distance from the crack tip. In the purely
linear elastic case, this is equivalent to the MTS-criterion.

The minimum strain energy density criterion, proposed by Sih [1974], is also used
frequently. Strain energy density, S, is given by:
S = a 11 K I + a 12 K 1 K 2 + a 22 K 2
2

(2.38)

where aij are given by Sih [1974] and Qian and Fatemi [1996]. Crack growth is predicted in
the direction along which the strain energy density factor S is minimised, and fracture
occurs when S Sc:
dS
=0
d

d 2S
>0
d 2

(2.39)

Several other criteria, based on slightly different principles, require a small kink to be
extended from the crack tip. In each case, the kink is extended and relevant parameters
calculated, for a number of kink angles, to enable the optimum angle to be determined. The
Gmax criterion predicts that propagation occurs such that mechanical energy release rate is
2-35

maximized [Nuismer, 1975]. G may be calculated from displacement correlation,

compliance change, J-integral or crack closure integral. The (KII)min criterion is based on
the assumption that cracks will always propagate in the Mode I direction, ie. the direction
in which crack-tip stresses are symmetrical and KII is minimised [Banichuk, 1970,
Goldstein & Salganik, 1970, 1974]. KII may be calculated from displacement correlation,
closure integral or J and product integrals. The Gmax and KII = 0 criteria are both equivalent
to the MTS-criterion, to the first order.

Implementation of these criteria within a finite-element model involves the extension of a

test-kink from the crack tip, whose direction is varied to attain a minimum for kII. The
successive deletion and remeshing necessitated by this approach extends computation time
significantly [30,31]. Linear interpolation, within a small angular range, around 5, has
been used by Becker et al. [2000] to reduce the number of test angles required.

T-stress is known to have a significant influence on crack path and path stability [Cotterell
1965, Cotterell and Rice, 1980, Karihaloo et al., 1980, Tong, 2002], though this can vary
between specimen geometries [Melin, 2002]. The MTS criterion has been modified to
incorporate the effect of T-stress on tangential stress at a characteristic distance, rc [Smith et
al., 2001]. The crack opening displacement criterion proposed by Sutton et al. [2000] is
based on a similar approach with deflection determined from crack-opening displacements
at a characteristic distance from the crack tip, and becomes identical to the MTS criterion
under linear elastic conditions. Using the Gmax or KII = 0 criteria with a test-kink, the Tstresses will automatically be taken into account when the test-kink deflects from the cracktip direction.

Other criteria include the J-criterion, dilational strain energy density-criterion, and
tangential stress- and strain-factor criteria [Qian & Fatemi, 1996]. No single criterion
provides satisfactory predictions for crack path under all load conditions, and a judicious
application of several criteria may be required to realistically predict a crack propagation
path. Predictions from these different criteria all tend to be fairly similar for crack-tip
mode-mixity values below 30 [Bittencourt et al., 1992, Becker et al., 2000, Ohtsuka et al.,
2-36

2002], however there is some divergence for higher mode-mixities or in the presence of
large parallel T-stress. Bergkvist and Guex [1979] showed that differences between
predictions from different criteria tended to average out over crack propagation.

The criteria discussed have most relevance for ideal isotropic brittle materials. In some
materials, a statistical approach based on the assumption of pre-existing flaws may be more
appropriate. Becker et al. applied statistical fracture methods to FGMs and predicted the
accumulation of damage at larger angles than those typically predicted for crack deflection
[2002]. In other cases, the presence of a process zone or non-linear material behaviour
could complicate crack deflection [Rubinstein, 2003, Pettit et al., 2001]. Kotoul and Urbi
[2001] have considered a bridged crack with a kinked tip under mixed-mode loading. It was
shown that bridging has a strong influence on crack kinking and fracture criteria.
Furthermore, anisotropy in elastic properties and/or toughness can have a profound effect
on crack path [Yang & Yuan, 2000]; the splintering of wood is a prime example. In this
case, the crack path is understood to be that which maximises (G Gc).
Traditionally these deflection criteria have been developed for predicting crack propagation
direction in brittle materials. Under small-scale yielding conditions, crack-tip stresses,
hence mode-mixity, should still be dictated by the stress intensity factors, so that deflection
criteria based on SIFs should still be applicable [Shih, 1974, Suresh & Shih, 1986]. More
significant yielding near the crack-tip has been predicted to alter the effective mode-mixity
there [Ma et al., 2000, Dhirendra & Narasimhan, 1998]. The MTS criterion has been shown
to be applicable for small-scale yielding [Pawliska et al., 1993], though it was less
successful when more significant yielding occurred [Khan & Kraisheh, 2000]. Pettit et al.
[2001] showed that process zone effects could influence crack-tip mode-mixity, and hence
influence crack-path or stability. Various deflection criteria have been developed for
elastic-plastic materials [Li et al., 2004, Khan & Kraisheh, 2004, Bian & Kim 2004], but
these have not provided significantly better results than the criteria already discussed.

2-37

2.3.6 Deviant cracks

The development of fracture mechanics has been based on straight cracks. Whilst an initial
flaw under mode I loading will generally propagate in a straight line, a crack under mixed
mode loading can deviate from its original direction, thereby undergoing a change in shape.
Other situations in which significantly deviant crack-shapes may conceivably arise include
intersection of two cracks or the deflection of a crack due to toughness anisotropy, such as
in a laminate composite. It is generally understood that expressions developed for stress
intensity factors (SIFs) of straight cracks cannot be applied to non-straight cracks.
Understanding of curved cracks under mixed-mode loading is an important issue for
structural integrity, particularly for structures subjected to complicated or variable loadings
[Miranda et al., 2003, Plank and Kuhn, 1999], and for cracks near material interfaces
[Erdogan, 1995b].

Various solutions have been obtained for kinked cracks [Kitagawa et al., 1975, Bilby &
Cardew, 1975, Vitek, 1977, Lo, 1978, Karihaloo et al., 1980, Cotterell & Rice, 1980,
Hayashi & Nemat-Nasser, 1981], circular arc cracks [Sih et al., 1962, Chen, 1994] and
slightly curved cracks [Goldstein & Salganik, 1974, Cotterell & Rice, 1980]. Examples of
these are illustrated in Figure 2.16. Solutions are usually obtained using complex stresspotential techniques [Muskhelishvili, 1953, Boresi & Chong, 1987]. For severely deviating
cracks with irregular shapes, as observed in highly asymmetric loading situations [Miranda
et al., 2003], solutions for idealised crack-shapes are no more useful than those for straight
cracks. Chen and co-workers have developed models for calculating SIFs for arbitrary
crack shapes [Chen et al., 1991, Chen, 1993, 1994, 1999], by dividing the crack into a
number of straight sections, and using complex stress potentials to relate loading tractions
to the crack-opening displacements for each section. Application of models of this type to
real specimen geometries has been limited. Alternative methods for calculating SIFs for
non-straight cracks include FEM involving a domain integral similar to Equation 2.34 (p. 231) [Gosz et al., 1998], weight-function techniques [Guagliano & Vergani, 2001], and the
path-independent F-integral approach of Eriksson [2002, Lorentzon & Eriksson, 2000].

2-38

Although a crack shape may be quite complicated, several researchers have suggested that
the stress intensity factor may be approximated by a function of several simple crack-shape
parameters. It has been shown to be appropriate for particular crack-shapes and loading
configurations, although reasons were not given for why this is so [Kitagawa et al., 1975,
Leevers et al., 1976, Alpa et al., 1980, Noda et al., 1994]. Kitagawa and co-workers used
stress potentials and conformal mapping to investigate various bent (kinked) and branched
internal cracks [1975]. They suggested the approximate replacement of bent cracks with
equivalent straight cracks, and showed that stress intensity factors could be predicted fairly
accurately over a wide range of kink lengths and angles. They found discrepancies between
the bent crack and equivalent straight crack were highest for small kink lengths, probably
due to the interaction between the kink and the main crack. They found that circular arc
cracks were not as accurately approximated, which was attributed to curvature at the crack
tip.
(a)

(b)

(c)
(d)

Figure 2.16: Crack shapes: (a) kinked crack considered by Bilby et al. [1969], Kitagawa et al.
[1975], Lo [1978], (b) circular arc crack considered by Sih et al. [1962], Chen [1994], (c) piecewise
crack considered by Noda et al. [1993,1994], (d) zigzag crack considered by Chen [1999].

Noda and co-workers [1993,1994] modeled various curved edge cracks using a singular
integral equation formulation. They showed that a curved edge crack under mode I loading
could be simulated by a simplified crack with the same projected length and crack-tip
angle, and thereby concluded that curvature at the crack tip has a negligible effect on the
stress intensity factors. They addressed the possibility of the crack deflecting from its
original orientation, and found for a circular arc edge crack that the SIF in the direction
perpendicular to tensile stress was very close to that of a straight edge crack with the same
2-39

projected length. The apparently contradictory conclusions regarding the effect of crack-tip
curvature may be due to differences in crack configuration (internal vs. edge cracks).
Kitagawa or Noda did not discuss the reason why straight cracks can be used to
approximate more complicated shapes in any detail. Rather it was implicitly assumed that if
a curved crack may be approximated by a simpler shape, then a straight crack would be the
logical shape to use.

Approximation to an equivalent straight crack simplifies calculation appreciably and may

often be sufficiently accurate. In fact, this approximation is made when treating any
macroscopically straight crack as being perfectly straight, despite the fairly ubiquitous
presence of crack deflection at a microscopic level. The validity of the approximation in
this case is generally not questioned. For more significantly deflecting cracks, this
approximation is clearly useful for simply estimating fracture parameters, and thus warrants
further investigation.

There has been various work in recent years on simulating propagation of cracks with
arbitrary shapes, in both two- and three-dimensional cases. The key challenge in this area is
that crack paths are not known a priori, so that a FE mesh would have to be adapted. The
work of the Cornell Fracture Group [Bittencourt et al., 1996, Miranda et al., 2003] on FE
simulations, usually involving mesh-adaption or re-meshing, warrants mention, along with
that of Stone and Babuska [1998] and Bouchard et al. [2003]. Three-dimensional FE
simulations have been conducted by Schollmann et al. [2003] and Lin and Smith [1999].
Kamiya et al. [2000], Sumi and Wang [2003] and Mishnaevsky et al. [2003] have
incorporated microstructural effects into propagation models. Propagation in elasto-plastic
is attended by the further complication of strain history, which must be transferred on
remeshing [Lim et al., 1996, Pedersen, 1998, Trdegrd et al., 1998, Zhu & Gotoh, 1999].
Several approaches have been used to reduce or circumvent the need for re-meshing,
including the element-free Galerkin method [Belytschko et al., 1994, 1996, Belytschko &
Black, 1999, Mos et al., 1999], the extended finite element method [Dolbow et al., 2000]
and the local mesh replacement technique [Rashid, 1997, 1998].

2-40

2.4

Crack Propagation Processes

The response of the material near the crack tip, to the concentration of stresses there,
dictates the resistance to further crack propagation. While this resistance may, in some
cases, be described by a single number, ie. the fracture toughness, a more detailed analysis
of the mechanisms of crack propagation and their dependence on applied loading is often
required. For brittle homogeneous materials, including composites, fracture behaviour may
be characterised by quantifying the intrinsic crack-tip toughness, crack-growth resistance
(R-curve) behaviour and fatigue resistance. This section reviews intrinsic and extrinsic
mechanisms associated with crack growth under monotonic and cyclic loading. Several
such mechanisms are portrayed in Figure 2.17.
Extrinsic Toughening

Intrinsic Toughening

Grain
bridging

Cleavage
fracture
Microvoid
coalescence

Fibre
bridging

Behind crack tip

(crack-wake zone)

Oxide
wedging

Plastic
zone

Ahead of crack tip

(crack-tip process zone)

Figure 2.17: Schematic illustration of intrinsic mechanisms and extrinsic mechanisms associated
with crack growth (from Ritchie et al. [2000]).

2.4.1 Intrinsic Toughness

The resistance of a material to fracture depends on the deformation behaviour around the
crack tip and the resulting fracture micromechanisms. Ductile materials such as metals and
polymers deform significantly at the crack tip, leading to a considerable amount of energy
absorption as the crack propagates. The deformation is visible on the fracture surface,
which is typically textured and torn. Consequently, materials undergoing ductile fracture
have high values of Kc, eg. 40 MPam for Al, 150 MPam for alloy steel [Ashby & Jones,
1996], and are used in applications requiring greater toughness. Certain polymers will
2-41

deform significantly and whiten around the crack tip, a phenomenon known as crazing,
attributed to the rearrangement of molecular chains under high stress. Although polymers
have much lower Kc values than metals, their low Youngs moduli means their Gc values
are similar to those of metals. Brittle materials on the other hand, such as ceramics, deform
negligibly at the crack tip, hence little energy is absorbed and the fracture surface appears
much smoother. Materials undergoing brittle fracture have low toughness as signified by
low values of Kc, eg ~3 MPam for alumina, ~1 MPam for concrete [Ashby & Jones,
1996].

Ceramics are held together by covalent or ionic bonds which are specific and directional,
hence dislocations are only able to move in certain directions, if at all. Consequently,
ceramics display negligible plastic deformation and fracture occurs in a brittle manner
[Lawn, 1996]. This is a major drawback for applications, so considerable efforts have been
focussed in the past two decades on methods for toughening of ceramics. Fracture in
ceramics may be intergranular or transgranular. It depends heavily on microstructure and is,
therefore, strongly influenced by processing conditions. Ceramics often rely on extrinsic
toughening mechanisms, such as grain-bridging, for their toughness, though crackdeflection, transformation toughening, and microcracking can also increase energy
absorption during fracture, and increase toughness [Rhle et al., 1986, Anderson, 1996].
Mixed-mode toughness can often be different to toughness under purely Mode I conditions,
particularly in the presence of material anisotropy [Richard, 2001, Fett et al., 1995, Fett &
Munz, 2002]. This may be further influenced by T-stresses also.

Composite materials display a range of fracture behaviours, depending on the constituent

materials and the composite configuration. Often, brittle materials are toughened by the
addition of fibres or particles of a more ductile material [Sigl et al., 1988, Ashby et al.,
1989]. Deformation of the ductile phase, debonding and friction between phases can
increase the energy required for crack propagation, making the material tougher.
Interpenetrating-network structured composites display particularly good toughness
properties as the crack must pass through both phases, unlike particle-reinforced
composites in which the crack may deflect around the reinforcement phase [Knechtel et al.,
2-42

1994, Prielipp et al., 1996, Garcia et al., 1998]. Prediction of the fracture toughness of a
composite is difficult, as it is strongly dependent on the specific micromechanisms of
fracture. In some cases, a linear volume-fraction based estimate of effective toughness may
be adequate, though in many composites it is not [Hashin, 1983]. Often composites contain
residual thermal stresses as a result of processing, which can also influence fracture.

In many ceramics and composites, fracture is a stochastic process, due to statistical

variations in toughness or in inherent flaw distributions. Probabilistic approaches to the
analysis of fracture originated with the pioneering work of Weibull [1939], though the
statistical nature of strength of materials was identified by Leonardo Da Vinci [Parsons,
1939]. Statistical approaches continue to attract considerable attention [Ritchie et al., 1973,
Lin et al., 1986, Rahman, 2001], particularly in application to complicated structures
[Rahman & Kim, 2001] and composite materials [Barbero et al., 2000].

The present review of concepts in crack propagation has treated the stresses at the crack tip
separately from the toughness, or crack growth resistance, and it has been assumed they
would be calculated separately. It is within this paradigm that the field of fracture
mechanics developed [Atkins and Mai, 1985], particularly linear-elastic plastic fracture
mechanics, and within which the present study was conducted. More recently, the stress
and material resistance aspects of crack growth have been brought closer together, by
considering the mechanics of decohesion together with those of deformation [Hutchinson &
Evans, 2000].

At larger scales, this is achieved by explicitly modeling the cohesive zone, assuming a
relationship between cohesive tractions and decohesion displacements [Needleman, 1987,
Hutchinson & Tvergaard, 1993, De Borst et al., 2004]. Early work on cohesive zone
models includes the Dugdale-Barenblatt model for elastic-plastic crack growth [Dugdale,
1960, Barenblatt, 1962b], and the process-zone model for fracture of concrete [Hillerborg,
1976]. This approach potentially avoids the use, or assumption, of a stress singularity at the
crack tip and the attendant computational difficulties. A cohesive zone is usually
incorporated into a computational model using special cohesive interface elements with the
2-43

traction-displacement relation designed into them [Ortiz and Pandolfi, 1999, Roy and
Dodds, 2001]. Cohesive zone models have been used to simulate cracks in ductile [Roy and
Dodds, 2001] and brittle materials [Guo et al., 1999], at interfaces [Hutchinson and
Tvergaard, 1993], in composites [Yang et al., 1999], in porous materials [Nakamura &
Wang, 2001] and under fatigue loading [Yang et al., 2001]. Although they are addressed
within a single formulation, the deformation and decohesion processes are still treated
separately. Bulk deformation is modeled by the regular finite elements, whilst decohesion is
modeled by the cohesive-zone interface elements which must be put into the model
explicitly. This can be problematic if the crack path is not known a priori, or if a diffuse
damage zone is known to develop. At smaller scales, an even more integrated approach is
taken: the attractive and repulsive forces between molecules or atoms, or clusters of them,
are modeled explicitly. This takes the process of fracture down to its most fundamental
scale [Gumbsch, 2001, Cotterell, 2002, Li et al., 2003].

2.4.2 R-curve effects

A number of researchers have shown that crack wake effects can lead to an increase in
effective fracture toughness with crack length, which is known as R-Curve behaviour.
There are various mechanisms that, acting away from the crack tip, can reduce the crack-tip
stress intensity factor. This increases the effective toughness and impedes crack growth.
These usually involve interactions in the crack wake, as shown in Figure 2.17, and
therefore depend on the length of the wake-zone. For long steady-state cracks, this is fairly
constant, whilst for short cracks, the wake length is limited by the crack-length. Therefore,
the influence of extrinsic toughening mechanisms can vary with crack-extension. This is
referred to as an R-curve effect: an increase in effective toughness with crack propagation,
as illustrated in Figure 2.18.

The value for stress intensity factor (SIF) at the crack tip, Ktip, is less than the applied SIF,
Kapp(a), due to shielding effects, KS(a), that are dependent on crack-extension.
K tip = K appl (a ) K S (a )

(2.40)

where a is the crack length and the magnitude of Ks increases with crack extension as, for
example, more bridges form in the crack wake. Addition of SIF components in this manner
2-44

is valid as SIFs of the same mode may be superposed linearly, under linear-elastic
conditions. As Ks is a shielding term, it is negative.
For fracture to occur, the SIF at the crack tip Ktip must be equal to intrinsic fracture
toughness, K0, so the required value of Kapp increases with crack propagation. This equates
to an increase in effective fracture toughness:
ie.

K appl (a ) K S (a ) = K 0

=>

K eff
C (a ) = K 0 + K S (a ) = K appl

(2.41)

where Ka is the applied SIF, Ko is the crack-tip toughness and Kceff(a) is the crack length
dependent effective toughness. Note that for homogenous materials Ko is constant.
There is a similar relation, in terms of mechanical energy release rate:
R o = G appl R s (a )

(2.42)

where R is referred to as the fracture resistance. Rearranging this:

G appl = G c (a ) = R 0 + R S (a )

(2.43)

where Rs accounts for the additional work done against bridging tractions as the crack

Crack Growth
Resistance, Gc = R [J/m2]

extends.
Steep
R-curve

Shallow R-curve

No R-curve

G0

ao

Crack length, a = ao+a

Figure 2.18: Schematic of R-curves

The presence of an R-curve effect leads to an additional criterion for crack propagation.
Differentiating Equations 2.42 and 2.44:
2-45

dK dK c (a )
>
da
da

or

dG dG c (a )
>
da
da

(2.44)

as K0 is constant with respect to crack length. R-curve behaviour in a material is

particularly advantageous as it can lead to a crack self-arresting if the toughness increases
more significantly with crack length than the stress intensity, as indicated by Equation 2.44.
For a more rigorous treatment of the topic of R-curves, refer to Mai and Lawn [1987],
Rdel [1992] and others [Fett et al., 2000, Steinbrech et al., 1990, Cox & Marshall, 1991,
Cotterell & Mai, 1996].

2.4.3 Crack-Bridging
R-curve effects are often a result of bridging of the crack by intact ligaments or crack-face
interactions behind the crack tip. Bridging of a crack behind the crack tip, by grains
[Knehans & Steinbrech, 1982, Swanson et al., 1987, Mai & Lawn, 1987, Rdel, 1992] or
an intact ductile phase [Budiansky et al., 1986], reduce the stress intensity factor at the
crack tip. The energy absorption leading to toughening occurs behind the crack tip and acts
to reduce the stress at the tip, this is termed as a crack-wake effect or crack-tip shielding.

Bridging has long been understood to occur in fibre- and particle-reinforced composites
[Cotterell & Mai, 1997]. It was first shown to occur in ceramics by Knehans and Steinbrech
[1982] who sawed-out the crack wake zone in alumina and observed the toughness
reverting to the initial value. Crack-bridging has since has been verified in numerous
experiments [Swanson et al., 1987, Hu & Wittman, 1989, Hu & Mai, 1992, Pezzotti et al.,
1999]. Theoretical work on bridged cracks has been dominated by Cox and co-workers
[Cox & Marshall, 1991a,b,c, Cox 1991, 1993, Cox & Lo, 1992, 1993, Cox & Rose, 1994,
1996]. Several useful overviews have been given [Lawn, 1987, Bao & Suo, 1992, Cox &
Marshall, 1994, Cotterell & Mai, 1996].

Bridging tractions are provided by frictional forces between grains or between fibre and
matrix, by mechanical constraining forces between interlocking grains or by tension in
intact bridges. Rdel [1992] suggested a classification system for types of bridges in
ceramics and composites:
2-46

elastic (intact) bridges, which deform reversibly and linearly;

plastic bridges, which are not intact, but provide tractions due to frictional forces during
pull-out;

elastic-plastic (mechanically interlocking) bridges, which provide a reaction force

against crack-face opening; and

ductile bridges, which remain intact but deform plastically.

A similar delineation was proposed by Fett and Munz [1995]. For cracks with bridgingzone size comparable with the specimen geometry, load and specimen configuration have a
major influence on bridging behaviour [Fett et al., 2000].

Fracture toughness enhancement may be predicted ab initio from models of bridging

ligament deformation behaviour, via several different approaches. For steady-state crack
propagation, the increase in critical energy release rate, Rs, is related to the bridging
tractions by [Dugdale, 1960]:
um

R s = 2 p y (u )du

(2.45)

where um is the maximum value of the crack opening displacement. Alternatively, the
contribution, Kbr, of bridging tractions to the crack-tip SIF may be determined using an
appropriate weight function [Cox & Marshall, 1991, Rdel et al., 1991]:
a

K br = p( x )h ( x, a )dx

(2.46)

The equivalence of energy release and stress intensity factor approaches for crack-bridging
problems was demonstrated by Srensen et al. [2000]. The latter approach requires that the
spatial distribution of bridging tractions, p(x), be known. A simple expression may be used:
p( x ) = p 0 (1 x / x 0 )

(2.47)

where p0 is the maximum traction, x is the distance from the crack tip and x0 is the
maximum length of the bridging zone [Moon et al., 2002]. The exponent defines the
shape of the bridging traction profile. More usually, however, the bridging traction is
known as a function of bridge extension, which equates to crack-opening displacement, ie.
p(u), as discussed below. Relating crack-opening displacement to position can be difficult,
as the closure tractions act to decrease the crack-opening in the bridging zone.
2-47

(a)
Far-field loading
Bridge

Crack-Opening
u (x)
Displacements I

Traction, p(u)

x
Crack tip

uII(x)
Far-field loading

(b)

Without bridging
Bridging-zone length

With bridging
Critical COD

Bridging-zone

Figure 2.19: Schematics showing (a) an example of a single fibre bridge, and (b) a bridging zone,
comprised of the smeared distribution of a large number of discrete tractions as in (a).

The interdependent relationship between closure traction profile and crack opening profile
necessitates a self-consistent solution. The effect of bridging on the crack profile is not as
simple as that on the SIF. Bridges act to close the crack within the bridging zone but
further away, the profile is similar in form to the non-bridged case.

Analytical solutions may be attained for particular simple forms of bridging traction [Cox
& Marshall, 1991a, Foote et al., 1987]. The crack-opening displacement at a point x may be

2-48

determined from the weight function and SIFs for the crack as it propagates from that point
to its present position:
a

2
(K app (a ' ) + K br (a ' ))h ( x, a ' )da '
u(x) =
E' a x

(2.48)

These SIFs may then be related to the applied traction distributions using the weight
function, resulting in a second-order integral equation in u(x):
u(x) =

a a '
4

(
a
,
x
'
)
p
(
u
(
x
'
))
h
(
x
'
,
a
'
)
dx
'
h ( x , a ' )da '

app
E ' a x a

(2.49)

In some cases, it is possible to remove the applied stress component from the integral, as
the COD profile due to applied stresses only (in the absence of bridging):
u ( x ) = u app ( x )

a
a a '

4
p
(
u
(
x
'
))
h
(
x
'
,
a
'
)
dx
'

h ( x , a ' )da '

E ' a x a

(2.50)

To predict the R-curve behaviour of a material, the relationship between the crack-face
closure tractions provided by grain or fibre bridges, and the crack-opening displacement
(COD) profile is required. As the deformation behaviour of small inclusions or fibres often
differs significantly from that of a bulk material [Krstic, 1983] and other types of crackface interaction are hard to predict, it is difficult to estimate an appropriate expression for
the bridging traction. Elaborate models have been derived with some success for ductile
and frictional bridging [Krstic, 1983, Mataga, 1989, Budiansky & Amazigo, 1989,
Ravichandran, 1992, Spearing & Evans, 1992, Xu & Ostertag, 1999]. In many cases, a
simple expression has been assumed: a constant traction, linearly increasing or decreasing
with extension, or a power-law relation [Foote et al., 1987, Mai & Lawn, 1987, Rose, 1987,
Rdel et al., 1990, 1992]. Several such expressions are illustrated in Figure 2.19. Sohn and
Lee [1995] suggested the use of a statistical distribution of simple traction functions,
similar to the above, which compared favourably with experiment. Such models are fitted
to experimental results by varying parameters associated with the microstructure.

Rather than estimating, the bridging tractions may be measured directly. This was done by
Hoffman et al. [1997] and Raddatz et al. [2000] for Al-Al203 composites. They propagated
a crack through brittle matrix material without breaking the bridges and then measured
2-49

stress-strain relation of bridging material only. Single fibre tests are a well-established
technique for assessment of fibre composites [Zhao et al., 2000], and have been utilised for
characterisation of bridging tractions [Jacobsen & Sorensen, 2001]. The post-fracture
tensile test developed by Hay and White [1997, Geraghty et al., 1999, Fett et al., 2000c,
2000d] has been used to measure bridging traction relations for different sections of the
crack-wake zone, enabling effects of degradation to be quantified. Pezzotti and co-workers
[1998, 1999, 2001] have used Raman microprobe spectroscopy to measure bridging
traction stresses on either side of the crack. Alternatively, tractions may be back calculated
from R-curve and COD profiles [Cox & Marshall, 1991b, Gilbert & Ritchie, 1998]. Hu and
Wittman quantified crack-wake interactions in cementitious materials by progressively
sawing out this region and measuring variation of compliance [1989, 1990, Wittman & Hu,
1991]. An alternative approach of compliance measurement without sawing was used by
Hu and Mai to quantify crack-wake interactions in alumina [1992].

Bridging Traction, p(u)

Linear-softening
Linear-elastic
Uniform traction

Common form for

cohesive tractions:
exponential strainsoftening tail
Extension, u

Figure 2.20: Examples of bridging traction relations commonly assumed in models.

Incorporation of crack-bridging behaviour into a finite-element model presents an added

challenge due to the requirements of self-consistency of solution. A number of FEA-based
models have been developed for bridged cracks under Mode I loading, generally utilising
non-linear spring elements or user-defined interface elements, similar to the cohesive-zone
models discussed in the previous section. Materials simulated in this manner have included
alumina [Guo et al., 1999], fibre composites [Cao & Sakai, 1996, Lindhagen et al., 2000]
and metal-matrix composites [Rauchs et al., 2002]. Several key models are summarised in
2-50

Table 2.2. The mechanics of intact ligament bridges have also been modelled using FEA
[Mataga, 1989, Hoffman et al., 1997].

Table 2.2: Summary of computational models of crack-bridging reported in the literature.

Researchers
Guo et al.
[1999]
Lindhagen et
al. [2000]
Rauchs et al.
[2002]
Chandra et al.
[2002]
Cai & Bao
[1998]
Jin, Dodds &
Paulino [2002]
Bao, Fan &
Evans [1992]
Suo, Bao &
Fan [1992]
Yan et al.
[2001]

Mode
I

Material
Alumina

Fibre-reinforced
polymer
Fibre-reinforced
MMCs
SiC-reinforced Ti

Greco et al.
[2002]

I/II

I
I
I
I
I/II
I/II
I/II

Ceramic-metal
FGM
Ti-TiB FGM
Laminated fibre
composite
Laminated fibre
composite
Laminated fibre
composite
Laminated fibre
composite

Bridging Relation
A priori unknown
obtained via model
Linear and linear
softening
Friction with wear
damage
Exponential and
bilinear
Uniform
Non-linear
cohesive
Uniform
Piecewise linear &
linear softening
Linear, damagable

Piecewise linear

Method Used
FEM non-linear
springs
FEM non-linear
springs
FEM interface
elements
FEM non-linear
springs
FEM defined
p(x,x)
FEM non-linear
springs
FEM uniform
applied stress
FEM non-linear
springs
FEM linear
damagable interface
elements
Analytical & FEM
with interface
elements.

Bridged cracks under mixed-mode loading have also been simulated by a number of
researchers, generally in the context of delamination of fibre composites. These are also
summarised in Table 2.2. Bao et al. modelled the delamination of a fibre composite beam,
applying a uniform bridging traction which simplified the problem significantly [1992].
Suo et al. also investigated delamination, using spring elements in an iterative process to
apply linear and linear-softening tractions in both Modes I and II (uncoupled) [1992]. More
recently, Yan et al. used linear damageable interface elements to incorporate R-curve
behaviour in a model of mixed-mode delamination fracture of laminated fibre composites
[2001]. Greco et al. adopted a similar approach using damage interface elements with
piecewise linear tractions [2002], in both analytical and FE models. Kotoul and Urbis have
2-51

analytically modelled cracks with bridged kinks and found that bridging influenced kink-tip
mode-mixity [2001].

2.4.4 Fatigue
One of the greatest causes of unexpected structural failure is cyclic fatigue. When a
material is loaded and unloaded repeatedly, even if the applied stress may be much less
than the fracture stress (and may even be less than the yield strength), damage occurs and
microscopic flaws are able to propagate. This progressive accumulation of damage can
often lead to failure, and is known as fatigue. As most engineering applications involve
cyclic loading of components, understanding of fatigue crack growth is crucial. For a more
detailed treatment of many of the aspects of fatigue, see Anderson [1995], Broek [1991] or
Suresh [1998]. Several general reviews have been given by Newman [1998], Schijve
[2003] and Vasudevan et al. [2001], along with more specific reviews covering fatigue in
different classes of materials [Fleck et al., 1994, Ritchie, 1999], in biomaterials [Teoh,
2003] and its effect on structural integrity [Berger et al., 2002].

Micromechanisms associated with fatigue damage differ between ductile, brittle and
composite materials. In metals, fatigue occurs due to the build-up of dislocations around the
crack tip, caused by repetitious plastic deformation. This leads to an alternating sharpening
and blunting of the crack tip, which causes a progressive growth of the crack. Ceramics
were originally believed not to undergo fatigue, as negligible plastic deformation occurs
[Sarkar et al., 1969]. However, it is now known that fatigue occurs in brittle as well as
ductile materials [Soma, 1988, Ritchie, 1999]. Fatigue behaviour of polymers is dependent
on their ductility, and can often be dependent on the frequency of cyclic loading, due to
viscoelastic mechanical behaviour. Composite materials are particularly susceptible to
fatigue as cyclic loading causes wear at the interfaces between phases and can lead to
degradation of reinforcing particles and fibres, resulting in a complex damage state
[Stinchcomb & Reifsnider, 1979, Poursatip et al., 1982, Reynaud, 1996, Ritchie et al.,
2000]. Numerous damage mechanisms are understood to contribute to fatigue degradation
in composites [Majumdar et al., 1995, Dvorak, 2000, Nairn, 2001]. The effects of fatigue
on crack-wake zones in composites and ceramics are discussed in Section 2.4.5.
2-52

There are two stages in fatigue behaviour: crack initiation, and crack propagation.
Specimens without initial flaws exhibit initiation-controlled fatigue, whereas those with
pre-existing flaws display propagation-controlled behaviour. Within crack propagation,
there is a notable distinction between short and long (steady-state) cracks [Suresh, 1998,
Taylor, 2002]. This can be related to microstructural interaction, the development of a
crack-wake zone or localised effects such as residual stress [Gilbert et al., 2000].

While it is well recognised that repeated sub-critical loading leads to material degradation
and crack growth, the magnitude and state of this degradation is difficult to predict in many
cases. Fatigue testing is carried out with loading and measurement apparatus similar to that
used for monotonic fracture testing. The applied loading cycle is described in terms of the
maximum and minimum stresses, max and min, or, more commonly, the stress amplitude
and load ratio, R (shown in Figure 2.21):

R=

min Pmin
=
max Pmax

Load or
Stress
Pmax (or max)

(2.52)

1/f

Pmin/Pmax = R

Pmin(or min)
Time
Figure 2.21: Load cycle parameters for fatigue characterisation.

Characterisation of fatigue behaviour of materials is often fairly empirical. Propagationcontrolled behaviour is described using graphs of crack propagation rate, da/dN, plotted
against SIF amplitude, K, where N is the number of cycles, a is crack length and K is
defined as:
K = Y a

(2.53)

2-53

These graphs, with both axes plotted logarithmically, typically exhibit a sigmoidal shape,
displaying 3 stages of fatigue, as in Figure 2.22. Stage 2 is of greatest interest as Stage 1 is
so slow that it rarely leads to failure while Stage 3 is so fast that component lifetime is
negligible. Fatigue crack propagation in Stage 2, as in Figure 2.22, is often described by
Paris law:
da
= A ( K ) n
dN

(2.54)

where A and n are empirically determined material constants. Numerous variations of this
relation have been used to fit various experimental results, and to account for stage 1 and
stage 3 behaviour. By integration of Equation 2.54 from initial to critical flaw size, the

Fatigue crack growth rate, da/dN (log scale)

total fatigue lifetime may be determined.

Stage 2:
Paris-law regime
Stage 1:
Threshold
regime

KTh

R = Kmin/Kmax
K = Kmax(1-R)
Stage 3:
Critical regime

KCr = Kc(1-R)

Applied SIF Amplitude, K (log scale)

Figure 2.22: Fatigue characterisation graph showing 3 stages of fatigue crack growth.

The Paris law relationship has been shown to apply for brittle materials [Gilbert et al.,
1997, Ritchie et al., 2000], albeit over a fairly narrow stress range, with very high values of
the exponent, m, observed. For such materials, the use of several variations of Paris law
has proved expedient. The first normalises the stress intensity range in terms of the
measured toughness of the material:
2-54

 K
da

= A '
dN
Kc

(2.55)

This non-dimensionalises K. The second variation acknowledges the influence of peak

SIF, Kmax, in addition to that of SIF range, K, on the fatigue damage accumulation:
da
= A ' ' (K ) n (K max ) m
dN

(2.56)

where the exponent m dictates the influence of Kmax. For constant R, this is essentially the
same as the original Paris law in 2.54, with n equal to the sum of exponents in 2.56 and
A related to A and R. A review of many of the other variations on Paris law was given by
Kamiski [1999], with an emphasis on probabilistic approaches. Various approaches take
the fatigue damage history into account, ranging from simple [Willenborg, 1971, Wheeler,
1972] to very complicated [Patankar, 1998, Vasudevan, 2001].

Fatigue crack propagation under mixed-mode loading conditions has been investigated
fairly extensively [Qian & Fatemi, 1996, Magill & Zwernemann, 1997, Plank & Kuhn,
1999, Taylor et al., 1999, Wang & Pan, 1999, Kah Soh & Bian, 2001] however, as in the
case of mixed mode fracture, a conclusive understanding of crack path criteria or crack
propagation rates has not yet been reached. These have been described using various
expressions for effective SIF and effective strain energy density factor, which are fairly
empirical and not universally applicable.

The accumulation and propagation of damage under cyclic loading frequently has a
significant stochastic component, especially in heterogeneous materials. There has been
significant work in this area [Sheikh & Younas, 1995, Peng et al., 1998, Kamiski, 1999,
Luo & Bowen, 2003] including statistical analyses of fatigue crack growth laws [Rocha &
Schueller, 1996, Bigerelle & Iost, 1999, Kim, 2000]. Multiscale models linking continuum
damage mechanics with larger-scale fatigue degradation and crack propagation have
proliferated in recent years [Fish & Yu, 2002, Le Pen et al., 2002].

2.4.5 Bridging and fatigue

In brittle materials, the degradation of the bridging zone under cyclic loading can be the
main mechanism underlying fatigue crack growth. Venkateswara Rao and Ritchie [1992]
2-55

observed a significant reduction in the bridging effect in TiNb-reinforced TiAl-matrix

composites, with a decrease in the bridging zone from 4-5 mm to <150 m. Similar
evidence was found by Gilbert and Ritchie [1998], who measured crack profiles and
propagation rates in silicon carbide under cyclic loading. They showed that the bridging
traction relation as a function of position degraded with cyclic loading, as shown in Figure
2.23. Theoretical modeling by Cox and Rose [1994] of degradable elastic bridges under
cyclic loading identified two rate parameters in the problem: the rate of crack advance and
the rate of bridging traction degradation.

Figure 2.23: Example of fatigue degradation of bridging traction (from Gilbert and Ritchie [1998]).
Bridging traction relations determined from fitting of crack-opening displacement profiles for
cracks grown (i) near the critical point for fast fracture (92% of Kc) and (ii) near the threshold point.

(75% of Kc).
The influence of crack-extension effects on crack propagation under cyclic loading is likely
to be more complicated than under monotonic loading, due to progressive fatigue
degradation of the bridging zone [Ritchie, 1999]. The shielding of the crack-tip would be
dependent on crack-extension as well as applied loading:
K tip = K app K br ( a , K app , R )

(2.57)

The dependence of Kbr on applied loading, Kapp, would be both direct, via crack-opening
which scales with applied loading, and indirectly, through crack propagation rate, which
defines the damage state of the bridging zone in steady-state conditions. The traditional
2-56

approach for quantifying fatigue behaviour by presenting crack propagation rates as a

function of SIF amplitude becomes insufficient when R-curve effects are present.

Alumina has been demonstrated to undergo fatigue crack growth [Sarkar, 1969, Vaughan
1987]. Hu and Mai [1992] identified frictional degradation of grain-bridges as the major
fatigue mechanism in alumina, and showed that the maximum bridging traction value was
reduced under cyclic loading. Dauskhardt [1993] provided a model for wear-degradation of
frictional bridges in alumina under cyclic loading. Kishimoto et al. [1994] found that larger
grain sizes lead to lower crack propagation rates, and that cyclic loading reduced the Rcurve effect. Gilbert and co-workers [1997] obtained similar results. Values for A and n of
22 m/cycle and 466 respectively were determined by Choi and Salem [1995], having
normalised K by dividing by Kc = 3.1 MPam, as in Equation 2.55. Geraghty et al. [1999]
experimentally demonstrated that the onset of frictional sliding was reduced and effective
compliance was progressively increased with fatigue.

Begley and McMeeking [1995] modeled bridging in metal-matrix composites under cyclic
loading. Using an integral equation formulation with assumed cyclic bridging laws, they
solved for crack-tip SIFs and crack growth rates. Rodopolous et al. [2001] conducted
similar simulations predicting fatigue life and fracture toughness for titanium-matrix
composites, considering degradation micromechanics of bridges and elastic-plastic crack
growth in the matrix. Connell and Zok [1997] measured the cyclic bridging traction law in
a titanium matrix composite by cutting a section from the wake-zone for multiple-fibre
pullout testing. They demonstrated a reduction in fibre pull-out strength after fatigue
loading, and applied this in simulations of crack growth. Zhang et al. [2001] measured the
cyclic bridging law for fibre-reinforced concrete, and observed wear degradation of the
friction bond between fibre and matrix. Rauchs and co-workers [2002a,b, [Preuss et al.,
2003] conducted simulations of fatigue crack growth in Ti/SiC composites, considering the
effects of load ratio, thermal expansion mismatch, interface friction and wear on crackbridging. They experimentally characterised fibre and matrix strains in similar composites
using synchrotron strain mapping and observed degradation of the interfaces.

2-57

2.5

Cracks in Interfacial and Layered Materials

Discussion of fracture in material containing interfaces or layers is relevant in the context

of graded materials for several reasons. Firstly, as gradient steepness increases, the stress
and fracture behaviour in the graded region approaches that of a bimaterial interface [Marur
& Tippur, 2000]. Secondly, a number of issues pertinent to the fracture behaviour of graded
materials have previously been encountered in fracture at material interfaces: anisotropic
fracture behaviour; asymmetric crack-tip stress fields and crack deflection; and residual
stresses. Finally, so-called graded materials produced experimentally are often comprised
of a series of steps rather than a smooth gradient, so they contain interfaces [e.g. Lin &
Miyamoto, 1999, Chapa-Cabrera et al., 2000, Uzun et al., 2001].

E1, 1, 1, Kc1, y1

Material 1
Interface

E2, 2, 2, Kc2, y2

Material 2

Figure 2.24: Schematic of interface between two materials showing mismatches in various
properties. While some properties may be closely matched, others will not.

Several issues differentiate cracks near or at material interfaces from those in

macroscopically homogeneous materials. The interface introduces anisotropy and a
reduction in symmetry. There will be certain differences in the thermomechanical
properties of the materials on either side of the interface, as illustrated in Figure 2.24. This
can lead to a stress concentration near the interface, with compressive stresses on one side
and tensile stresses on the other. When a crack is present, the relative position and
orientation of the crack relative to the interface will determine the shape of the crack-tip
stress-field and the resultant fracture behaviour. Several possible configurations are shown
in Figure 2.25. A distinction may be made based on crack-tip position: cracks near the
interface, as in (a) and (c) in Figure 2.25, vs. those with the crack tip at the interface, as in
(b) and (d). Alternately cracks may be distinguished based on orientation. Typically, the
orientation dependence is explicated by considering the two extreme cases: cracks normal
2-58

to the interface, shown in Figure 2.25(a) and (b); and those propagating parallel to or along
the interface, as in (c) and (d).
Normal to the interface
(a)

(b)

Cracks
near
interfaces (c)

(d)

Cracks at
interfaces

Parallel to the interface

Figure 2.25: Schematic of various interface crack configurations: (a) approaching, or propagating
across, normal to the interface; (b) intersecting normal to the interface; (c) near to, and parallel to,
the interface; and (d) along the interface.

Various crack/interface configurations have been investigated, due to their relevance to

engineering applications. Fracture and fatigue experiments have been conducted on:

bimaterial specimens, as in Figure 2.26(a) [Sugimura et al., 1995b, Pippan et al., 2000,
Jiang et al., 2003];

sandwich (brittle/ductile/brittle trilayer) specimens, as in Figure 2.26(b) [Cao et al.,

1988, Reimanis et al., 1991, Ritchie et al., 1993, Kruzic et al., 2004, McNaney et al.,
1994, 1996, Hasegawa et al., 1998]; and

multilayered (> 3 layers) specimens, as in Figure 2.26(c) [Shaw et al., 1993, Lee et al.,
1996, Blanks et al., 1998, Lugovy et al., 1999, 2002, Clegg, 2000, El-Shaer & Derby,
2004].

Theoretical and computational models have been developed for:

idealized bimaterial interfaces (ie. semi-infinite planes on either side ) [Williams, 1959,
Dundurs, 1969, He & Hutchinson, 1989, 1989b];
2-59

bimaterial specimens [Sugimura et al., 1995a, Kim et al., 1997, Pippan et al., 1998,
Joyce et al., 1998];

sandwich/trilayer specimens [Charalambides et al., 1990, McNaney et al., 1994,

Mintchev et al, 1998, Kang & Beom, 2000]; and

multi-layered specimens [Shaw, 1998, Selvarathinam, 1998, Jha & Charalambides,

1998, Lugovy et al., 2005].

The following sections address the issues of stress fields near interfaces, cracks at and near
to interfaces, then cracks in layered materials.
(a)

(b)

(c)

Figure 2.26: Schematic illustration of several crack configurations in interfacial and layered
materials. (a) bimaterial interface specimen, (b) sandwich specimen showing possible crack
positions and (c) a layered material showing possible crack propagation directions, parallel,
perpendicular and inclined to layers.

2.5.1 Stresses near interfaces

The stress and strain fields near an interface are dictated by the mismatch, in thermal,
elastic and plastic properties, between the materials on either side of the interface.

Prior to the application of external loads, an interface between two materials with differing
thermal properties will often contain residual stresses, as a result of processing at high
temperature [Hu et al., 1998]. Mismatch in thermal properties will tend to lead to residual
stresses parallel to the interface, compressive on one side and tensile on the other. This
situation was illustrated in Figure 2.2(a) on page 2.2.
2-60

When an external load is applied, the elastic mismatch at an interface can lead to altered
stress fields. Stresses acting normally across the interface will not be affected, but stresses
parallel to the interface and shear stresses can be significantly altered, as compared with
stresses in a homogeneous material structure. The less compliant side will take a greater
proportion of the load than the less compliant side.

Mismatch in yielding behaviour can influence stress fields if stresses are high enough to
induce plastic deformation [Yang et al., 1997]. Yielding essentially contributes to
compliance, leading to an increase in the stresses on the side with less yielding. In the
presence of high thermal mismatch, plastic deformation or creep during cooling can act to
reduce the magnitude of thermal stresses in this region. Furthermore, stresses resulting from
elastic mismatch will have an influence on plastic deformation.

2.5.2 Cracks at interfaces

Cracks at sharp interfaces between materials differ fundamentally from those in
homogeneous materials [Williams, 1959, England, 1965, Rice & Sih, 1965, He &
Hutchinson, 1989, Dundurs, 1969, Geubelle & Knauss, 1994, Sih & Chen 1981]. This topic
has been summarised by Rice [1988]. When the crack tip is at the interface, the stress field
nearby displays a non-square-root singularity, hence the SIF becomes invalid for
quantifying fracture behaviour [Rice, 1988, Erdogan, 1995b]. The mechanical energy
release rate is still valid, and is related to the crack-tip stresses and properties of each
material. The crack-tip stresses will be dictated by the mismatch in elastic and plastic
compliance between the materials on each side of the interface, and can result in stress
concentration and mode-mixity.

For a crack normal (or oblique) to the interface with crack tip at the interface, as shown in
Figure 2.25(b), the stress field will be symmetrical. The crack can either propagate across
the interface into the second material, or deflect and travel along the interface [He &
Hutchinson, 1989] or, in some cases, the crack could arrest at the interface [Mintchev et al.,
1998]. The path taken will depend on the relative resistance to crack propagation of the
interface and the bulk material, and the direction of applied loading. A similar criterion may
2-61

be used to predict whether a crack meeting a reinforcing fibre will deflect [Selvarathinam,
1998].

For a crack propagating along a bimaterial interface, as shown in Figure 2.25(d), the
mismatch between elastic properties leads to a change in mode-mixity, which may be
determined from stresses near the crack tip. This means that, under applied tensile (mode I)
loading, mixed-mode loading may occur locally at the crack tip [Hutchinson & Suo, 1992].
This mixed-mode loading may lead to crack deflection and interfacial debonding depending
on the toughness variation across the interface [He & Hutchinson, 1989b]. Furthermore, the
mode-mixity can influence fracture toughness, especially if process zone effects are present
[Tvergaard, 1999, 2003]. The mechanical energy release rate for a crack along an interface
between two materials is generally higher than for an equivalent crack in a homogeneous
material [Rice, 1988].

The path of the crack will depend on the mode-mixity and on the relative toughnesses of
each material and the interface. If the interface is significantly weaker than either material,
the crack will propagate along it under mixed-mode conditions. A criterion for determining
crack path was given by He and Hutchinson [1989b]. Deflection criteria for homogeneous
materials, presented in Section 2.3.5 (pp. 2-34), are not necessarily applicable for layered
materials due to toughness anisotropy at the interface and differing crack-tip stress field
[He & Hutchinson, 1989b, Oh et al., 1988]. The toughness and fatigue resistance of
bimaterial interfaces have been investigated by Oh et al. [1988], Evans et al. [1990] and
Cannon et al. [1991]. Influences of properties of each material, mode-mixity and interface
roughness were demonstrated.

Cracks along interfaces between brittle and ductile materials are of particular interest, and
have been investigated extensively, both experimentally [Reimanis et al., 1990, Cannon et
al., 1991, Dalgleish et al., 1991, McNaney et al., 1996] and theoretically [Rice, 1988, Shih
& Asaro, 1988, Zywicz & Parks, 1992, Tvergaard & Hutchinson, 1993, 1994, Deng, 1995,
Sham et al., 1999, Bose et al., 2004]. For crack growth along the interface between brittle
and ductile materials, the effective toughness, |K|R, increases with crack extension from Ko,
2-62

the toughness in the absence of plasticity [Tvergaard & Hutchinson, 1993, 1994]. The
increase results from the development of a plastic zone on the ductile side of the interface,
and depends on yielding behaviour, crack-tip mode-mixity and crack-tip toughness [Liang
& Liechti, 1995]. This toughening effect has been observed experimentally for cracks on
interfaces [McNaney et al., 1996, Kruzic et al., 2004, Hasegawa et al, 2003].

2.5.3 Cracks near interfaces

Fracture near interfaces is dependent on crack orientation relative to the interface. Here we
consider the two extreme cases: cracks normal to the interface, as shown in Figure 2.25(a);
and those parallel to the interface, as in Figure 2.25(c) [Rice, 1988]. This configuration is
pertinent to the study of cracks in FGMs, addressed in detail in Sections 2.6 to 2.8. In each
case, the crack will be influenced by the property mismatch, but the crack-tip mechanics
will be the same as in homogeneous materials (ie. square-root singularity at crack tip),
unlike a crack at the interface.

A crack approaching an interface perpendicularly on the less compliant side, as in Figure

experiences an increased SIF, due to stress concentration at the interface [Bleek et al.,
1998], whilst one on the more compliant side experiences a decreased SIF [He &
Hutchinson, 1989b]. This effect has been described as elastic shielding or anti-shielding of
the crack-tip. Residual thermal stresses at and near interfaces in layered materials can have
a significant influence on the measured fracture behaviour, due to modification of the
crack-tip stress field [Bleek et al., 1998, Hu et al., 1998, Mintchev et al., 1998].

Sugimura et al. [1995a] investigated, through FEM and experiments, the influence of
plasticity mismatch on cracks normal to interfaces. They showed that the driving force for
crack growth is strongly influenced by the interaction between the near-tip plastic zone and
the interface. For a crack in the weaker (ie. more easily yielding) material, the near-tip Jintegral is reduced compared to the applied loading, whereas the J-integral at the tip of a
crack in the stronger (higher yield stress) material is increased. They showed that the
amplifying or shielding effect is greater: (i) for lower yield stress, (ii) when the crack tip is
closer to interface, (iii) for higher crack-tip toughness (ie. greater stresses around crack tip)
2-63

and (iv) for greater plastic zone radius, which is essentially a product of the first three. They
predicted amplification factors up to 2.6 and shielding factors less than 0.4. Furthermore
they showed that negative (compressive) T-stress could increase the effect. The plastic
shielding or anti-shielding effect resulting from mismatch in plastic compliance is
essentially equivalent to the elastic shielding (or anti-shielding) effect described above.
Sugimara et al. [1995b] further demonstrated the shielding/antishielding effect for cracks
approaching plastic interfaces at an angle. Cracks approaching interfaces have since been
investigated with computational simulations [Kim et al., 1997, 1999, Pippan et al., 1998,
Simha et al., 2003, Joyce et al., 2003, Kolednik et al., 2005] and experiments [Pippan et
al., 2000, Jiang et al., 2003], with findings in agreement with Sugimura and co-workers.

The case of a crack oriented parallel to the interface a small distance away from it has
received significantly less attention than that of a crack along the interface. A crack parallel
to interface will tend to experience mixed-mode loading due to the asymmetry in
compliance [Hutchinson et al, 1987]. This is likely to result in crack deflection, away from
the interface if the crack is in the more compliant material, and towards it for a crack in the
stiffer material. Compliance mismatch can also lead to stress concentration or shielding,
depending on whether the crack is in the less or more compliant side. Fleck and co-workers
[1991], considering brittle fracture paths in thin adhesive layers, showed that, due to elastic
shielding of the adhesive by the more rigid substrate, the apparent toughness is higher than
the real toughness by a factor of (Es/Ea)1/2, where Es and Ea are the Youngs moduli of the
substrate and adhesive layers respectively. For a glass alumina system they examined, the
factor was 2.4. Conversely, a less rigid substrate would lead to an amplification of crack-tip
stresses, effectively an anti-shielding effect. Residual stresses may also have an influence.

For a crack in a ductile material near the interface with a brittle material [Liu et al., 1996],
plasticity near the crack tip was influenced, firstly, by the mode-mixity resulting from
elastic mismatch across the interface. At higher loads, constraint became an issue as the
plastic zone reached the interface. The plastic zone spread along the interface affecting
crack-tip mode-mixity. Similar results were obtained by Kang & Beom [2000] for a crack
in a ductile layer between brittle solids.
2-64

The concentration of stresses near the crack tip with applied loading can induce plastic
deformation in the nearby ductile section [Charalambides et al., 1990]. On one hand, this
could lead to increased stress concentration at the crack tip due to the added compliance as
discussed above [Sugimura et al., 1995a, Kim et al., 1997]. On the other hand, the energy
absorption associated with crack propagation could be increased [Charalambides et al.,
1990]. Charalambides et al. [1990] used a simple analytical model and a rigorous FE model
to analyse the mechanical energy release rate for a crack growing parallel to an elastically
constrained ductile layer. They showed that an increase in effective fracture resistance
could arise, due to energy absorption by plastic deformation.

There appear, therefore, to be two possible effects of plasticity in a neighbouring ductile

layer on a crack parallel to an interface: stress concentration and mode-mixity due to
compliance mismatch; and increased toughness due to accumulation of plastic strains.
These two effects have not been explicitly addressed together previously. In addition,
plastic yielding around the crack-tip can alter the crack propagation behaviour under
mixed-mode loading.

Mode-mixity due to compliance mismatch could result in crack deflection either toward or
away from the interface. Sugimara et al. [1995b] also observed deflection of cracks initially
oriented parallel to the interface. Deflection has also been observed for cracks near the
interface in brittle/ductile/brittle sandwich specimens [Ritchie et al., 1993, McNaney et al.,
1994, Kruzic et al., 2004]. If the crack is in a yielding material, crack deflection under
mixed-mode loading could be affected [Dhirendra & Narasimhan, 1998, Ma et al., 2001,
Bose et al., 2004]. Work on predicting elastic-plastic crack propagation paths near
interfaces between ductile and brittle materials has been limited [Tvergaard, 2002, ChapaCabrera & Reimanis, 2002a,b, Rashid & Tvergaard, 2003].

The deflection criteria discussed in Section 2.3.5 should be applicable to cracks near
interfaces as the crack-tip stress field has the same form as in homogeneous materials. As
mentioned in Section 2.3.5, plastic deformation could influence propagation direction; this
2-65

may occur at the crack-tip or in the neighbouring region. The tendency of cracks near
interfaces to deflect means they can quickly become cracks at interfaces, and vice versa [He
& Hutchinson, 1989, 1989b].

2.5.4 Cracks in layered materials

Materials comprised of several discrete layers, as in Figure 2.26(c), contain a number of
bimaterial interfaces, which is reflected in their fracture behaviour [Mujul, 2000]. Layered
composites also display anisotropic fracture behaviour [Marshall et al., 1991], as cracks
parallel to the layers will behave differently to those directly through the layers. Cracks in
layered materials represent a combination of the cases discussed above, of cracks at and
near interfaces, depending on the location of the crack-tip. Apart from the local variation at
the scale of the layers, the properties are uniform across the structure, so that property
mismatch effects are not as pronounced as in bimaterial specimens. Failure of layered
materials can involve multiple mechanisms including interfacial debonding, crack
renucleation across intact layers and ductile deformation of intact layers [Shaw et al., 1993,
Selvarathinam, 1998]. The interface between layers plays an important role in the
distribution of stresses and overall tensile strength of layered materials [He et al., 1993].

Design strategies for multilayered composites involve one or more of the following [Chan,
1997]:

weak interlayers or interfaces, so that cracks will deflect rather than travel directly
through [Clegg, 2000, Leguilon et al., 2005, Lee et al., 1996, Blanks et al., 1998, Lee,
2002, Ma et al., 2004b];

controlled residual stresses to increase effective fracture toughness [Moon et al., 2001,
Lugovy et al., 2005]; or

tougher interlayers to increase energy absorption and inhibit crack growth [Shaw et al.,
1993, Huang & Zhang, 1995, Hwu & Derby, 1999, El-Shaer & Derby, 2004].

For propagation perpendicular to layers, crack-tip stresses and propagation resistance will
vary periodically [Cao & Evans, 1991]. Crack progression occurs either by the extension of
a dominant, nearly planar crack or by the formation of a zone of periodically spaced cracks
2-66

[Shaw et al., 1993]. Shaw [1998] used a probabilistic approach to analyse the resistance of
layered materials to cracks travelling normal to the layers. Compliance mismatch, due to
differing elastic and yielding properties, can result in cracks growing ahead in one material,
whilst the other material remains intact, effectively bridging the crack.

For layered ceramic composites, Marshall and others have showed that the effective
fracture toughness and R-curve behaviour was strongly affected by the presence of residual
stresses [Marshall et al., 1991, Lakshminarayanan et al., 1996, Moon et al., 2001]. This is
illustrated in Figure 2.27, which displays R-curves for a homogeneous material and a
layered composite [Moon et al., 2001]. The intrinsic fracture toughness does not vary
significantly throughout the layered composite, so the significant variation in effective
toughness is attributed to residual stress associated with the thermal expansion mismatch

1/2

[MPa m ]

between the layers.

5
80% Alumina
Homogeneous

Effective Fracture Toughness K

90%
Alumina

80% Alumina

Layered Composite
3

Compressive
Residual
Stress

Tensile Residual
Stress

200

400

600

800

Crack Extension, a [m]

Figure 2.27: The experimentally measured ({) and theoretically calculated ( ) R-curves for a
layered 90/80%alumina-10/20%zirconia composite compared with the experimentally measured
(z) and theoretically calculated (---) R-curves for a homogeneous 80%alumina-20%zirconia
sample. [Moon et al., 2000].

2-67

For a crack initially propagating parallel to the layers, there are several possible crack path
options: propagation along an interface or propagation in one or other of the materials. The
path taken will depend on both the relative toughness of each, and the stiffness mismatch
between the materials [Jha & Charalambides, 1998]. Furthermore, branching and
wandering of the crack may occur. In some cases, propagation parallel to the layers is
preferable to through the layers; this can be induced by weak or porous layers or weak
interfaces [Clegg, 2000, Ma et al., 2004b].

Crack propagation parallel to the layers in sandwich specimens, as shown in Figure 2.26(b),
has been investigated by Ritchie and co-workers [Ritchie et al., 1993, McNaney et al.,
1994, 1996, Kruzic et al., 2004] have conducted extensive investigations into cracking of
ceramic/metal/ceramic sandwich specimens. Cracks in the ceramic region, initially within
500 m of the interface, deflected towards the ductile layer, which led to crack paths that
alternated between the interface and the ceramic region. Crack jumping from one interface
to the other was observed, and subsequent bridging, though this was understood not to
contribute significantly to effective toughness [McNaney et al., 1996]. Cracks initially
situated further from the interface did not deflect appreciably. For glass/copper specimens,
cracks situated in the glass deflected away from the copper layer due to the higher
compliance of the glass. Observed crack paths were governed by a combination of elastic
and plastic properties [McNaney et al., 1994]. Effective toughness and fatigue crack growth
resistance was influenced by plastic deformation in the ductile layer [Kruzic et al., 2004]

R-curve behaviour can occur for crack propagation parallel to the layers, as a plastic zone
develops near the crack and/or in neighbouring layers [Hwu & Derby, 1998, Charalambides
et al., 1990, Kruzic et al., 2004]. If a plastic zone develops in the ductile layer, the extent of
plastic deformation could be constrained by the interface with the brittle region. This is
predicted to limit the toughening effects of plastic deformation for cracks along interfaces
[Tvergaard & Hutchinson, 1994, Kohnle et al., 2002] and for cracks situated in the ductile
region [Liu et al., 1996, Kang & Beom, 2000]. This constraint effect has been demonstrated
experimentally for interface cracks in sandwich specimens [McNaney et al., 1996,

2-68

Hasegawa et al., 2003, Kruzic et al., 2004]. Narrower ductile layers were shown to result in
less toughening, and plastic deformation spread along the interface in front of the crack.

Cao et al. [1988] showed that the cracking behaviour of brittle solids bonded by a ductile
layer is influenced by the thermal expansion coefficients, elastic properties and fracture
resistance (in mixed-mode) of each material and the yield strength and thickness of the
ductile layer. They concluded that there are no unique trends in these variables, and that the
fracture characteristics of the material combination under consideration must be identified.

For crack growth at an inclined angle to the layers, a combination of effects could occur.
Crack path will be influenced by relative toughness of each material and the interfaces, and
by compliance mismatch and external loading. This could lead to a zigzag crack path, as
the crack extends along a weaker plane, inclined to the macroscopic propagation direction,
then corrects by cutting back across several layers. Furthermore, the crack could jump
across layers, resulting in bridging by intact layers, or undergo branching [Selvarathinam,
1998].

2-69

2.6

Cracks in Graded Materials

Introduction of a continuous gradient in material composition and properties alters the

fracture behaviour in four key ways, as compared to homogeneous materials.

Altered Stress Field

The spatial variation in mechanical properties, E and , leads to a stress field around a
crack that differs from that in a homogeneous material.

Under particular loading

conditions, this leads to altered values of SIF and mode-mixity, as compared with a cracked
homogeneous material under the same conditions. Consequently, fracture loads and crack
paths are affected by the presence of a material gradient.

Spatial Variation in Crack-Tip Toughness

The variation in composition across a graded region is often associated with a changing
intrinsic fracture toughness, K0, which varies as a function of position within the gradient,
ie. K0(a).

Spatial Variation in Crack Extension Toughening

The variation in composition and microstructure across a graded region, results in a spatial
variation in the crack-growth toughening mechanism characteristics. The closure tractions
that act along the crack length therefore depend, not only on location behind the crack tip
(as in homogeneous materials), but also on position of the crack within the graded region.

Residual and Thermal Stresses

As an FGM undergoes a change in temperature, it may experience a stress differential due

to spatially varying thermomechanical properties, resulting in residual thermal stress
distributions, and thermal loading. In FGMs the residual thermal stresses, which can vary
in sign (compressive or tensile) and magnitude across the graded region [Freund, 1993,
Shabana & Noda, 2001, Itoh & Kashiwaya, 1992], have been shown to influence crack-tip
stress fields and the fracture behaviour [Lin & Miyamoto, 1999, Chung et al., 2002].
Residual stresses in FGMs have been calculated using finite element analysis [Williamson
et al., 1993, Dao et al., 1997, Becker et al., 2000] and analytical methods [Freund, 1993,
Ravichandran, 1995, Zavalangios, 1997, Xu et al., 1999, Shabana & Noda, 2001,
Pitakthapanaphong & Busso, 2002, Tsukamoto, 2003].

2-70

(a)

(b)
Stress
Contour
Gradient
Stress
Contour

Gradient
Figure 2.28: FGMs with cracks initially oriented (a) parallel to the gradient and (b) perpendicular
to the gradient. Representative constant crack-tip stress field contours are shown. Material gradient
introduces an asymmetry that can lead to skewing of the crack-tip stress-fields.

2.6.1 Stress Singularity

It has been shown by Delale and Erdogan [1983] and Eischen [1987] that, for a continuous
and piecewise differentiable material property variation, the (1/r) crack-tip stress
singularity is the same in graded materials as in homogeneous materials (cf. Equation 2.20),
due to similarly structured integral equations defining the crack problems for each case.
This was confirmed by Jin and Noda [1994] for elastic and plastic crack-tip fields, who also
showed that the angular distribution functions, fij(), are the same.
Eischen [1987a] and Jin and Noda [1994] showed that crack-tip stress, ij, and
displacement, ui, fields in FGMs have the same form as those in homogeneous materials.
The latter are expressed in Cartesian coordinates as:

ui

K (2 + 2 tip ) 8r ( )
g i ()

E tip

i = 1,2,3

(2.57)

where K ( = I,II) are the stress intensity factors (SIFs), r is the distance from the crack
tip, is the angle from the crack-tip axis in the x1 direction, is the Kronecker delta, T is
the transverse (T) stress, parallel to the crack-tip axis, ECT and CT are Youngs modulus
and Poissons ratio at the crack tip and gi()() are angular functions given inter alia by
Anderson [1995].
2-71

These results imply that the Stress Intensity Factor remains valid as a determinant fracture
parameter. Similar results were found for dynamically propagating cracks by
Parameswaran and Shukla [1999].

The anomalous stress behaviour around cracks at

bimaterial interfaces is avoided by a continuous, piece-wise differentiable variation in

properties [Chen & Erdogan, 1996]. Additionally, the effect of spatial variation of
Poissons ratio on stress singularity was found to be negligible [Delale & Erdogan, 1983].
Consequently, a constant value of Poissons ratio may be assumed without inheriting
significant errors.

The SIFs for a crack in a graded material depend on the mechanical property distribution
across the gradient and the local properties at the crack tip. Accordingly, SIFs is influenced
by crack direction, property profile, crack-tip position and specimen geometry.

2.6.2 Stress Field Asymmetry

The variation in material properties across a graded region represents a material anisotropy,
which results in orientation-dependent fracture behaviour. Konda and Erdogan [1994]
considered the dependence of SIF on relative angle between the crack and the material
gradient, and predicted an asymmetry in the stress field around a crack, potentially resulting
in mixed mode loading. The orientation-dependent fracture behaviour in a graded region
may be illustrated by considering the following two limiting cases:

Crack propagation parallel to the gradient direction: The symmetric crack-tip stress
field (Figure 2.28(a)) means that the crack does not deflect when extending through the
graded region, although a variation in effective fracture toughness occurs.

Crack propagation perpendicular to the gradient direction. The asymmetric stress field
around the crack tip (Figure 2.28(b)) induces mixed-mode loading at the crack-tip
causing crack deflection. As the crack extends asymmetrically, the crack shape will
change along with the local properties experienced by the crack tip.

The theoretical and experimental results for each of these cases will be treated separately in
Sections 2.7 and 2.8.

2-72

2.6.3 Calculation of Stress Intensity Factors

Confirmation of the validity of the SIF as a determinant fracture parameter in graded
materials and subsequent prediction of SIF values for various loading and gradient
situations has constituted the major area for theoretical work regarding FGMs so far. A
variety of approaches have been used, generally involving specialised mathematical
techniques, including integral equation methods, finite element modelling (FEM) and
weight functions.

A number of researchers [Gibson, 1967, Plevako, 1972, Delale & Erdogan, 1983, Kassir,
1972, Lekhnitskii, 1968, Sih & Chen, 1981, Dhaliwal & Singh, 1978, Atkinson & List,
1978, Gerasoulis & Srivastav, 1980, Delale & Erdogan, 1988, Erdogan, 1985, Ang &
Clements, 1987] have considered deformation and crack problems in inhomogeneous
media, some long before the term FGM came into use. They generally assumed a simple
spatial variation of elastic properties, from which analytical solutions were obtainable.
More recently, the determination of stress intensity factor expressions via analytical
methods for various cracks in FGMs has been lead by Erdogan and co-workers. Alternative
perturbation-based approaches have been used by Ang and Clements [1987], Gao [1991]
and others [Jivkov, 1999, Jivkov & Stahle, 2003, Erguven & Gross, 1999].

The weight-function technique for estimating SIF developed by Bueckner [1970] has been
used by Fett and co-workers for graded materials [Fett & Munz, 1997]. They showed that
weight functions can be derived for specific gradient profiles, accounting for the influence
of the elastic property variation, and the results will differ from those calculated using the
weight functions for homogenous materials [Fett, 1992, Fett et al., 2000]. Similarly, they
showed the Petroski-Achenbach technique may be applied to graded materials [Fett et al.,
2000]. A similar weight function approach was used by Santare and co-workers [2002].

Finite element modelling has been used for simulating fracture in graded regions [Bao &
Wang, 1995, Marur & Tippur, 2000, Chapa-Cabrera & Reimanis, 2002, Becker et al., 2001,
Jin et al., 2002]. Certain techniques within finite element modelling, such as domain
integral [Gu et al., 1999] and multiple isoparametric methods [Zou et al., 2000] and
2-73

internally graded elements [Santare & Lambros, 2000], have been developed specifically
for application to FGMs. In particular, methods for determining SIF and J-integral have
been modified to account for the material gradient [Dolbow & Gosz, 2002, Dolbow et al.,
2000, Chen et al., 2000a, Wu et al., 2002]. These were summarised by Kim and Paulino
[2002a].

In calculating fracture parameters for graded materials, certain issues arise. The techniques
used for homogeneous materials assume a uniform stress-strain relation around the cracktip, however in graded materials this may not be the case. This discrepancy must either be
ignored, thereby reducing the accuracy of modelling predictions, or accounted for by
modifying the calculation techniques.

There is a term in the derivation of the J-integral that depends on material property
gradient. In homogeneous materials this is equal to zero and the J-integral is pathindependent, however in graded materials this is not the case. Gu et al [1999] used a
domain integral formulation similar to that in homogeneous materials, calculated over a
very small domain near the crack tip, over which the gradient may be ignored. This yielded
accurate results, however a very fine, and computationally expensive, mesh was required
near the crack tip. The advantage of the J-integral as a remote method is lost in this
particular formulation for FGMs. A modified J-integral for graded materials has been
proposed by Eischen [1987], Chen et al. [2000] and others. This includes a term involving
material stiffness-gradient adding to Equation 2.29, thereby restoring path independence:

J = ( W1i ij
O

du j
dx 1

)n i d

1
2

ij

dC ijkl
dx 1

kl d

(2.58)

where C is the elastic property tensor, is the strain and summation applies over i,j,k,l in
the second integral. This modification may be obtained by subtracting the path-dependent
term from the original J-integral [Chen et al., 2000], and was also derived more rigorously
by Eischen [Eischen, 1987]. The modified J-integral has been applied in element-free
Galerkin method [Chen et al., 2000] and FEM formulations, and extended to mixed-mode
loading situations [Kim & Paulino, 2002a] and dynamic fracture [Wu et al., 2002]. The
drawback of the modified J-integral is that the stiffness-gradient term must be integrated
2-74

across a domain surrounding the crack tip. Accordingly a path integral formulation is no
longer adequate and calculation becomes more involved. Similarly, a modified crackclosure integral has been used with functionally graded materials [Eischen, 1987, Becker et
al., 1987, Kim & Paulino, 2002a, 2003b].

For mixed mode loading, a second integral, J2, called the product integral, has also been
derived [Eischen, 1987b, Raju & Shivakumar, 1990]. From J1 (the original J-integral) and
J2, both KI and KII may be extracted through iteration [Kim & Paulino, 2002a]. Another
method involves the interaction integral, or M-integral, which relates the J-integrals for two
different stress states [Yau et al., 1980]. This has been applied to FGMs in both FEM
[Dolbow & Gosz, 2002, Kim & Paulino, 2003b] and Galerkin method [Rao & Rahman,
2002] formulations.

Kim and Paulino [2002a] compared the displacement correlation, modified crack closure
integral and modified J-integral techniques for cracks in FGMs. These all appeared to
provide reasonably similar results in agreement with theoretical models. Subsequent work
showed that the interaction integral also provides similar results [2003c]. These methods
discussed have been applied extensively to straight cracks, though their application to
curved cracks has been very limited, which raises the question of their applicability to
curved cracks. Certainly calculations at the crack tip (DCT and crack closure) are not
influenced by crack curvature; however, the J-integral may not be appropriate for curved
cracks, especially over larger domains.

In the finite element modelling of graded materials, the assignment of material properties
should reflect the property distribution in the FGM structure being simulated. Whilst
significant efforts have been made to accurately incorporate continuous property variation
into finite element formulations, many experimentally produced FGMs exhibit a stepwise
variation in properties and should be modelled as such, bearing in mind that effective
property smoothing may be applicable (cf. Fig 2.7). The inclusion of spatially varying
properties in the FE formulation does not present a computational problem, as the stiffness
matrix may be determined by averaging across each element [Kim & Paulino, 2002c].
2-75

Some FE software packages allow graded elements to be explicitly defined though others
do not, necessitating circumventive measures.

The simplest method involves the assignment of properties to elements individually. This
leads to a discontinuous step-type variation in properties, though this may provide a more
appropriate representation of a real graded component than an ideal smooth gradient, as
mentioned in Section 2.1. The practicality of assigning element properties individually, or
dividing a structure into numerous areas then assigning properties to areas, can be
problematic [Bao & Wang, 1995, Bleek et al., 1998, Li et al., 2000]. Santare and Lambros
[2000] developed a formulation for graded elements, which automatically interpolate
material properties within the element. These were partially successful in improving
calculation accuracy, as compared with theoretical results, though not in all cases. Li et al.
[2000b] and Kim and Paulino [2002c] have also investigated elements with an internal
property gradient, utilising generalised isoparametric formulations, and reached similar
conclusions. Mean properties, calculated by integration within each element, were used for
the stiffness matrix.

A novel technique used by Rousseau and Tippur [2000] and others [Leck, 2002] involves
definition of properties as function of temperature then assignment of temperature values to
nodes. As the FE formulation leads to an interpolation of temperatures within the elements,
this results in a continuous variation in properties. Artificially assigning temperatures
across a structure is inherently aphysical; hence temperature is no longer a physically
meaningful property. In this case, coefficient of thermal expansion is set to zero to avoid
thermal stresses.

As discussed in Section 2.4, the stress intensity factor (or J-integral) may be circumvented
by directly incorporating crack-tip cohesive tractions. This approach has been used by Jin
and Dodds to model mode I crack growth in TiB/Ti composites and FGMs [2002, 2004],
and by Wang and Nakamura [2004] for elastic-plastic FGMs.

2-76

Various experimental techniques have been used to validate theoretical predictions of stress
fields and SIFs in graded materials. Along with conventional fracture and fatigue testing
methods, stress and strain field measurement using Moir interferometry [Chapa-Cabrera et
al., 2000, Steffler et al., 2000, Winter et al., 2000], coherent gradient sensing [Rousseau &
Tippur, 2000, Xu et al., 1999] and digital image correlation [Lambros & Abanto-Bueno,
2002] have been used to obtain additional information. Experimental results have generally
shown good agreement with theoretical and computational predictions.

2.7

Crack Propagation Parallel to Gradient

For crack propagation parallel to the gradient direction (Figure 2.28(a)), the stress field
around the crack tip will be symmetrical and the crack is expected to continue to extend
directly through the gradient. A surface crack in a graded coating is one example of this
type of crack configuration. The results of a number of studies investigating the theoretical
[Erdogan & Wu, 1997, Jin & Batra, 1996, 1998, Bao & Wang, 1995, Bleek et al., 1998]
and the experimental [Carpenter et al., 1999, Li et al., 2000, Chung et al., 2001, Lin &
Miyamoto, 1999, Moon et al., 2001, 2002a, Rousseau & Tippur, 2001a, 2001b, 2002, Hill
et al., 2002] fracture behaviour of this crack configuration are discussed below.

2.7.1 Stress intensity factor calculations

Predicted values for the stress intensity factor for cracks parallel to the gradient differ from
those in homogeneous materials and near interfaces. The SIF and crack-tip stress fields in a
graded material depend on the direction of crack propagation. The SIF for a crack oriented
towards the stiffer material tends to be higher than for a crack in a homogeneous material,
whereas a crack oriented towards the more compliant material has a lower SIF [Rousseau &
Tippur, 2000, Jin & Batra, 1996, Delale & Erdogan, 1983, Konda & Erdogan, 1994, Jin &
Batra 1998, Rousseau & Tippur, 2001]. This is illustrated in Figures 2.29(a) and 2.29(b),
which show the results of Delale and Erdogan [1983] for internal cracks and Erdogan and
Wu [1997] for surface cracks respectively. These results have been verified via FEM [Kim
& Paulino, 2002a, Anlas et al., 2000].

2-77

Increasing positive
stiffness gradient

Gr

Normalised Stress Intensity Factor, K /K

Hom

1.5

1.25

E(x) = Eoex

2a

0.75
Increasing negative
stiffness gradient

0.5
0

0.2

0.4

0.6

0.8

Normalised Crack Length, a/l

1.4

Gr

Normalised Stress Intensity Factor, K /K

Hom

1.5

E /E = 10

1.3

1.2

E /E = 5
2

1.1

E2

E /E = 1
2

E1
E /E = 0.2
2

0.9

E /E = 0.1
2

0.8
0

E(x)

0.1

0.2

0.3

0.4

0.5

Normalised Crack Length, a/h

Figure 2.29: Stress intensity factor (SIF) for a crack in a graded material, KGrad, normalised with the
SIF for a corresponding crack in an homogenous media, KHom, vs. normalised crack length, a, for
(a) an internal crack [Delale & Erdogan, 1983] and (b) an edge cracked beam [Erdogan & Wu,
1997]. Crack length normalised by (a) 1/ and (b) h. Crack is oriented parallel to an exponential
gradient as in Equation 2.1, with x1 = x.

2-78

Comparing cracks in FGMs to those in bimaterials, the crack-tip stresses within a graded
region has been found to be considerably less than those for a crack situated near a
bimaterial interface [Erdogan, 1995b, Bao & Wang, 1995, Bleek et al., 1998]. This is
demonstrated in the results for a thermally loaded graded interlayer [Bleek et al., 1998] in
Figure 2.30, and in similar results for periodic surface cracks [Erdogan & Ozturk, 1995]
and edge cracked beams [Jin & Batra, 1998, Bao & Wang, 1995] under mechanical
loading. Although results for thermal loading and mechanical loading situations may differ
somewhat [Bao & Wang, 1995, Jin & Batra, 1996, 1996b], the tendency for reduction of

Normalised Stress Intensity Factor, K/K

norm

crack-tip stresses compared to bimaterial interfaces remains consistent.

1

Surface Layer

Graded Interface

0.8
Crack approaching
Bimaterial Interface

n=2

0.6

n=1
n = 0.5

0.4

0.2

0
0

0.02

0.04

0.06

0.08

0.1

Normalised Crack Length a/h

Figure 2.30: Stress intensity factor, K, under thermal load for cracks approaching and within a
graded interface normalised by a thermal stress intensity factor, Knorm = ETh, vs crack length, a,
normalised by total specimen thickness, h. and E are the thermal expansion coefficient and
Youngs modulus of the surface layer. Gradient profile given by Equation 2.2 [Bleek et al., 1999].

Material gradient profile and crack position within the gradient have been predicted to
influence the SIF significantly. Generally, the deviation of SIF from that in a homogeneous
material increases with increasing gradient steepness [Erdogan, 1995b, Jin & Batra, 1996,
Delale & Erdogan, 1983, 1998a, Wang & Gross, 2000], as shown in Figure 2.30. Gradient
shape also has a significant effect on SIF, with exponential-type (n>1) gradients predicted
2-79

to yield lower SIFs than logarithmic-type (n<1) gradients [Bao & Wang, 1995, Wang &
Gross, 2000, Bleek et al., 1998]. For a given gradient shape, the crack-tip position within
the graded region will have a major influence on SIF [Jin & Batra, 1998, Bleek et al.,
1998]. The results of Bao and Wang [1995] for periodically edge-cracked graded coatings,
shown in Figure 2.31, illustrate the effect of crack length and gradient shape on mechanical
energy release rate. Furthermore, the influence of the graded profile on the variation of SIF
with crack length, as compared to the homogenous situation, is dependent on the specimen
geometry and loading configurations [Jin & Batra, 1996, Delale & Erdogan, 1983, 1988b,
El-Borgi et al., 2001a]. The theoretical results of Jin and Batra for an edge-cracked beam
suggest a difference in the extent of deviation between the homogeneous and the graded
material for differing loading geometries [Jin & Batra, 1996, 1998].

Gr

Normalised Energy Release Rate G /G

Hom

14

n=2
12

n=1
10

n = 0.5

0.2

0.4

0.6

0.8

Normalised Crack Length, a/b

Figure 2.31: Mechanical energy release rate, GGr, for periodic cracks in a graded ceramic/metal
surface layer normalised with mechanical energy release rate for cracks in a homogenous medium,
GHom, vs. crack length, a, normalised by surface layer thickness, b. Gradient shape is described by
Equation 2.2 [Bao & Wang, 1995].

2-80

2.7.2 Intrinsic fracture toughness variation

The effect of the spatial variation of composition and corresponding microstructure, on the
effective fracture toughness for crack propagation, KR(a), has been modelled [Jin & Batra,
1996, 1998, Bao & Wang, 1995, Cai & Bao, 1998] and experimentally verified [Li et al.,
2000, Chung et al., 2001, Moon et al., 2001, 2002a, Fukui et al., 1997].
2000

100

Nickel-rich

1500
Toughness - FGM

1/2

Remanent Strength Homogeneous Composite

80

60
1000
Toughness - Homogeneous Composite
40

Remanent Strength, [MPa]

Fracture Toughness, K [MPa m ]

Alumina-rich

500

Remanent Strength - FGM

20

0.1

0.2

0.3

0.4

0
0.5

Normalised Crack Length, a/h

Figure 2.32: Fracture toughness, KC, and remanent strength, R, of an edge-cracked beam of Al2O3Ni loaded in bending, vs. crack length, a, normalised with beam height, h. Comparison is made
between beams of graded and homogenous composition. [Jin & Batra, 1998].

For the case in which the crack travels through a graded region from the more brittle to the
more ductile material, the increasing intrinsic fracture toughness, K0(a), will cause Kc(a) to
increase with crack length, resulting in a rising R-curve behaviour.

Note that in a

homogeneous material, Ko is constant and does not contribute to the R-curve behaviour of
the material. Variation in fracture toughness can lead to an enhanced R-curve, which may
result in stable crack propagation and crack arrest and hence an increase in the remnant
2-81

strength of a cracked component. Enhanced R-curve behaviour was predicted by Jin and
Batra [Jin & Batra, 1996] as shown in Figure 2.32. It has been experimentally observed in a
wide range of FGMs: irradiated ECO polymer [Li et al., 2000], co-continuous Al-Al2O3
[Chung et al., 2001], Al-Al3Ni [Fukui et al., 1997], PSZ-steel [Tohgo et al., 2001] and TiTiB [Hill et al., 2001]. In each case, the crack propagated from the more brittle to the more
compliant material. Cracks in the opposite direction tend to experience a diminishing
toughness [Jin & Batra, 1996], which would generally result in unstable crack propagation.
Similar results have been obtained by Wang and Nakamura [2004] and Jin and Dodds
[2004], for elastoplastic fracture in FGMs simulated with FE and cohesive-zone models. A
variation in ductility and cohesive traction profile resulted in a variation in crack growth
resistance as the crack propagated through the gradient. This means that steady-state
propagation is not reached as it would be in a homogeneous material [Wang & Nakamura,
2004]. Jin and Dodds predicted three-dimensional effects for cracks in four-point bend
specimens, with the crack tunnelling ahead inside the specimen and lagging behind at the
surfaces. Further studies of crack propagation in FGMs by Ma et al. [2004], Lee [2004] and
Batra and Love [2005] have yielded similar results.

2.7.3

Variation in crack-extension toughening

FGMs have a spatially varying composition and microstructure hence the magnitude of the
crack closure stresses, resulting from wake effects along the crack length, will be dependent
on the compositional profile. Spatially-dependent crack bridging has been incorporated into
several models [Jin & Batra, 1998, Moon et al., 2002a, Cai & Bao, 1998] and its resultant
influence on the crack growth resistance behaviour derived. Cai and Bao [1998] predicted a
significant increase in effective fracture toughness due to bridging, as shown in Figure 2.33,
thus increasing the possibility of stable crack growth or crack arrest. This has been
observed experimentally in a number of studies [Chung et al., 2001, Moon et al., 2002a].
Jin and Batra predicted a significant dependence of R-curve behaviour on initial flaw size
due to spatial variation in bridging (Figure 2.34). As crack extension toughening effects
differ greatly between materials, their influence on the effective toughness will vary also,
and will be less universal than the influence of spatial variation of intrinsic toughness.

2-82

No bridging

Gr

Normalised Energy Release Rate, G /G

Hom

3.5

2.5

= 2.5

1.5

appl

=5

appl

= 7.5

0.5

appl

0
0

0.2

0.4

0.6

0.8

Normalised Crack Length, a/b

Figure 2.33: Mechanical energy release rate, GGr, for bridged periodic cracks in a graded
ceramic/metal surface layer on a metal substrate (normalised with energy release rate for cracks in
an homogenous layer, GHom) vs. crack length, a, normalised by surface layer thickness, b. Applied
stress, appl, is constant. The effect of bridging traction stress, 0, in terms of appl is shown. [Cai &
Bao, 1998].

35
30

1/2

Fracture Toughness, K [MPa m ]

40

25

20
15
10
5
0
0

0.1

0.2

0.3

0.4

0.5

Normalised Crack Length, a/h

Figure 2.34: Effect of initial crack size on R-curve behaviour for bridged cracks in ZrO2-Ti graded
beams under bend loading. Initial crack sizes, a0/h, are A: 0.01, B: 0.1, C: 0.2. [Jin & Batra, 1998]

2-83

2.7.4 Residual stress distribution

The spatial variation of composition within an FGM composite may result in residual
thermal stress distribution across the entire composite, which can have a significant effect
on the effective fracture toughness. Enhanced fracture toughness has been partially
attributed to residual stresses in Al2O3/TiC/Ni FGMs [Lin & Miyamoto, 1999], Al-Al2O3
FGMs [Chung et al., 2001, 2002] and alumina-zirconia composites [Moon et al., 2001,

1/2

40
Surface Layer
35

(a)

Effective Fracture Toughness, K [MPa m ]

2002a, 2002b].
Intermediate
Layer

Central Layer

30
25
20
15
Homogeneous Al O
2

10

5
0
0

0.1

0.2

0.3

0.4

0.5

600

NT

Residual stress
400
10
200

Notch-tip stresses
Tensile

Compressive

-200

0
0

0.1

0.2

0.3

0.4

Residual stress, [MPa]

Normalised Notch-tip Stress, /

(b)

15

-400
0.5

Normalised Notch Length, a/h

Figure 2.35: (a) Effective fracture toughness, KR, and (b) residual stress, R, and normalised notchtip stress, NT/R, for a graded Al2O3-Ti/Ni-C edge-cracked beam, as a function of notch length, a,
normalised by beam thickness, h. NB: This testing was conducted using multiple samples with
varying initial notch depths, so there was no influence of crack bridging. [Lin & Miyamoto, 1999].

2-84

As residual stresses vary with composition, a crack along the gradient direction will
experience a variation in residual stress as it propagates.

Thus, the residual stress

distributions can also contribute to the anisotropic nature of the FGM fracture behaviour.
For a crack propagating through a graded region in the direction of decreasing compressive
stress, decreasing crack growth resistance behaviour may result, whereas an increasing
compressive stress leads to enhanced crack growth resistance. This is apparent in the results
of Lin and Miyamoto [1999] for alumina-titanium carbide-nickel FGMs, as in Figure 2.35.
In this case, compressive residual stresses in the surface and intermediate layers produced
enhanced R-curve behaviour (compared with the toughness of monolithic alumina).

2.7.5 Resultant fracture behaviour

Experimental observations of fracture in FGMs are a product of most if not all of the
factors discussed above. Spatial variation of composition has been considered to affect the
measured toughness behaviour by influencing [Moon et al., 2002a]: (i) intrinsic fracture
toughness, Ko(a), (ii) crack bridging toughening, Kbr(a,a), and (iii) residual stress
toughening, Kr(a). The resulting effective fracture toughness, for mode I loading, may be
described as a modification of Equation 2.42 on page 2-45:

K eff
c ( x, a) = K 0 ( x tip ) + K br ( x br , a) - K r ( x tip )

(2.59)

where each component is a function of the location within the graded region. Such an
expression allows one to observe the influence of the crack-tip location within the graded
region on each component and on the resulting effective fracture toughness (R-curve).

This analysis was utilized by Moon et al. who investigated the crack growth resistance
behaviour of homogeneous and graded alumina-zirconia composites [Moon et al., 2002a,
2002b]. They determined that the material gradient had a significant influence on the
measured R-curve behaviour and, by partitioning the contributing components as in
Equation 2.59, that the residual stress distribution dominated measured crack growth
resistance behaviour for this material system, as shown in Figure 2.36.

2-85

First Layer

Layer
Interface

2
Gradient Direction
1
0

200

400

600

Crack Extension, a [m]

1/2

Stress Intensity Factor Components [MPa m ]

Second Layer

(a)

1/2

Effective Fracture Toughness, K [MPa m ]

K(b)
o

K
1

-1
0

br

200

400

600

Crack Extension, a [m]

Figure 2.36: (a) Measured effective fracture toughness, KR, and weight-function based predictions
versus crack extension, a, in alumina-zirconia graded layered composites. (b) Stress intensity
factor representation of components of the predicted KR; intrinsic toughness K0, bridging Kbr, and
residual stress Kr, versus crack extension, a. Note the significant effect of residual stress upon
predicted results. [Moon et al., 2002].

Chung et al. [2001, 2002] measured the effective crack growth resistance behaviour of AlAl2O3 FGMs, with an interpenetrating network structure (Figure 2.37). They found that the
2-86

presence of a compositional gradient influences the measured R-curve, as compared with

that of the homogenous material, and that the gradient shape had a large influence. Crack
bridging was shown to occur in homogeneous samples, whilst modifications of this R-curve
were attributed to the variation in property gradient in the different FGM samples (with
n>1, n=1, and n<1, as in Equation 2.2). Observed crack growth resistance behaviour was
attributed to the fracture toughness variation, crack bridging variation and also to residual
stresses [Bleek et al., 1998].

1/2

Effective Fracture Toughness, K [MPa m ]

15

FGM, n = 1/3

Homogeneous
composite
10

FGM, n = 1
FGM, n = 3
0
0

10

12

14

16

Crack Extension, a [mm]

Figure 2.37: Effective fracture toughness, KR, vs. crack extension, a, for Al-Al2O3 composites.
Shown are results for a homogeneous composite (13% Al), and graded composites (FGM) with
compositional profiles, n=1/3, n=1 and n=3 as per Equation 2.2. [Chung et al., 2000].

2-87

2.8

Crack Propagation Perpendicular to Gradient

An initial crack or flaw orientated perpendicular to the gradient (Figure 2.28(b)) can
experience gradient-induced mixed-mode loading at the crack tip, which may result in
deflection of the crack as it extends. Factors influencing the initial crack extension and
continued crack propagation include crack deflection criteria and crack morphology, along
with variations in intrinsic fracture toughness, bridging stresses, and residual stresses. This
fracture configuration has been investigated theoretically [Erdogan, 1995b, Gu & Asaro,
1997a, Konda & Erdogan, 1994,153], experimentally [Chapa-Cabrera et al., 2000,
Rousseau & Tippur, 2000, Hoffman et al., 2001] and with numerical modelling [Marur &
Tippur, 2000, Hoffman et al., 2001, Chapa-Cabrera & Reimanis, 2002, Becker et al., 2001,
Kim & Paulino, 2004].

Gr

1.4

0.25

1.2

0.2

1
0.15

0.8

0.6

0.1

0.4

0.2

0.2

2a

0.4

Mode Mixity, [radians]

Normalised Stress Intensity Factor, K /K

Hom

E(y)

0.05
II

0.6

0.8

Normalised crack length, a/l

Figure 2.38: Stress intensity factors, KI and KII, for an internal crack oriented perpendicular to the
gradient direction normalised with stress intensity factor for a homogenous material, KHomog and
mode mixity, , vs. crack length, a, normalised by l =1/. Gradient has an exponential shape, as
given by Equation 2.1, with x1 = y. Note that both KI and KII, along with mode mixity, , increase
with an increase in gradient steepness. [Konda & Erdogan, 1994].

2-88

2.8.1 Stress intensity factor calculations

The existence of a material gradient has been predicted to modify the mode-mixity and
influence the magnitude of the mechanical energy release rate [Gu & Asaro, 1997a, Marur
& Tippur, 2000, Konda & Erdogan, 1994] and has been experimentally verified [ChapaCabrera et al., 2000, Butcher et al., 1999]. Due to the asymmetric stress-field, the effective
mode-mixity at the crack tip often differs from the applied mode-mixity; for a crack under
Mode I applied loading, this leads to a mode II loading at the crack tip, as shown in Figure
2.38. In addition, the mode I SIF tends to increase, as compared with a similar crack in a
homogeneous material, which leads to an increase in mechanical energy release rate.
Crack-tip stresses are reduced however, when compared to a crack along or parallel to a
bimaterial interface [Gu & Asaro, 1997a, Chen & Erdogan, 1996]. As the gradient
steepness increases, the crack-tip stresses approach those at a bimaterial interface, whereas
a decrease in steepness results in stresses approaching those of a homogeneous material
[Marur & Tippur, 2000] (Figure 2.39).
1.5
Bimat

: Bimaterial Interface

: FGM, ~ 0.5

Gr

1
1.25

: FGM, ~ 0
0.8

: FGM, ~ 1

0.6

0.75

0.2

G: FGM, ~ 1
0.5
0

20

40

60

80

100

120

140

Bimat

G: FGM, ~ 0.5

Gr

G: FGM, ~ 0
0.4
G: Bimaterial Interface

Normalised Mode-Mixity, /

Normalised Energy Release Rate, G /G

1.2

0
160

-1

Gradient Steepness, 1/w [m ]

Figure 2.39: (a) Mechanical energy release rate for a crack within a linear gradient region, GGrad,
normalised with mechanical energy release rate for an equivalent crack in a bimaterial interface,
GBimat, and (b) normalised mode mixity, Grad/Bimat, plotted vs. gradient steepness, given as the
inverse of gradient width, w-1. Cracks are perpendicular to the gradient, with a range of values of
normalised crack position, . [Marur & Tippur, 2000].

2-89

60

4PB, = 0.1

(a)
Mode-Mixity Phase Angle, [deg]

4PB, = 0.5
40

4PB, = 0.9
DCB, = 0.9
20

DCB, = 0.5
0

DCB, = 0.1
-20

-40
0

0.4

0.8

1.2

1.6

2.4

2.4

Gradient Steepness, w
5

Normalised SIF Magnitude, |K|/K

Norm

(b)

DCB, = 0.5
4PB, = 0.5

DCB, = 0.3
2

DCB, = 0.1
4PB, = 0.3

4PB, = 0.1
0

0.4

0.8

1.2

1.6

Gradient Steepness, w
Figure 2.40: (a) Stress intensity factor magnitude, K =

K I2 + K 2II , normalised as in [26], and (b)

mode mixity, , plotted vs. gradient steepness, w, for cracks of different relative crack position,
, in 4-Point Bending (4PB) and Double Cantilever Beam (DCB) specimen. Note effect of
specimen geometry. [Gu & Asaro, 1997a].

2-90

The influence of material gradient on SIF magnitude and mode-mixity has been predicted
to depend strongly on the localised property variation, which in turn depends on global
property profile and crack position within the gradient [30,153]. An increase in gradient
steepness results in an increase in the effective mode-mixity phase angle as KII increases
more significantly than KI [Gu & Asaro, 1997a, Konda & Erdogan, 1994]. The results of
Gu and Asaro (Figure 2.40) indicate that variation in crack position may result in
substantial variation in both mechanical energy release rate and mode-mixity, so that a
crack located on one side of the gradient would experience a totally different environment
to a crack located on the other side [Gu & Asaro, 1997a].

Crack Deflection Angle, [degrees]

60

4PB ( = 0.5)

50

CCP ( = 0.9)
40
30

DCB ( = 0.5)

20
10

CCP ( = 0.5)

CCP ( = 0.1)
-10
-20

0.4

0.8

1.2

1.6

2.4

Gradient Steepness, w
Figure 2.41: Crack deflection angle, k, (from MTS criterion) plotted against gradient steepness,
w, for cracks of different relative crack position, , in Centre-Cracked Plate (CCP), 4-Point
Bending (4PB) and Double-Cantilever Beam (DCB) specimens. Note effect of specimen geometry
and crack position on crack path. [Gu & Asaro, 1997b].

The influence of a gradient on SIF values and mode-mixity has also been shown to depend
on specimen geometry [Chapa-Cabrera & Reimanis, 2002, Cai & Bao, 1998] and loading
configuration [Konda & Erdogan, 1994, Jin & Batra, 1998, Chapa-Cabrera & Reimanis,
2002]. Gu and Asaro calculated the SIF for 4-point bending (4PB) and double cantilever
2-91

beam (DCB) specimens, and demonstrated a strong influence of specimen geometry

(Figure 2.42) [Gu & Asaro, 1997b].

(a)

1/2

SIF Magnitude, |K| [MPa m ]

25

20

With residual stress

15

10

Without residual stress

0
0

50

100

150

Applied Stress,

Appl

[MPa]

(b)
SENT
4PB
CCT

60

Crack Deflection Angle, [deg]

90

200

30

-30

-60

-90
0

50

100

150

Applied Stress,

Appl

200

[MPa]

Figure 2.42: (a) Stress intensity factor magnitude, K =

K 2I + K 2II , vs. applied stress, appl, for

single edge-notched tension (SENT) sample with and without the effect of residual stress and (b)
crack deflection angle, k, as a function of applied stress, Appl, for SENT, 4-point bend (4PB) and
centre-cracked tension (CCT) specimen incorporating the effect of residual stress. Calculations at
constant crack length, a/h = 0.4. [Chapa-Cabrera & Reimanis, 2002a].

2-92

2.8.2 Crack deflection

Mixed-mode SIF has been predicted to cause crack deflection in homogeneous [Karihaloo
et al., 1980, Nuismer, 1975, Erdogan & Sih, 1963] and graded materials [Gu & Asaro,
1997, Bao & Wang, 1995, Marur & Tippur, 2000, Chapa-Cabrera & Reimanis, 2002,
Becker et al., 2001] and has been confirmed experimentally in homogeneous materials
[Williams & Ewing, 1972] and various types of FGMs [Chapa-Cabrera et al., 2000,
Rousseau & Tippur, 2000, Hoffman et al., 2001]. Prediction of deflection angles for cracks
in graded materials under mixed-mode loading presents a further challenge. This was
considered by Kim and Paulino [2002a] and by Becker et al [2001].

Stress acting transverse to the crack direction (T-stress) is known to affect crack deflection
in homogeneous materials [Karihaloo et al., 1980, Cotterell & Rice, 1980]. Using FEM,
Becker et al. [2001] showed that T-stress influences deflection in graded materials, and that
crack-path predictions are dependent upon crack-extension increment size.

As the mode-mixity (Equation 2.35 on page 2-34) at the crack tip increases, the crack
deflection angle also increases. Accordingly, the subsequent crack deflection behaviour will
be dependent on factors that influence the crack-tip mode-mixity: gradient shape and
steepness; crack position within the gradient; specimen and loading geometry; residual
stress distribution; and crack-bridging tractions.

As the gradient becomes steeper the mode-mixity at the crack tip increases and the crack
deflection angle is predicted to be larger. Likewise, the initial position of the crack within
the graded region dictates the mode-mixity at the crack tip, and thus influences the crack
deflection angle, as shown in Figure 2.41. Additionally, different specimen and loading
geometries produce different loadings at the crack tip, influence the mode-mixity and
produce different crack deflection angles as also demonstrated in Figure 2.41.

The influence of residual stress distribution on the crack-tip mode-mixity and resulting
crack deflection angle has been investigated by Chapa-Cabrera and Reimanis [2002]. They
showed that for an initial crack extension from a pre-existing flaw, the presence of residual
2-93

stresses modified the stress distribution around the crack tip, which changed the mixedmode phase angle and influenced the crack kinking angle, as shown in Figure 2.42.
Additionally, they showed that the predicted crack deflection angle was also dependent on
the specimen and loading geometry. Figure 2.42(b) shows that predicted deflection angles
for 4PB and single-edge notch tension (SENT) geometries were opposite in sign to those
predicted for the centre-crack tension (CCT) testing geometry.

Crack-bridging is understood not to influence the initial crack deflection angle from a notch
but has an increasing influence on the subsequent crack extension. The influence of crackbridging on crack extension profile was investigated by Hoffman et al. [2001] with
experiments and analytical modelling. Due to greater shielding of Mode II SIF than Mode I,
bridging was predicted to reduce the mode-mixity and consequently reduce the kink angle.

2.8.3 Resultant Fracture Behaviour

Crack deflection has been observed experimentally in several types of FGMs [ChapaCabrera et al., 2001, 2002b, Rousseau & Tippur, 2000, Hoffman et al., 2001] with cracks
initially oriented perpendicular to the gradient, as in Figure 2.28(b). Chapa-Cabrera et al.
[2001, Chapa-Cabrera & Reimanis, 2001], investigating Cu-W FGMs with linear
composition gradient (E2/E1 ~ 3) in a 4-point bending configuration, observed significant
crack deflection, from 0o to 50o, towards the more compliant material (Cu). Asymmetric
stress fields at the crack tip were observed using Moir interferometry, confirming the
presence of mixed-mode loading at the crack tip as predicted by FEA [Also: Winter et al.,
2000, Steffler, 2001]. They observed that that crack deflection angles decreased with
decreasing gradient steepness, and that the crack deflection angle was dependent on the
crack location within the graded region. The variation of mode-mixity was primarily
attributed to elastic property mismatch although the influences of residual stress and plastic
deformation were acknowledged. In a later work [2002b], they incorporated plastic
deformation into finite element simulations and obtained predictions with and without
residual stress for the plane stress and plane strain cases. Residual stress dominated the
crack-tip stresses at low loads but became less important as applied load increased.

2-94

Plasticity was predicted to increase deflection angle. These results showed good agreement
with experimental results, as shown in Figure 2.43.

Figure 2.43: Comparison of experimental results for layered Cu-W composite FGMs with FE
predictions. Kink angle as a function of applied bending load, P/2, predicted by the finite element
model for the (a) 2 mm and (b) 4 mm thick layer specimens when the crack is within the 20%W
layer. In each figure the squares represent plane stress and the circles plane strain conditions. The
stars with error bars represent the experimental measurements. [Chapa-Cabrera, 2002].

2-95

Similar results were obtained by Hoffman et al. [2001] for alumina-polyester stepwisegraded composites, with E2/E1 ~ 100, loaded in tension. Initially cracks deflected with
angles around 40 toward the more compliant material, as shown in Figure 2.44. The
deflection angle then decreased with crack extension. This was attributed to the effect of
bridging on crack-tip mode-mixity. Alternatively, this could have been due to sample
geometry or toughness anisotropy.

Polymer

Interface
= 0.75

Region
Ceramic

Figure 2.44: Crack deflection in an alumina-composite-epoxy step-wise graded composite. Single

edge-notched specimen under tensile loading. [Hoffman et al., 2001].

Rousseau and Tippur [2000] investigated crack deflection in glass-sphere-filled epoxy

FGMs, with E2/E1 ~ 3.3, with no expected residual stresses, under 4-point bending and
observed crack propagation toward the more compliant material. The measured kinking
angles compared well to those predicted by analytical and FEM techniques using the KII =
0 criterion [Cotterell & Rice, 1980]. As seen in Figure 2.45, the crack deflection angle, as
predicted with the local symmetry criterion, was dependent on the initial notch location
within the graded region, and the crack propagated toward the more compliant material.

2.8.4 Crack propagation path

Very little work has been done on continued crack propagation in graded regions after the
initial crack deflection has occurred. Crack propagation will become more complex after
deviation from a straight path. Accurate models for the continued propagation of cracks
that were initially perpendicular to the gradient will require a method for calculating SIFs
which incorporates the effects of variations in local mechanical properties, crack-tip
toughness, bridging and residual stresses, as well as changing crack shape. To the authors
knowledge, no such model yet exists. However, a number of models have been used to
2-96

calculate SIF for curved and kinked cracks in homogeneous materials, which may be
adapted to graded materials.

Theoretical Predictions
Experimental Observations
Computational Predictions

Crack Deflection Angle, [deg)

10

0.2

0.4

0.6

0.8

Crack position,
Figure 2.45: Crack deflection angle, K, vs. normalised initial crack position ( = x 1 d ) in a
graded glass sphere/epoxy composite with crack propagating perpendicular to the gradient:
experimental observations and analytical and numerical predictions. [Rousseau & Tippur, 2000].

Kim and Paulino [2004] have applied their previous work of cracks under mixed-mode
loading in FGMs to predicting propagation paths and critical load profiles. Their
simulations of the experiments of Rousseau and Tippur [2000] showed good agreement
with the observed crack trajectories. They considered linear-elastic fracture without any
process-zone or bridging effects.

An extensive experimental investigation was conducted by Lyndal Rutgers, in parallel with

the present study, into stable crack propagation in alumina-epoxy composites [2004]. As
the data analysis for those experiments was conducted using the finite element model
developed in the present study, the findings are outlined in Chapter 10 rather than in the
present literature review.
2-97

2.9

Fatigue in graded materials

As discussed previously, fatigue behaviour can be complex, and spatial material variation
will further complicate cyclic fatigue predictions for FGMs. The limited work done on
fatigue crack propagation in graded materials has tended to focus on thermal fatigue
damage in the context of thermal barrier coating applications [Kawasaki & Watanabe,
1999]. Material gradients can influence cyclic deformation and crack initiation, due to
modification of the stress field and variation in local deformation properties. Hofinger et
al. [16] found that the introduction of a material gradient, to zirconia-alumina thermal
barrier coatings, reduced the amount of damage under cyclic thermal loading, by reducing
interface stresses. Similar results were reported by Balke et al. [2000, Bahr et al., 2003].
Variation in material composition was observed by Yamashita et al. [2000] to influence
cyclic deformation and crack initiation in Al-Al3Ti.
Results for crack problems apply equally well to cyclic loading as to monotonic loading
situations. Hence the SIF predictions discussed in earlier sections may be applied to fatigue
cracks also.

Fatigue crack propagation in homogeneous materials under mixed-mode

loading conditions has been investigated fairly extensively [Qian & Fatemi, 1996, Magill &
Zwerneman, 1997, Plank & Kuhn, 1999, Taylor et al., 1999, Wang & Pan, 1999, Kah Soh
& Bian, 2001]. As in the case of mixed-mode fracture however, a conclusive understanding
of crack path criteria or crack propagation rates has not yet been reached. Criteria have
been defined using various expressions for effective SIF and effective strain energy density
factor; however, they are often empirical and certainly not universally applicable.

Crack propagation under fatigue loading is influenced by material gradient, due to the
altered stress field throughout the loading cycle, and to spatial variation in fatigue
resistance. Siber et al. [1999] observed a decrease in fatigue crack propagation rate in
graded aluminium alloys as crack length increased along the gradient direction. This was
attributed to spatial variation in fatigue resistance, which correlated with hardness variation.
Cracks parallel and perpendicular to the gradient in Cu-Ni alloys were observed by Blumm
et al. [1995]. Crack kinking was observed in the latter case. Residual stress and variation
in the crack-tip plastic zone were seen to be influencing factors. Crack propagation under
2-98

fatigue loading is also influenced by residual stress, due to the modification of crack-tip
stresses under a given applied loading cycle. Berg and Wagner [1998] observed that a
surface microstructural gradient in -Ti alloy influenced both the fracture toughness and the
fatigue resistance, which was attributed to residual stresses. Residual stress was also
observed to affect fracture strength and fatigue resistance in mullite-alumina FGMs by
Bartolom et al. [1998].

Forth et al. [2003] have produced titanium alloys which, after solution-treatment and overaging, exhibited a distinct microstructural gradient near the surface. Coarse grains at the
surface resulted in higher hardness and fatigue resistance there, whilst smaller grains
resulted in greater toughness internally. This differs from the typical Hall-Petch relation
between hardness and grain-size [Ashby & Jones, 1996]. They conducted mode I and
mixed-mode fatigue tests of specimens from the surface, internal and transition regions and
observed crack growth from inclined flaws and interaction between the crack path and
microstructure. Crack paths generally agreed with MTS criterion predictions, but the
material was essentially elastically homogeneous so mixed-mode loading was not a result
of material gradient. They did not look at crack growth through a graded region or attempt
to quantify the effect of spatial microstructural variation in fatigue crack growth resistance.
They concluded that the surface treatment method resulted in improved performance under
cyclic loading.

Uzun et al. [2001] investigated fatigue crack growth in layered SiC-particle-reinforced

aluminium FGMs. By cold-pressing of two aluminium pieces, one reinforced with 10% SiC
particles, then hot-extruding, they produced cylinders with a radial compositional variation.
The residual stresses were calculated with an FE model. Fatigue cracks growing from the
reinforced to the non-reinforced side exhibited acceleration in fatigue crack growth rate,
while those growing from the non-reinforced to the reinforced side experienced a reduction
in fatigue crack growth rate. This was explained by the SiC particles causing more
deflection of the crack, which resulted in more roughness-induced crack closure, in concord
with the findings of Suresh [1985]. There were also influences of residual stress and
compliance mismatch [Kim et al., 1997].
2-99

Xu et al. [2003a, 2003b, 2004] continued research in this area and produced layered SiCparticle-reinforced aluminium matrix via powder metallurgy, resulting in gradients of width
2-10 mm, comprised of 6 steps. Fatigue cracks were grown parallel to the gradient into the
more SiC rich region. Retardation in fatigue crack growth rates was observed as the crack
grew, along with increased deflection and branching. It was concluded that effective
shielding of the crack tip, due to plastic mismatch, promoted deflection [Sugimura et al.,
1995, Kim et al., 1997] and reduced crack growth rates. Comparing the fatigue resistance
of the FGMs with equivalent homogeneous composites, they found the FGMs performed
relatively better with increasing crack extension [2003b]. They also investigated the effect
of load-ratio R on fatigue crack growth rate, and found that larger R resulted in faster crack
growth, concordant with conclusions that roughness-induced crack closure has a dominant
effect on fatigue resistance in the composite [2004]. It is noted that the variation in elastic
modulus across the gradient would have been around 1.6, but they did not appear to include
the influence of this when calculating SIFs. Comparison with theoretical results, in Figure
2.29 for instance, suggests this could have a slight (10%) influence on SIF magnitude.
2.10 Final remarks
The work summarised in this review is not an exhaustive review of the field of crack
propagation in graded materials. Erdogan and co-workers have considered a range of
specific problems not discussed here, including axisymmetric cracks [Ozturk & Erdogan,
1996], fracture under contact loading [Erdogan & Dag, 2000, Guler & Erdogan, 2004] and
under thermal loading [Lee & Erdogan, 1995, El-Borgi et al., 2000b]. Paulino and coworkers have used gradient elasticity theory to analyse a number of crack problems
[Paulino et al., 1999, Chan et al., 2000] and also considered fracture in viscoelastic media
[Paulino & Jin, 2000] which features time-dependent SIF values. Other issues investigated,
via theoretical and computational modelling, include dynamic loading [Banks-Sills, 2002]
and crack propagation [Meguid et al., 2002], thermal loading [Jin & Batra, 1996b,
Fujimoto & Noda, 2001] and plasticity effects [Tvergaard, 2002, Rashid and Tvergaard,
2003]. A wide range of mathematically intensive techniques have been used in modelling
FGMs, including group theory [Epstein & de Len, 2002], gauge field theory [Bilotsky &
Gasik, 1997], higher order tensor theory [Aboudi et al., 1999], constitutive modelling
2-100

[Nakagaki & Wu, 2001], element-free Galerkin method [Peixiang et al., 2001, Rao &
Rahman, 2003], fuzzy logic [Rabach & Lehnert, 1999] and genetic algorithms [Shimojima
et al., 1999]. Experimental work not covered here includes work on contact [Low, 1998]
and impact damage [Shah et al., 2000], initiation effects [Rousseau & Tippur, 2002] and
dynamic crack propagation [Rousseau & Tippur, 2001]. Investigation of the effects of
microstructure on crack propagation has been limited [Cannillo & Carter, 2001, Becker et
al., 2002] and is a potential field for considerable further work.

There has been some theoretical and experimental work on design optimisation [Grujicic &
Zhao, 1998, Ferrari et al., 1999, Nadeau & Ferrari, 1999, Afsar & Sekine, 2002, Bahr et al.,
2003, Cho & Shin, 2004, Shimojima et al., 1999], but this is an area which warrants further
attention. Ferrari and Nadeau considered the optimisation of FGMs in several structural
mechanics problems [1999]. They identified two complementary approaches:
(1) solving the macroscale optimisation problem for material property distribution, then
addressing implications for variation of microstructure; and
(2) considering the microstructural variation directly, then calculating the corresponding
macro-property variation, and subsequently optimising.
Shimojima et al. [1999] have used a genetic algorithm approach to find the optimal
composition profile for reducing peak thermal stress at an interface between Mo and MoSi2
subjected to prescribed thermal boundary conditions. They assumed a phenotype
comprised of a series of steps of different thickness and composition. The thickness of each
step, yi, and the weight % of MoSi2, xi, comprised the genotype. They utilised crossover
(breeding), mutation and selection operators in the genetic algorithm formulation, and
evaluated the fitness of each gradient profile using FEA. The resulting optimised gradient
profile with only four steps outperformed a linear gradient with 20 steps by 15% and an
exponential gradient with 20 steps by 2%.

In summary, the nature of crack-tip stress fields in graded materials is the same as in
homogeneous materials, which allows SIFs to be used analytically to characterise fracture
behaviour. Material property variation influences SIFs, generally tending to increase it
compared to cracks in homogeneous materials, and both the shape and scale of material
2-101

gradient have an effect. Crack position, specimen geometry and loading, residual stresses,
T-stress and crack-wake effects all influence crack growth. The crack extension direction
with respect to the gradient is a determining factor in the fracture behaviour and should be
clearly distinguished. A crack propagating parallel to the gradient experiences a variation
in crack-tip environment, including crack-tip toughness, composite properties, bridging and
residual stresses, all potentially leading to enhanced R-curve behaviour.

A crack

perpendicular to the gradient experiences an asymmetric stress field, resulting in mixedmode loading which leads to crack deflection and subsequent asymmetric propagation.

Work thus far has been predominantly based on modelling, both analytical and FEM, and
has focussed on SIF calculation for cracks in graded regions. Experimental work has been
limited by processing difficulties, though early experiments have provided some validation
to theory. Although significant advances have been made recently in the investigation of
fracture in graded materials, a consolidated understanding of the field remains elusive.

2-102

3 Investigation
This chapter describes the outstanding issues in the area of fatigue in FGMs and proposes a
number of hypotheses regarding these issues. The approach used in the present study to
address these hypotheses is then outlined.
3.1

Outstanding issues

From the review of the literature in the previous chapter, several outstanding issues
regarding crack propagation in graded materials become apparent. Further experimental
verification of analytical and computational modelling results for a range of FGMs is
clearly required. In particular, observation of crack paths, quantification of the effects of
bridging and residual stresses, crack propagation under cyclic loading, crack initiation,
effects of microstructure, plastic and viscoelastic fracture represent areas for future
experimental investigation. Key areas for further modelling work include: prediction of
crack propagation paths; interactions between different effects on crack propagation; and
comparison of existing models with experimental results to attain a more consolidated
theoretical understanding of FGMs.

Addressing all of these would be an ambitious undertaking. The present study will focus on
those issues relating directly to propagation of fatigue cracks in graded composites that
exhibit predominantly brittle failure behaviour. Several key questions that are yet to be
answered are:

How does toughness variation predict crack path? The presence of a spatial
toughness variation could be expected to influence crack growth direction, as cracks
will preferentially extend in materials of lower toughness. This may be summarised by
assuming that crack deflection follows the path where (G-R) is maximised. This is
taken into account by deflection criteria for cracks at interfaces, which also take
toughness mismatch into account, but not by criteria used for cracks in homogeneous or
graded materials. The possible influence of toughness variation must be considered, and
the effect of including it in crack deflection predictions should be quantified.

How does non-linear material behaviour influence crack path? Non-linear

deformation such as plasticity, ahead of the crack tip, or bridging, behind the crack tip,
3-1

could affect the effective mode-mixity acting at the crack tip and therefore influence
crack propagation direction. This has received limited attention for cracks in
homogeneous materials. These possible influences should be investigated.

What is the effect of crack shape and how might this be included in models? After
initial deflection, cracks in FGMs will no longer be straight, yet existing theoretical and
computational results have focussed exclusively on straight cracks. The effect of crack
shape on stress intensity factors should be quantified and comparison

What are the similarities and differences between stepped and continuous
gradients? Experimentally produced FGMs are often comprised of a series of
compositional steps separated by relatively narrow interface regions, yet theoretical
models have usually assumed continuous variation of composition. The influences of
steps in (i) elastic properties and (ii) crack propagation resistance must be examined.

How does crack propagation in graded structures under cyclic differ from that
under monotonic loading? Crack growth under monotonic and cyclic loading differs
in non-graded materials, and is expected to do so in FGMs. However, further
experimental evidence of this is required. Furthermore, crack propagation paths could
differ between monotonic and cyclic loading cases, particularly due to influences of
non-linear material behaviour as discussed above. This should be investigated
experimentally.

How may these effects be incorporated into a model for the crack propagation
process to provide accurate predictions? Simulation of crack propagation is required
for prediction and assessment of failure resistance. A theoretical or computational
model that takes into account all the pertinent influences on the crack propagation
process is required. The model should be efficient yet rigorous, and be verified by
comparison with experiments.

How could a graded join be designed to optimise its reliability under cyclic
loading? While introduction of a material gradient offers improvements in component
performance by reducing stress concentration and increasing failure resistance,
understanding of how to tailor the gradient in order to optimise these attributes is
limited. Insights gained in addressing the above issues should be utilised to develop
guidelines for optimal design of graded components.
3-2

3.2

Hypotheses

In addressing the questions above, a series of hypotheses are posed:

1. Alumina-epoxy graded specimens may be used as a model system to examine the
influences of (i) elastic gradient and (ii) fatigue crack propagation resistance gradient,
on fatigue crack growth in graded materials.
2. Fatigue cracks will deflect due to elastic gradient and this deflection may be predicted
from stress field calculations and an appropriate deflection criterion.
3. As a result of deflection, cracks will experience a variation in fatigue crack propagation
resistance as they propagate across the gradient.
4. Effective property relationships obtained for homogeneous composites can be used for
graded composites.
5. Bridging influences crack path and effective resistance to crack growth
6. Spatial variation in toughness and fatigue crack propagation resistance can have an
influence on crack propagation path, as propagation through material with lower
resistance to cracking is more energetically favourable.
7. After crack deflection, the crack-shape has an effect on crack propagation direction and
on stress concentration at the crack tip.
8. Crack propagation behaviour in stepped gradients will differ from that in continuous
gradients.
9. Crack propagation under cyclic loading will differ from that under monotonic loading.
10. The fatigue crack propagation process can be modelled, taking the various influences
into account, and providing accurate predictions for crack propagation path and failure
loads.
It is noted that the first four are generally accepted, but provide the basis for the present
study and therefore warrant statement, whilst the remaining five have not been investigated
previously.

3-3

3.3

Approach

The above hypotheses were addressed through experiments and computational simulations.

1. Finite element simulations were used for investigation of specific influences on the
crack propagation process, as it enables specific aspects of the process to be isolated and
investigated in depth. These aspects included: elastic gradient, curvature, toughness,
bridging and non-linear properties.

2. Experimental investigations focus on observing and quantifying crack propagation under

cyclic loading in a particular graded composite system. The alumina-epoxy composite
system was chosen, for several reasons:
(i) The large disparity in elastic properties, as shown in Table 1.1, which enables high
elastic property gradients to be obtained over relatively wide regions.
(ii) The absence of thermal stresses which simplifies mechanical and fracture behaviour.
(iii) Relative ease of processing.
(iv) As both phases are brittle, linear elastic fracture mechanics theory should be
applicable, leading to simple fracture behaviour. The disparity in failure strain does
present the possibility of crack bridging by the epoxy phase.
Whilst alumina-epoxy composites may not experience significant demand in industrial
applications, this is an excellent model system for investigating the effect of extreme elastic
property mismatch on crack propagation, along with the influence of crack-extension
toughening behaviour.

In the context of crack propagation, a gradient may be initially characterised in terms of the
mismatch in elastic and thermal properties, as shown in Figure 3.1. Whilst further
distinctions may be made based on mismatch of yielding and failure behaviour, this
diagram serves to illustrate the uniqueness of alumina-epoxy FGMs for providing an
extreme elastic mismatch without thermal mismatch. Accordingly, the results of this
investigation should be interpreted as a model benchmark, against which results for systems
with less elastic mismatch, but more complicated thermal, yielding and fracture behaviour,
may be compared.
3-4

When choosing a particular graded composite system for experimental investigation, the
question arises as to how the findings for that system may be applied to other types of
FGM. This is addressed in the present study through finite element simulations of other
types of FGM system.
Elastic mismatch,
E2/E1

Al2O3-epoxy

Epoxy-glass
Al2O3-Al
Al2O3-Nb

Cu-W

Al2O3-Cu

TiB-Ti
Irrad. ECO
STOA Ti
Negligible
thermal stresses

Al2O3-Zr2O
SiC-Al2O3
High thermal
stresses

Thermal mismatch,
|1-2|T

Figure 3.1: Characterisation of graded material systems in terms of elastic and thermal mismatch.
Systems included have been investigated experimentally [Al2O3-ZrO2: Moon et al., 2001, Al2O3-Cu,
Al2O3-Al: Moon, 2004, Al2O3-Nb: Matterson et al., 2004, Epoxy-glass: Rousseau & Tippur, 2000,
TiB-Ti: Carpenter et al., 1999, Cu-W: Chapa-Cabrera et al., 2000, 2002b, STOA Ti: Forth et al.,
2003, SiC-Al2O3: Xu et al., 2003a, Irradiated ECO: Li et al., 2000].

Samples were produced with an interpenetrating network structure, through a multi-step

infiltration technique. Fatigue crack propagation was observed in graded specimens.

3. Homogeneous composite specimens were also produced. Their elastic properties were
measured using the impulse excitation technique and fatigue crack propagation behaviour
was characterised. This provided comparison with graded materials, and to provide
property values for the finite element model.

4. Parallel finite element simulations of experimental specimens provided predictions for

comparison with experiment, as well as being used for analysis of experimental data.

3-5

3.4

Framework

This section introduces the framework used for describing and investigating the crack
propagation process in graded materials, which is shown in Figure 3.2. Each attribute of the
process is explained, along with the approaches used to relate them.
1 Crack Shape

& Position

6 Applied

2 Compositional

Loading

Distribution

3 Elastic Property

4 Intrinsic (CT)

Distribution
7

Deformation
& Stress Fields

Toughness

5 Extrinsic Fracture

Processes

8 Fracture

Parameters

9 Crack Propagation

Path & Rate

10 Residual Strength

& Life Predictions

Figure 3.2: Outline of the problem of crack-propagation in a graded material structure. Numbers
refer to the accompanying explanation in the text.

The problem considered is that of a graded material structure, with a crack at a given
position. At this stage, it is assumed that material deformation is linear elastic, so that crack
propagation may be analysed within the linear-elastic fracture mechanics regime. It is
assumed that the distribution of composition in the structure, denoted (1) in the diagram, is
known. In practice, this is achieved either through an understanding of the relation between
processing method and resultant composition, or through microstructural characterisation.

The compositional distribution (2) may be used to determine the distribution of effective
mechanical properties, ie. Youngs modulus and Poissons ratio (3), and also, along with
the crack position (1), the effective fracture properties at the crack tip (4,5). Assuming the
structure is subjected to a known applied loading (6), this will result in stresses and
3-6

deformations across the structure (7), which will depend on the applied load, specimen and
crack geometry and the distribution of elastic properties. The concentration of stresses
around the tip of the crack may be described in terms of fracture parameters (8), ie. stress
intensity factors or the mechanical energy release rate, which are calculated from the stress
or deformation fields. When the pertinent fracture parameter approaches the critical value,
which depends on intrinsic crack-tip toughness (4) and extrinsic toughening (5),
propagation can occur. The direction and rate of propagation (9) may then be determined
from the fracture parameters (8) and propagation resistance (4,5). Propagation of the crack
will lead to a modification of the crack geometry (1), so that an iterative analysis scheme is
required. Analysis of the overall process can allow predictions of residual strength and life
(10), and eventually lead to recommendations for optimised design against failure for
graded material structures.

Crack Shape
& Position

Applied
Loading

Compositional
Distribution

I Mixing Laws,

I Experiment, Composite

Fracture Models

Experiment 9
II Finite

Element
5 Model

Elastic Property
Distribution
6

Deformation III Fracture

& Stress Fields
Parameters

Intrinsic (CT)
Toughness
IV Failure &
V Deflection

5
VI Re-meshing

routine

Criteria

Extrinsic Fracture
Processes 8

Experiments
on Graded
Specimens 10

Crack Propagation
Residual Strength &
Path & Rate
Life Predictions
6,10

Figure 3.3: Outline of the methods used in the analysis of crack-propagation in graded material
structures. Roman numerals refer to the accompanying explanation in the text, whilst circled
numbers indicate the Chapter/s in the thesis in which the issue is primarily addressed.

3-7

The key to developing such an analysis is an appropriate choice of methods for linking each
attribute to its dependent attributes. The important relationships to be understood, and the
method/s used in the current investigation, are shown in Figure 3.3, and are as follows:
I.

Effective composite property calculations require relations between composition and

pertinent properties, which could be obtained from a relevant rule of mixtures, or from
experimental measurements on homogeneous specimens. In the current investigation,
experimental results from homogeneous specimens were used, and were compared
with theoretical mixing-law predictions. This enabled determination of the elastic
property distribution around the crack, and the local values for crack-tip toughness and
extension toughening. Production and characterisation of homogeneous specimens are
described in Chapter 4 and results are presented in Chapter 9.

II. Calculation of stress and strain fields may be conducted using an analytical or
computational model, and could be confirmed experimentally using a technique such
as photoelasticity, Moir interferometry or Raman spectroscopy. In this work, the
computational modelling approach, using finite element (FE) analysis, was chosen for
its flexibility and ease of use. Details of FE simulation methods are given in Chapter 5
and results are given in Chapters 6 to 8. Moir interferometry was used to provide
confirmation of the FE calculations, as described in Chapter 9.
III. Extraction of fracture parameters from stress and strain fields may be achieved
through a number of approaches. While several of these were compared, the
displacement correlation technique was used in general. The implementation and
validation of this are described in Chapter 5.
IV. Predictions of propagation direction can be obtained from a number of criteria, with
the local symmetry criterion being used predominantly here. This is discussed in
Chapter 5.
V. Propagation rate prediction requires an understanding of the fracture and fatigue
behaviour, in particular the crack-extension toughening effects if they are present.
Experimental characterisation of homogenous specimens was used to obtain such an
understanding for the composites studied, and this was used to form empirical rate
relations which were incorporated into the finite element model.

3-8

VI. Changes in crack geometry that accompany crack propagation require that the analysis
method be capable of handling variable crack geometry, and preferably updating
crack shape automatically. This was achieved with an automated remeshing
procedure was developed within the finite element model.

The development of this analysis methodology for the crack propagation process was
conducted in parallel with experiments on graded specimens, and results of each were
compared.

3-9

4 Procedures
This chapter describes the various experimental and computational methods employed in
the investigation; results from these procedures are presented separately in subsequent
chapters. The production and microstructural characterisation of the alumina-epoxy
composites are detailed in Sections 4.1 and 4.2 respectively. Sections 4.3 and 4.4 describe
procedures used for elastic property and fatigue crack propagation testing respectively.
Results from these experiments are presented in Chapter 9 for homogeneous composite
specimens and Chapter 10 for graded specimens. Results from Moir interferometry tests,
described in Section 4.5, are also presented in Chapter 10. Computational methods are
outlined in Section 4.6; these are addressed in significantly greater detail in Chapter 5.
4.1

Sample Processing

Alumina-epoxy composites were produced by infiltration of epoxy into open porosity

alumina preforms, which were produced by infiltrating alumina slip into pieces of opencelled polyurethane foam. This resulted in interpenetrating-network (IPN) structured
composites in which both phases are continuous [Clark, 1992, Prielipp et al., 1996, Lange
et al., 1990]. Processing of graded composites by this method was first conducted by
Cichocki et al. [1998]. Since then Neubrand and co-workers produced graded aluminiumalumina interpenetrating composites [Neubrand et al., 2002, Chung et al., 2001, 2002], and
Lyndal Rutgers adapted the approach to produce alumina-epoxy graded structures [Kidson,
1998, Rutgers, 2004]. As discussed in Section 2.2.1, composites with an interpenetratingnetwork structure differ from traditional fibre/matrix or particle/matrix composite
structures. Overviews of the production process for graded composites are shown
schematically in Figure 4.1. While the process for and homogeneous (non-graded)
composites was similar, there were several key differences as shown in Figure 4.2. Details
of each processing step follow.

4-1

Compress
Polyurethane
Foam

Pour in Al203
slurry dries

Infiltrate with
Epoxy Resin

More
Grinding,
Polishing &
Cutting

Burn out foam &

fire ceramic @
1500C

Grinding

Graded Sample

Figure 4.1 Outline of the production process for graded composite specimens.

Pour in Al203
slurry to single
foam piece

Infiltrate single
plate with
Epoxy Resin

Homogeneous plates

Cut into beams

Figure 4.2 The production process for homogeneous composite specimens was very similar to that
in Figure 4.1 for graded composites, with several exceptions.

4.1.1 Foam preparation

Open-celled polyurethane (PU) foam (Bulpren S-31048, Eurofoam, Troisdorf, Germany, 90
pores/inch) was used as an imprint for the interpenetrating network structure. It was
compressed from an initial density of 2.5% to obtain the required foam volume fractions
(2.5% to 55%). The foam was immersed with water and frozen to retain shape, then was cut
to size, as calculated to obtain the required compression ratios, with a bandsaw (ESA5
Heska Maschinenfabrik, Germany). After thawing and drying, the foam pieces were
4-2

compressed and heated in a hot press (Carver Lab Press 2697, W. Carver & Co., Melbourne
w/ Eurotherm controller, Stuttgart) to retain their compressed shape. The final height of the
foam was set with the use of aluminium spacers. A pressure of 300 kPa was applied. The
temperature was ramped up to 160C at 1C/minute, then held for 1 hour. After cooling to
below 50C, at approximately 1C/minute, the foam was removed from the hot-press. Some
buckling occurred during compression, resulting in unevenness at the edges, so these were
cut off before the next stage. For foam pieces with high compression ratios (> 5), initial
compression was done inside a metal tube with rectangular cross-section (60 x 40 mm2) to
reduce the extent of buckling. The foam was compressed inside the tube and heated in a
box furnace (HPFT 6kW, Ward, Melbourne with HE388Z 240V electronic controller,
Harco Electronics, Melbourne).

4.1.2 Slip-casting & drying

A colloidal suspension of alumina particles was infiltrated into the PU foam under varying
pressure, cast and allowed to dry. The slip contents were as follows:
142.2 g

Distilled water

2.00 g

Dolapix CE64 (Zschimmer and Schwarz, Lahnstein, Germany)

0.50 g

Ammonia solution (25% concentrate)

4.00 g

Duramax B-1052 (Rohm & Haas, Lauterborg, France)

0.96 g

Glydol N 109 NEU (Zschimmer and Schwarz, Lahnstein, Germany)

0.32 g

Contraspum KWE (Zschimmer and Schwarz, Lahnstein, Germany)

200 g

Alumina Powder (99.99% purity Al2O3, Taimicron TM-DAR, Taimei

Chemicals Co Ltd, Japan, powder size 1 m)

These were added in the above order whilst stirring with a magnetic stirrer (Labec MS5),
resulting in 200 mL of slip with 25vol% solids loading. The slip was mixed using the
magnetic stirrer and an ultrasonic bath (Unisonics Ultrasonic Cleaner FXPSM). Alternating
between these, ie. 10 minutes on the magnetic stirrer, then 10 minutes in the ultrasonic
bath, several times resulted in satisfactory mixing. The slip was produced several days
before casting, to enable a degree of aging to occur. Immediately before casting, the slip
was mixed again with the magnetic stirrer and the ultrasonic bath.

4-3

For homogeneous specimens, a single piece of foam (8 x 40 x 60 mm3) was used. For
graded specimens, a series of smaller pieces of different compression ratios (10 x 35 mm2
cross-section and thickness 1 to 4 mm) was used for the gradient region, along with a piece
of uncompressed foam (10 x 35 x 40 mm3) for the alumina-rich region, as shown in Figure
4.3. Initially, attempts were made to glue the foam together to ensure phase connectivity
across the step-interfaces. The glue tended to cause warping of the foam, which lead to
delamination. Further attempts to adhere the foam layers by holding them together with a
nominal applied pressure then heating to 160C in a box furnace (HPFT 6kW, Ward, as
above) were not regarded as successful. It was later found that placing the foam pieces
together, without glue, and applying a slight pressure in the gradient direction when placing
them into the mold resulted in good continuity of alumina and porosity across the step
interfaces. Accordingly, this approach was used for all subsequent samples.

The foam piece/s was/were arranged on top of a plaster block and were surrounded with
perspex moulds, as shown in Figure 4.4. These were coated with Teflon tape (Ebony PTFE
Tape, J. Blackwood & Son, Sydney, Australia) to enhance watertightness. The perspex
moulds were kept in place with a rubber band. The plaster-of-Paris block was used to draw
water down thereby hastening the slip-casting process. A microporous cellulose-based
membrane (0.002 mm membrane, Millipore) was placed between the foam and the plaster
to reduce the likelihood of contamination of the slip by particles or soluble ions from the
plaster.
30 to 45 mm

1 to 4 mm
10 mm
40 mm

Figure 4.3 Schematic of foam pieces for graded specimen, showing approximate dimensions.

Slip casting was carried out in a vacuum chamber (Epovac, Struers, Denmark, and DVP
Vacuum Technology, Bologna). The slip was poured over the foam, then pressure was
lowered to around 100 mbar, causing air bubbles to leave the foam. When the pressure was
returned to atmospheric pressure, the slip was forced into the foam. The pressure was
cycled several times to minimise the amount of air bubbles trapped inside the foam.
4-4

Perspex moulds

PU foam

Epoxy

Teflon
tape
Silicon mould
Plaster block

Alumina slip
Sintered
alumina pieces

Final FGM specimen

Figure 4.4: Samples and equipment at different stages of production: Alumina, plaster block,
perspex moulds, PU foam, teflon tape from casting step; alumina pieces after burn-out and sintering
steps; epoxy and silicon mould from epoxy infiltration step; and final FGM specimen after cutting
and machining steps.

The casting process took several (2-4) days. As water was absorbed, the slip volume
decreased, so that additional slip was required. After the samples had hardened sufficiently
to retain their shape, they were removed from the plaster/perspex set-ups and placed on top
of foam pieces in drying containers. Drying, which was conducted artificially slowly to
reduce the likelihood of cracking, took up to 10 days and was controlled by keeping
specimens inside sealed containers for the first few days, then gradually opening up the
boxes over a period of 3-5 days. After all discernible moisture had evaporated, final drying
was conducted in an environmental chamber at 65C for 24 hours.

4.1.3 Foam burn-out & sintering

The foam was then pyrolysed at 800C, leaving a ceramic green-body (approximate
dimensions: 50 x 30 x 6mm3) with a network of interconnecting pores. This was conducted
in a box-furnace (HPFT 6kW, Ward, Melbourne with HE388Z 240V electronic controller,
Harco Electronics, Melbourne). The following temperature regime was used: 25C
4-5

235C at 30C/hour, 1 hours hold; 400C at 35C/hour, 1 hour hold; 800C at

40C/hour, 1 hour hold; natural cool (around 15 hours). This is shown in Figure 4.5.

Foam & Binder Burnout

Sintering

1200
Natural Cool

Temperature [ C]

1500

900

Ash Burnout

600
Binder Burnout
300
Foam Burnout
0
0

10

20

30

40

Time [hours]
Figure 4.5: Temperature-time profiles for foam and binder burn-out step and sintering step.

The ceramic was sintered in a high-temperature computer-controlled furnace (Ceramic

Engineering, Melbourne) at 1500C for one hour, with heat-up at 100C/hour and natural
cool (12 hours). Sintering shrinkage of ~ 10% occurred in each dimension, resulting in
densified alumina pieces several of which are shown in Figure 4.4.

4.1.4 Polymer infiltration

Epoxy resin (Epofix, Struers, Denmark) was infiltrated into the ceramic under varying
pressure then cured at room temperature. A two-component epoxy, comprised of 15 parts
resin (Bisphenol-A-epichlorhydrine, molecular weight > 700) and 2 parts hardener
(Triethylaminetetramine), was used and mixed according to the manufacturers directions.
Homogeneous ceramic preforms were placed in folded aluminium foil moulds, whilst
graded preforms were positioned in pairs a pre-determined distance apart in a silicone
mould. The epoxy resin was stirred vigorously then poured into the mould. The mould was
then placed in a vacuum chamber (DS40 Single-stage vacuum pump, Wombat Pumps,
4-6

Smithfield) and subjected to several pressure cycles from 100 to 1000 mbar to allow the
resin to fully infiltrate into the pores, and for air bubbles to escape. The pressure was
reduced very close to the point at which the epoxy would start to boil (50 mbar). The
pressure was returned to atmospheric pressure for curing, which took approximately 24
hours at room temperature. Negligible shrinkage of the epoxy occurred during curing, so
that homogeneous and graded specimens did not contain residual stresses, as noted in
Section 3.3 on page 3-4. After curing, excess epoxy was removed with coarse sandpaper or
a steel file.

4.1.5 Grinding
Specimens were then ground to size (50 x 30 x 5 mm3) with a 600 grit diamond grinding
wheel (KGS-250AH, Kent Industrial, Kent, UK) with semi-automated xyz-stage fitted with
a magnetic vice. Specimens were held in place in the grinding machine by fixing them,
with hot wax or double-sided tape, to steel platens. It was considered that the elevated
temperatures required to melt the hot wax (around 150C) could degrade the epoxy, and it
was observed that the double-sided tape tended to lose its adhesive capacity quickly.
Subsequently, specimens were placed directly on the grinding machine plate and wedged
from each side with steel pieces. Once the magnetic chuck was engaged, these pieces held
the specimen firmly in place. This method was found to be effective and more efficient than
the use of tape or wax. Grinding rates varied between the different materials, from 0.4
mm/pass for epoxy to 0.03 mm/pass for porous alumina to 0.01 mm/pass for aluminaepoxy composites.

The sides of the graded samples, which were usually fairly uneven after infiltration with
polymer, were evened out by grinding. Samples were held between two square crosssectioned steel pipes that were bolted together. Low grinding rates, around 0.01 mm/pass,
were employed, as the samples were fragile in this orientation.

4.1.6 Polishing
Specimens were polished with an automatic polisher (LaboPol-5 with LaboForce-3,
Struers, Denmark) and diamond paste (Microid Diamond Component, Leco, Michigan/DP
Paste, Struers, Denmark) down to a diamond particle size of 1 m in the final step.
4-7

Polishing pads (MD Piano (120 m), Allegro (15 m), Pan (6 m, 3 m), Dur (1 m),
Struers, Denmark) and lubricant/extender (Microid Diamond Compound Extender, Leco,
Michigan) were used. Due to their relatively large size, polishing of graded specimens
could only be achieved by taping two discs an appropriate distance apart onto the graded
specimen. Polishing was conducted at 300 rpm with applied force of 20N for homogeneous
samples and 40N for graded samples.

4.1.7 Cutting
Homogeneous plate specimens were later cut into small beams (50 x 5 x 4 mm3 approx.)
with a low speed diamond saw (Model 650, South Bay Technology, California) for further
testing. This was a gravity-feed configuration with cutting rates varying significantly
between different compositions. This was also used for sectioning of the larger graded
specimens.

4.1.8 Other specimens

Several samples of pure alumina, pure epoxy and alumina-powder/epoxy-matrix
composites were also produced, to enable comparison with the interpenetrating-network
composites. Alumina specimens were produced from a suspension of alumina particles,
which was slip-cast as described above, dried and sintered at 1500C for one hour. Epoxy
specimens were produced from resin cured at room temperature. For the powder
composites, alumina powder was mixed with epoxy resin by stirring before curing. Due to
the viscosity of the mixture, air bubbles introduced during stirring were difficult to remove,
which lead to some porosity in these samples. Grinding and cutting were conducted as
described above.

Sample production was hindered by a number of processing difficulties, which were

eventually addressed. These are summarised in Table 4.1. Once the process was
established, samples, like that in Figure 4.6, were produced with an >80% success rate.

4-8

Table 4.1: Summary of processing difficulties encountered during sample preparation.

Problem

Processing Step

Solution

Incomplete slip infiltration

Casting

Mixed slip several days before.

Separation of foam layers

Casting/drying

No glue, slight pressure on foam pieces

Cracking

Drying/burn-out

Slower drying

Bubbles/Incomplete
infiltration

Epoxy infiltration

Use of a stronger vacuum pump, more

pressure cycling.

Damage to specimens

Grinding

More careful & slower grinding

Failure of interface b/w

epoxy & graded region.

Grinding/Testing

Roughened interface: cut small grooves

in alumina pieces before epoxy
infiltration

Length 130 mm

35 mm

100%
Epoxy

Graded Regions

95% Alumina
5% Epoxy

25 mm

95% Alumina
5% Epoxy

Crack Path

Notch 12 mm

Figure 4.6: Photograph of graded specimen after fatigue testing, with notch and crack path.

4-9

4.2

Microstructural analysis

Optical microscopy on the polished surfaces was used to quantify the fractions and
morphologies of the alumina and epoxy phases and the extent of porosity. Images were
obtained using a Nikon 200 microscope and digital camera, then processed using Adobe
Photoshop and Image Processing ToolKit (Version 3.0) [Russ, 1998]. Examples of
microstructures for a range of compositions, for both the interpenetrating-network and
powder composites, are shown in Figure 4.7. The continuity of each phase in the
interpenetrating microstructures cannot be observed in two-dimensional micrographs,
although 4.7(a) is comparable to the 3D interpenetrating-structure in Figure 2.6(a) [Wegner
& Gibson, 2000]. Differences between (a,b) and (c,d) are attributable to the buckling of
foam ligaments during compression which resulted in increasing microstructural
irregularity in the alumina phase (light regions) for higher foam compression ratios.
(b)

(a)

200 m

200 m

(d)

(c)

200 m

200 m

Figure 4.7: Optical microscopy images of homogeneous composites with compositions (a) 5%
epoxy, (b) 15% epoxy, (c) 30% epoxy and (d) 50% epoxy.

4-10

The scale and morphology of the composite microstructure reflected that of the original
polyurethane foam. This process resulted in two continuous interpenetrating phases; the
interconnecting porosity of the foam perform was mimicked by the alumina, which is a
robust structure after firing, while the complete penetration of the epoxy with no
discernable porosity in the composite confirmed the complete interpenetration of this phase.
The observed microstructures are very similar to those presented by others for similar IPN
structure composites [Neubrand et al., 2002, Chung et al., 2001]. Micrographs of powder
composites are shown in Figure 4.8, for comparison with the interpenetrating-network
structured composites. There is a marked difference between the micrographs of the
powder and IPN structured composites.
A

Figure 4.8 Powder composites: A) 75% epoxy. B) 90% epoxy. Alumina appears as the grey phase,
epoxy is the light phase and the darker spots are pores.

The samples tested varied in composition across the range 5%-50% epoxy for the
interpenetrating structured composites. For samples containing pores, measurement of
porosity volume fraction was incorporated into the compositional analysis, by a second
thresholding process. Although a small amount of porosity was observed in most samples,
this was generally < 2% and at most 5%. Error margins for composition measurements
were estimated via perturbation during the thresholding step and from variation in
compositions observed across single samples. The thresholding process is illustrated in
Figure 4.9, which shows a microstructure before and after thresholding. Figure 4.10 shows
micrographs on planes parallel to the direction of foam compaction. Comparison of these
with micrographs in Figure 4.7, which were on planes perpendicular to the direction of
foam compaction, indicates there is negligible anisotropy in these composites.
4-11

200 m

200 m

Figure 4.9 Microstructural image for 20% epoxy sample. A) Grey-scale image and B) Black and
white image after thresholding.

200 m

200 m

Figure 4.10 Representative microstructural images, on planes parallel to the direction of foam
compaction for A) 25% epoxy sample and B) 10% epoxy specimen.

After cutting and polishing, the densities were determined from mass and dimensional
measurements. Composite densities were compared with those predicted using
compositions obtained from image analysis and measured densities of constituent phases,
as shown in Figure 4.9(a). The observed disparity between the two density measurements,
particularly for more epoxy-rich samples, is most probably due to a higher volume fraction
of epoxy and pores near the surface where image analysis was undertaken. Several samples
were progressively ground down, polished and microanalysed at several depths, and
slightly higher alumina volume fractions were observed internally. For example, the density
for the sample containing ~40% epoxy was calculated, from surface microstructure, as 2.55
and, from internal microstructure, as 2.7. Similarly the calculated densities for the sample
containing ~12% epoxy were 3.5 at the surface and 3.65 internally.
4-12

Measured density [g/cm ]

(a)
3.5

2.5

2
2

2.5

3.5

Microstructural density [g/cm ]

100

Measured Alumina Volume Fraction [%]

(b)

90

80

70

60

50

40
40

50

60

70

80

90

100

Theoretical Alumina Volume Fraction [%]

Figure 4.11 Comparison of (a) directly measured density with that calculated from microstructures
and (b) theoretical and measured compositions.

4-13

It was noted that the compositions predicted from the initial PU foam compression ratios
differed significantly from those realised experimentally, as shown in Figure 4.11(b). The
assumption that final epoxy volume fraction is equal to initial PU foam volume fraction
appears to over-simplify the process of slip-casting and drying. It is possible that
infiltration of the PU foam with alumina slip is incomplete due to trapped air at vertices of
foam ligaments, or that shrinkage during drying increases porosity in the ceramic body.

Microstructures from graded specimens are shown in Figures 4.12 and 4.13. These show
that, within each step, the microstructures was very similar to those in the homogeneous
composites. Furthermore, at the interfaces between steps, phase continuity may be clearly
observed. Due to this continuity, the calculation of effective composite properties in the
interface region may result in a smoothing of the interfaces, as discussed in Section 2.2.6.
6%

16%

200 m

27%

200 m

16%

27%

200 m

40%

27%

40%

200 m

Figure 4.12 Optical microscopy images of interfaces within a graded composite structure, between
compositional steps. Approximate compositions of steps are given (as volume fraction of epoxy)
and positions of the interfaces are indicated.

4-14

4.3

Elastic property characterisation

Elastic properties of the homogeneous composites were measured with the impulse
excitation technique (IET). For measuring elastic moduli, the IET provides an alternative to
tensile or flexural loading, either of which may damage brittle material samples. This
technique involves lightly striking a rectangular-shaped specimen to stimulate vibration in
its natural modes [ASTM Std. C1259-98, Heritage et al., 1988, Atri et al., 1999], as shown
in Figure 4.14. Vibrational frequencies were measured with a microphone, amplifier and
signal analyser (Analyzer2000 V5+, www.brownbear.de). Samples were tested both as
plates (50 x 30 x 5 mm3 approx) and later as beams (50 x 5 x 4 mm3 approx), to investigate
the possibility of variation in properties across samples.
Microphone
Sample
To computer

Figure 4.14: Schematic of Impulse Excitation Technique apparatus, showing testing method.
Specimens were supported on a large piece of foam, enabling then to vibrate in their natural modes.

Values of Youngs modulus, E, and shear modulus, , were calculated from flexural and
torsional frequencies, fflex and ftors, respectively:
m L3
2
E = 0.9465
T1 f flex
3
bt
=4

Lm B
2

f tors
b t 1+ A

(3.1)
(3.2)

where L is the specimen length, b is the width, t is the thickness, m is the mass and T1, A
and B are geometry constants determined from L, b and t [ASTM Std. C1259-98]. Striking
in the corner, and along the main axis, preferentially excited the torsional, and flexural,
modes respectively. Poissons ratio was subsequently calculated as:
=

E
1
2

(3.3)

Error margins in modulus values were estimated from errors in dimensional measurements
and resonant frequency readings. Errors in Poissons ratio results were relatively high due
4-15

to build-up of errors from Youngs modulus and shear modulus calculations. Results from
IET tests are presented in Chapter 9.

4.4

Fatigue testing

Fatigue crack propagation was investigated in both homogeneous and graded specimens.
These were tested in a four-point bending configuration, in accordance with the standard
[DIN51109], as discussed in Section 2.3.4. Homogeneous specimens, with approximate
dimensions of 50mm x 5mm x 3mm, were tested using a support span of 40 mm and a load
span of 20 mm, as illustrated in Figure 4.15(a). Graded specimens, with approximate
dimensions of 130mm x 30mm x 6mm, were tested using a support span of 120 mm and a
load span of 60 mm, as in Figure 4.15(b). Asymmetric graded specimens, with approximate
dimensions of 90mm x 30mm x 6mm, were tested using a support span of 80 mm and a
load span of 40 mm. This enabled both gradient sections of each graded specimen to be
used for fatigue testing.

(b)

(a)
P/2

Sin = 20mm

P/2

Sin = 60 mm

P/2

P/2

Sout = 40 mm
Sout = 120 mm

Figure 4.15 Four-point bending specimen configuration, showing approximate dimensions, used
for fatigue-testing of (a) homogeneous and (b) graded composite specimens.

Beam specimens with were cut for fatigue testing. These were polished on both sides with
diamond paste down to a diamond particle size of 1 um. Notches of approximately 1 mm
length were introduced with a diamond saw. The notch root was sharpened with a razor
blade and diamond paste (3 m), according to the method of Damani et al. [1996], using a
specially constructed notching machine. This was comprised of a handsaw (4320, Makita,
Yokohama), with the saw-blade replaced by a wooden popstick with a razor blade (Gillette)
glued to it. Notching speed was controlled with a variable-voltage power supply (Rotary
Regavolt 72A, Berco, Melbourne). The blade was moved back and forth at approximately 2
4-16

Hz for 30-120 minutes, until the notch was sharpened across its entire width. This enabled
notch root radii around 10-30 m to be obtained. Given the relatively coarse microstructure
of these composites as shown in Figures 4.5, 4.10 and 4.11, this was regarded as adequate
for crack initiation.
Light
source

Microscope

CCD camera
Specimen
fixture

MTS 810
Movable
x-y stage
To image
analysis
computer
Figure 4.16: Photograph of MTS set up for fatigue testing of homogeneous composites, showing
microscope and digital camera on movable stage, 4PB fixture and load cell.

Fatigue testing was conducted using a hydraulically driven 810 Material Testing System
(MTS Systems Corporation, Minnesota) in load control mode, with compressivecompressive loading at a frequency of 8Hz and a load ratio, R = Pmin/Pmax of 0.1. Crack
propagation was observed using an Olympus SZ11 microscope, with additional lens
(73AL1-5x WD36), and an Optronics Magnafire CCD camera on a movable x-y stage, as in
Figure 4.16. This allowed the real-time monitoring of crack-extension on a computer
screen, using the Magnafire SP image software (Version 1). Red penetrant dye (Ardrox
907PB, Surechem Industries Pty Ltd, NSW) was used to improve visibility of the crack
propagation. Crack lengths were measured on both sides in between sets of load cycles.
This was done by moving the camera via the x-y stage aligning points on the specimen with
a reference point on the computer monitor. As crack length can only be measured at the
4-17

surface with this technique, the possible errors due to uneven crack front through the
specimen were much higher than the error in the crack length measurements themselves.
The testing set-up is illustrated in Figures 4.16 and 4.17.

A 5 kN load cell was used for graded specimens, whilst a 1 kN cell was used for
homogeneous specimens. Initial pre-loading of the specimen was conducted in
displacement control until a set load was reached, then the cyclic loading was applied in
load control, with a maximum displacement limit. Load values were very accurate (< 0.5%
error, in accordance with the specifications of the load cells) except at very low loads
(<20N), when errors could be up to 5%, due to clipping of the sinusoidal loading cycles.

crosshead
load cell
sample

computer
piston

controller

Figure 4.17: Schematic of MTS set-up for fatigue testing of homogeneous and graded composites,
four-point-bending specimen fixture, load cell, controller and computer (from Pfeiffer [2003]).

The specimens conformed to the standard single-edge V-notched beam (SEVNB) geometry
described in the literature [Anderson, 1992] and in standards [DIN51109, ASTM Std
E1820-01]. Values for stress intensity factor, K, for homogeneous specimens were

4-18

calculated from specimen and crack dimensions using the standard expression for this
specimen geometry in four point bending [DIN51109]:
P( S out S in ) 1.5
(3.49 0.68 + 1.35 2 ) (1 )
K=
1.9887 1.1326

t h1.5
(1 + ) 2
(1 ) 3

(3.4)

where h is the specimen height, = a/h is the relative crack length, t is the specimen
thickness, Sout and Sin are the outer and inner load spans and P is the applied load. This
expression is not applicable to graded specimens, as the elastic property gradient modifies
the concentration of stresses at the crack tip. As there is no standard expression, or even
closed form solution, for a crack of arbitrary shape in a graded specimen, with arbitrary
variation of elastic properties, finite element modelling was required.

The initial load for crack initiation was estimated from the known toughness values for the
constituent phases. Initial estimates were extremely conservative. After a number of
specimens had been precracked successfully, predictions for initiation loads improved
markedly. Cyclic loading was applied, with Kmax equal to the estimated initiation
toughness, for 3000 cycles. Estimated load was increased by approximately 2% in
successive steps until initiation occurred. The precrack was typically extended to several
hundred microns to ensure that initiation had occurred uniformly across the width of the
specimen.

After precracking was achieved, the specimen was subjected to a series of cyclic load steps,
generally comprised of 1,000 to 100,000 cycles at a particular load. Crack growth was
monitored throughout on one side, and was measured on both sides between load steps.
Applied loading was varied throughout fatigue tests in such a manner as to ensure that
crack growth rates varied non-systematically with crack extension. As crack extension
during load steps altered the SIF, crack lengths were measured frequently throughout
testing to ensure the accuracy of data obtained.

In order to examine the possibility of crack-extension effects in fatigue behaviour, as

discussed in Sub-section 2.4.5, the applied SIF amplitude was varied during fatigue testing
so that the correlation between crack growth rates and crack extension would be negligible.
4-19

This was achieved by varying the applied SIF amplitude, within the range resulting in crack
growth, several times throughout the crack propagation. It was intended that any influences
of crack-wake zone damage history would not be correlated with crack extension. While
these influences may occur, they would contribute to noise in the data rather than resulting
in a systematic trend.

The fatigue testing procedures were fairly standard, and appear to have been appropriate for
this investigation. A more sensitive and reliable method for detecting crack growth would
have been beneficial. Continuous monitoring of the crack via digital camera throughout the
fatigue tests was psychologically demanding and was not adequate to prevent fast fracture
in some cases. For less brittle composites, this would not be such a problem. Fatigue crack
propagation testing results are presented in Chapters 9 and 10 for homogeneous and graded
specimens respectively.

4.5

Moire interferometry

To provide a verification of the finite element predictions, a specimen was examined with
Phase-shifted Moir interferometry (PSMI) [Steffler 2000, Post et al., 1994]. This was
carried out at the Idaho National Engineering Laboratory in Idaho Falls, with the assistance
of Keith Rozenburg and Eric Steffler. PSMI has been used previously to characterize FGMs
by Winter et al. [1999] and Chapa-Cabrera and co-workers [2001, 2002], as mentioned in
Section 2.8.3 (p. 2-95).

Moir interferometry is based on the fringes that appear when two gratings are
superimposed. The relative density of Moir fringes is related to the density of each grating
and their relative displacement, so, if the grating densities are known, the relative
displacement may be determined from the fringe density. Optical Moir interferometry
involves the use of an optical interference pattern for one of the gratings, the fixed or
reference grating, with the other grating is attached to the specimen. The high grating
densities used allow high resolution in displacement measurements. The use of phaseshifting, essentially moving the reference grating by a very small fixed amount several
4-20

times, allows even better resolution. In the present study, qualitative results were obtained
only so resolution was not so important.

An asymmetric graded specimen (30x8x6mm) with gradient width 8.7mm was tested in a
20-10mm span four-point-bend configuration. A grating, with a pitch of 1200 lines/mm,
was aligned using a preliminary interferometer set-up and was then glued to the specimen.
The specimen was loaded to failure and the displacement fields around the crack were
observed. For more details on the Phase-shifted Moir Interferometry procedure, refer to
the work of Eric Steffler [2000] and Post et al. [1994]. Results from Moir interferometry
tests, described in Section 4.5, are presented in Section 10.6.

4.6

Computational procedures

Computational simulations, utilising finite element (FE) analysis, comprised a key

component of the investigation. FE simulations were used for three different purposes:

to systematically investigate the effects of material gradient, and related issues such as
crack-bridging and curvature, on crack-tip stresses and crack propagation;

to predict the loads required for crack-growth and the likely crack propagation paths in
graded specimens, as a guide for experiments and to provide a comparison with
experimental results; and

to analyse the experimental data obtained, to enable calculation of stress intensity

factors from applied loads and crack and specimen geometry, which would not be
possible by the standard methods used for homogeneous specimens.

The benefits of finite element analysis over other theoretical or computational approaches
are its flexibility, commercial availability and ease of use. The FE software used in this
investigation, ANSYS 6.1 [ANSYS, 2002], is particularly helpful as it allows models to be
input as macros. This allows geometry or material parameters to be changed much more
efficiently than if commands were input through the graphical user interface. Macros,
written in ANSYS Parametric Design Language, enable easy implemention of loops,
thereby allowing the iteration and incrementation routines required for automatic crack
growth simulations. Details of FE procedures and validation are given in Chapter 5.
4-21

In addition to the FE analyses, numerical modelling, data analysis and presentation were
conducted using Excel (1997 Version, Microsoft Corporation, Seattle), MathCad (Version
7, 1997, Mathsoft Inc. USA) and Kaleidagraph (Version 3.5, 2000, Synergy Software,
USA), for a range of issues including effective composite properties, SIF calculations, and
crack growth predictions and analysis. Details of these calculations are given in the relevant
sections of the thesis.

4-22

5 Computational Simulations: Methods & Validation

Fatigue crack propagation in graded materials is understood to be a complex process with
numerous influencing factors. In order to deconvolute these influences, by investigating
them individually, and to demonstrate that an understanding of the overall process has been
attained, a method for modelling crack growth in graded specimens was required. Finite
element analysis (FEA) was chosen as it is flexible and and models may be modified
without difficulty. FEA has been applied extensively to FGMs, to predict composite
properties [Reiter et al., 1997], residual stress distributions [Williamson et al., 1993, Bleek
et al., 1998, Chapa-Cabrera & Reimanis, 2002], crack-tip stresses [Eischen, 1987a, Marur
& Tippur, 2000, Anlas, 2001, Kim & Paulino, 2002] and fracture behaviour [Cai & Bao,
1998, Becker et al., 2001, Hoffman et al., 2001]. To account for material inhomogeneity,
certain modifications must be made to methods for analysis of homogeneous structures.

As mentioned in Section 4.6, FE simulations were used in the present study to investigate
influences on crack propagation, to provide predictions for comparison with experiments
and to analyse experimental data. FE simulation methodologies and their validation are
presented in this chapter. The core of the modelling was focussed on linear-elastic graded
specimens containing cracks under four-point bend loading. Accordingly, the development
and validation of the model is explained in this context. Other aspects, including effects of
crack-shape, bridging, plasticity and residual stress, were also investigated. To investigate
these, alternate specimen and loading configurations were used in some cases; details are
given in the pertinent sections in Chapters 7 and 8.

The implementation of a model for crack propagation in graded materials presents several
key challenges: projection of the material stiffness and toughness variation onto the model;
calculation of fracture parameters; prediction of propagation direction; and extension of
crack path. These have variously been addressed via a number of approaches, each of
which involves certain advantages and disadvantages. While the first of these is specific to
graded materials, the other issues have also been faced in simulating crack propagation in
homogeneous materials [Bittencourt, 1996, Bouchard, 2003].

5-1

5.1

Methods

Finite element simulation was conducted using ANSYS (Version 6.1), a commercial FE
software package. The general sequence for FE modelling is:
1. Definition of parameters
2. Definition of keypoints, lines and areas
3. Assignment of material properties
4. Meshing
5. Application of boundary conditions
6. Solution
7. Post-processing
Incremental or iterative approaches, involving looping in this sequence, are often utilised in
simulating more complicated situations. These approaches were employed in the current
investigation. This section gives an overview of modelling procedures, with limited specific
references to the ANSYS code used. A sample macro is given in Appendix B, with
accompanying comments, to provide more insight into the programming side of the
simulations.

Specimens were modelled in two dimensions under plane stress conditions. Generally,
compositionally symmetrical samples, consisting of two graded sections situated between
materials 1 and 2, as in Figure 5.1(a), were considered. The load was applied in a four-point
bending configuration. Equal loads were applied directly downward at two points on the
top of the specimen, whilst displacement constraints were applied at points on the base.
Crack propagation was assumed to initiate from a notch in the graded section, oriented
perpendicular to the gradient direction. Mixed-mode loading is expected to occur and
cracks should deflect towards the more compliant side [Erdogan, 1995, Gu & Asaro 1997a,
1997b]. The notch was described by its relative position across the gradient, = x/w, and
its length, a0. These parameters are shown in Figure 5.1(b).
Geometry was defined in terms of keypoints then areas. Particular lines, specifically
referred to later in the macro, were defined so that their line-numbers were set, whilst other
5-2

lines were created automatically during definition of areas. For cracks of arbitrary shape,
defined by a variable number of keypoints, the definition of areas presented a challenge.
This was achieved by defining triangular areas from pairs of adjacent points on the crack
and a reference point away from the crack path, then adding these progressively to build up
the total area. This section of the macro in Appendix B is accompanied by detailed
comments.
P/2

(a)

(b)

P/2
Sin = 60mm

Material 2

Material 1

Material 2
w

E() = E1 + (E2 - E1)n

h= 30mm
ao
Sout = 120mm

y
x

=x/w

Figure 5.1: a) Four-point bending configuration for crack-growth simulation, with b) close-up of
gradient region.

The material gradient was described by the magnitude of elastic property mismatch, RE =
E2/E1, the width, w, and the shape of the elastic property profile, defined by the exponent,
n, in a power-law expression. Continuous and stepped gradients respectively, as shown in
Figure 2.4(a) on page 2.7, were defined by:

and

E() = E 1 (1 + (R E 1) n )

(Continuous)

(5.1)

E () = E 1 (1 + ( R E 1)( Int[N S ] / N S ) n )

(Stepped)

(5.2)

where Ns is the number of steps across the gradient and Int[#] represents the operation of
rounding to the nearest integer. Stepped gradients with smoothing of effective properties at
step-interfaces were also considered. As discussed in Sub-section 2.2.6 and shown in
Figure 2.7 on page 2.19, and further mentioned in Section 4.2 in relation to the observed
microstructures in Figure 4.12. The smoothed, stepped gradients were defined as:

5-3

E ( ) =

1
E 1 (1 + (R E 1)(Int[N S ( + )] / N S ) n ) 1 ( ) 2 d

(5.3)

where 2 is the width of the smoothing region and is an integration variable. As linear
elastic behaviour was assumed, the stress distribution is independent of the stiffness, E1, so
this was kept at a constant value in all cases.

Meshing was conducted under free-meshing conditions using the AMESH command, with
isoparametric quadrilateral Plane 82 elements, which have 8 nodes: 4 vertex nodes and 4
midside nodes. The crack-tip singularities were modelled using degenerate triangular
quarter-point elements, which enabled crack-tip fields and fracture parameters to be
obtained. An example of meshing is given in Figure 5.2(a). To improve solution accuracy,
the mesh was refined around the crack (Figure 5.2(b)), and the crack tip in particular
(Figure 5.2(c)), with crack-tip element size a0/1000 where a0 is initial crack length, as
shown in Figure 5.1. Approximately 3500 elements were used in total. Mesh density and
patterns were controlled to an extent using the SMRTSIZE and LESIZE commands. This
allowed the high mesh density required at the crack tip and lower mesh density in other
regions, maintaining computational efficiency.

5-4

(a)

Load Point

Graded Regions

Load Point

Notch

(b)

Work-Integral Path

(c)

J-Integral Path

Quarter-point elements
Crack
Displacement
Correlation Points

Figure 5.2: Representative meshing for finite element model, showing (a) entire specimen and
close-ups of meshing (b) around the crack-front and (c) at the crack tip.

5-5

5.2

Modelling Issues

This section describes the methods used in the present study to address issues associated
with modelling of cracks in graded materials, which were introduced in Section 2.6.3.

5.2.1 Application of material gradient

As ANSYS does not offer variation in assigned material properties across elements
directly, the material property gradient was applied via a spatial variation in assigned nodal
temperatures. This method has been used previously by Rousseau and Tippur [2000] and, at
UNSW, by Kong Meng Leck [2002], as mentioned previously.

The Youngs modulus was defined as a linear function of temperature. The coefficient of
thermal expansion was generally set to zero to avoid the presence of thermal residual
stresses. The inclusion of residual stresses with this approach, via linear and higher order
temperature-dependence coefficients, is addressed in Section 8.3. Youngs modulus values,
calculated from compositional distribution, were assigned by defining temperaturedependent elastic properties:
E(T) = E0 + E1T

(5.4)

(T) = 0 + 1T,

(5.5)

then assigning spatially-dependent nodal temperature values, T(x). The code used to
achieve this was as follows:
________________________________________________________________________
*GET,NMAX_,NODE,0,NUM,MAX
*GET,NNUM_,NODE,,COUNT,
NDNEXT_ = 0

! Get the maximum node number

! Get # nodes selected
! Next Highest Node

*DO,intI,1,NNUM_,1
! loop over all selected nodes
*GET,NDNEXT_,NODE,NDNEXT_,NXTH
! get the next node
*IF,NX(NDNEXT_),le,lcer,THEN
prop=Ec
! ceramic properties
*ELSEIF,NX(NDNEXT_),lt,lcer+wgrad,THEN
postep=NINT(((wgrad+lcer-NX(NDNEXT_))/wgrad)*nsteps+0.5)
prop=Ep+(Ec-Ep)*((2*postep)/(2+2*nsteps))**nprof
*ELSE
prop=Ep
! polymer properties
*ENDIF
*SET,T_(NDNEXT_), prop-3.4
BF, NDNEXT_, TEMP, T_(NDNEXT_)
! set temperature
*ENDDO

________________________________________________________________________
5-6

This approach was judged to be less time-consuming than assigning properties to elements
individually, and to be equivalent in accuracy to the use of explicitly graded elements.

5.2.2 Calculation of fracture parameters

Several different approaches, introduced in Sections 2.3.3 and further discussed in 2.6.3,
including crack-tip displacement correlation, compliance change and J-integral, were used
to calculate stress intensity factors, KI and KII, and mechanical energy release rate, G.
Results were compared to confirm consistency. Stress intensity factors were determined
using the KCALC command in ANSYS, which invokes an automatic use of the displacement
correlation technique, as defined in Equation 2.32 on page 2-31. These are obtained from
crack-opening displacements, ui, of nodes near the crack tip:

u i E tip
8
K lim

( )
r 0 (1 +
() r
tip )g i

{,i} = {I,2},{II,1},{III,3}

(5.6)

The angular distribution function, gi()(), evaluated at , corresponding to points on the

crack face, has a value of 8(1-tip) for Modes I and II. This is a straightforward method for
calculating SIFs, KI, KII and KIII, from which mechanical energy release rate or SIF
magnitude may be calculated. For the two-dimensional simulations in the present study, the
former was calculated with Equation 2.28 (p. 2-28) with KIII = 0, whilst, the latter was
calculated as:

K = K 2I + K 2II

(5.7)

As the SIF values output by ANSYS are given as magnitudes, the sign of KII had to be
determined separately, also by displacement correlation. Displacement correlation points,
situated within a distance of approximately a0/1000 behind the crack tip, were selected
considering several issues:
1. Points were not directly adjacent to the crack tip due to possible inaccuracy associated
with the high stress gradients [Becker et al., 2001].
2. Points were situated within the K-dominant zone, which has an approximate radius of
a0/100, so that higher order terms would not influence results [Anlas et al., 2002].
3. Points were sufficiently close to the crack tip that the material property gradient would
not influence results, with less than 0.5% variation in properties occurring over the
5-7

distance a0/1000.
The chosen points were defined as keypoints a fixed distance behind the crack tip, as shown
in Figure 5.2(c), within 20 m of the crack tip, equivalent to ~0.2% of initial crack length.
The nodes associated with these would change with remeshing, and were thus determined
each time SIFs were calculated. It is acknowledged that irregularities near the crack tip,
related to meshing and the high stress gradient, represent a key source of error in
displacement correlation results. Crack-tip meshing (Figure 5.2(c)) was kept as consistent
as possible, though the effect of variation in crack-tip mesh configuration was investigated
and is reported in Section 5.3.

Load-point compliance, J-integral and modified J-integral approaches were also calculated
to verify the accuracy of the DCT values. The compliance method was also used to
determine mechanical energy release rate, Gcom, as in Eq. 2.33 on page 2-31. For the fourpoint bend loading configuration studied, this is a convenient technique to determine the
stored elastic energy within the specimen. It is similar to methods for calculating the Jintegral from loading conditions [Rice, 1968].

A routine has been developed in ANSYS for calculating the J-integral in the form of
Equation 2.29 on page 2-29 [ANSYS, 2002]. This was integrated around a path close to the
crack tip (30 m radius, equivalent to 0.3% of original crack length) to obtain Jhom, the
value assuming homogeneity of properties, which is close to the actual value, as in Gu et al.
[2001]. The routine was adapted to also calculate mechanical energy release via a modified
J-integral approach, involving both path and domain integrals, similar to Equation 2.58 on
page 2-74. In this case, the energy contained within a domain around the crack was
calculated according to the expression:
U = Wd ij u j d
O

(5.8)

where W is the spatial strain energy density, is the stress acting at the boundary of the
domain and u is the displacement at the boundary. The J-integral modified for graded
materials, Jgr, was calculated from the values of U before and after the extension of a test
kink of length, a:
5-8

J gr =

U post U pre

(5.9)

This approach, equivalent to Equation 2.58, was adopted for its simplicity, and to avoid
complications that may arise in directly calculating the J-integral for deflecting cracks. In
this case, the integral was calculated over a region of radius 10% of original crack length,
a0, while the extension length a was around 0.5% of original crack length.

The methods discussed for calculation of fracture parameters have been applied extensively
to straight cracks. It was noted, however, that their application to curved cracks has been
very limited. This raises the question of their applicability to curved cracks. Certainly
calculations at the crack tip (ie. DCT) are not influenced by crack curvature, however the Jintegral may be inappropriate for curved cracks, especially when calculated over larger
domains.

5.2.3 Simulation of crack propagation

Crack propagation was simulated incrementally. At each increment, fracture parameters
were calculated and crack propagation direction was determined, then the crack tip was
advanced and geometry and mesh redefined. Several of the deflection criteria discussed in
Section 2.3.4 on pages 2.32 to 2.35 were considered. The maximum tangential stress
(MTS) criterion allows prediction of deflection angle directly from fracture parameters.
Implementation of the maximum energy release rate (Gmax) and local symmetry (KII = 0)
criteria required a small kink be extended from the crack tip, as shown in Figure 5.3, at a
number of different kink angles, to enable the optimum angle to be determined. The
predicted angle was refined by a bisection method until an appropriate degree of accuracy
was attained. This process required remeshing and reassignment of nodal temperature
values for each angle value. As the extension of the test kink requires numerous iterations,
with geometry redefinition and remeshing, computation time is extended significantly.
Becker et al. [2001] showed that test-kink length must be small relative to the crack length
for accurate predictions, due to the influence of higher order stress-field terms. The local
symmetry criterion was used for most simulations due to its simplicity and accuracy. An
improved algorithm, to obtain predictions more efficiently, is described in Section 5.4.
5-9

The crack increment size, a, used was 500 m, equivalent to 5% of the original crack
length, whilst the test kink used in determining propagation direction was much smaller,
aK = 50 m. The value of relative kink length, aK/a0 = 5x10-3, whilst larger than the ideal
value of 1x10-4 suggested by Becker et al. [2001], was chosen due to the practical
limitations of remeshing. This issue is examined further in Section 5.3. Furthermore, it was
assumed that after initial deflection the crack would not experience any sharp changes in
direction. This justified the use of increment sizes significantly larger than the test kink
size, thereby saving computation time.
(b)

KI

(a)

KII
r

2v(x)

Crack

Kink ak kI

x1

Crack

2u(x)

kII
kII

KII

kI
KI

Figure 5.3. Schematic of crack tip (a) before and (b) after kinking has occurred.

When simulating crack propagation, the general scheme shown in flowchart form in Figure
5.4 was used. This may be compared with the framework shown in Figure 3.2 on page 3-6.
The distinction made between different geometries is illustrated in the adjacent diagrams.
Specimen geometry refers to the entire specimen outside the region in which the crack may
propagate. Crack geometry refers to the region in which the crack may propagate, and
includes the crack itself, except for the crack tip and a small region around it, which
comprise the crack-tip geometry. Dividing up the geometry like this reduced the area to be
deleted then redefined and remeshed when simulating crack propagation, thereby reducing
computation time. This re-meshing approach is similar to that used by other researchers to
model crack growth in non-graded materials [Bittencourt et al., 1996, Miranda et al., 2003,
Bouchard et al., 2003], as mentioned in Section 2.3.6 on page 2-40. Also mentioned there,
propagation simulations in elastic-plastic materials require that plastic strain history be
retained when the mesh is redefined. This is addressed in Section 8.5.3.
5-10

Define & mesh specimen geometry

Define & mesh crack geometry
Apply load & Solve
Obtain SIFs and G

Loop until crack

reaches edge of
specimen
Loop until
G = Gc

Specimen
geometry

Delete crack mesh and geometry

Redefine crack mesh & geometry

Crack-tip
geometry

Crack
geometry

Define crack-tip mesh & geometry

Apply load & solve
Obtain SIFs and G

Loop to
determine
propagation
direction

Calculate next test-kink angle

Delete crack-tip mesh and geometry
Delete crack mesh and geometry

Figure 5.4: Schematic diagram of the sequence of events in simulations of crack propagation (cf.
Fig. 3.2). Distinctions between specimen, crack and crack-tip geometry are illustrated on the right.

5-11

5.3

Validation of simulation methods

Values for mechanical energy release rate, G, calculated with displacement correlation,
load-point compliance and J-integral, and predicted deflection angle, k, for an initially
straight crack ( = 0.5, E2/E1 = 100, a0/h = 0.27, w = 15 mm, Nsteps = 6, aK = 50 m), and
for the same crack after it has deflected and propagated for several increments, are
presented in Table 5.1. The mechanical energy release rate values are consistent, with less
than 5% variation between the different approaches. The close agreement of values
obtained from displacement correlation, regular and modified J-integral and load-point
compliance is particularly encouraging as these values were calculated via fundamentally
different approaches. This was further validated by comparison with the results of Konda
and Erdogan [1994], shown in Appendix C.
Table 5.1: Comparison of mechanical energy release rate values and predicted deflection angles
calculated from different approaches, for straight and curved cracks.

Code
MTS
KII
Gdct
Gcom
Jhom
-

Fracture Parameter
Calculation Method
DCT
DCT
DCT
Compliance
J-integral Jhom
Modif. J-integral Jgr

Deflection
Criterion
MTS
(KII)min
G max
G max
J max
J max

Straight Crack
G [J/m2] k deg]
44.7
32.0
45.1
34.6
46.8
41.3
44.5
35.3
44.8
34.7
47.4
43

Curved Crack
G [J/m2] k [deg]
36.8
47.4
36.9
51.1
36.9
51.1
35.7
48.9
36.7
55.6
37.2
47.6

The deflection criteria for the straight crack are illustrated in Figure 5.5(a), in which values
for KII, GDCT, GCom, and Jhom (defined in Table 5.1) are shown for a range of kink-angles.
The angular resolution of each of the criteria is apparently limited by fluctuations in the
pertinent parameter, which are attributed to the limitations of the finite element mesh. KII
appears to display the most significant and systematic variation with test-kink angle, and
has only minor fluctuations near the optimum angle. It was thus concluded that use of the
(KII)min criterion would provide the most computationally robust predictions for
propagation direction.

5-12

48

1/2

4 10

46

com
44

3 10

II

(a)

Mode II SIF, K [Pa m ]

GDCT

42

hom

2 10

40

1 10

38

II

20

25

30

35

40

36

45

50

Mechanical Energy Release Rate, G [J/m ]

5 10

Kink Angle, [deg]

K

y [mm]

(b)

MTS
K

II

com
DCT

hom

4
3
2

Edge of graded region

1
0
-1

x [mm]

Figure 5.5: Comparison of deflection criteria. (a) Variation of deflection parameters with kink
angle and (b) subsequent crack propagation path predictions, from different deflection criteria, for
an initially straight crack in a stepped gradient ( = 0.5, E2/E1 = 100, a0/h = 0.27, w = 15 mm, Nsteps
= 6, aK = 50 m). MTS, (KII)min and MERR (with G calculated from compliance, Gcom, from
displacement correlation, GDCT, or from J-integral, Jhom) criteria are compared. (KII)min criterion
predictions appear to be precise and consistent with those of other criteria.

Crack propagation paths predicted using the different criteria are shown in Figure 5.5(b).
The differences in angle predictions, for a given increment (Table 5.1), appear to average
5-13

out so that crack path predictions obtained via the different approaches are fairly similar, as
noted by Bergkvist and Guex [1979] for cracks in homogeneous materials. An apparent
exception to this is the MTS criterion which diverges noticeably from the others as the
crack approaches the interface with material 1 (x = 0.08). This is attributed to the presence
of T-stresses which influence the crack path, and which are taken into account when a testkink is used though not when the MTS criterion is applied. It was concluded that the
predictions from displacement correlation and the KII = 0 criterion were sufficiently
accurate and consistent with those from other methods to be used throughout the
investigation.

Due to the limitations of the finite element model in accurately simulating the high stress
gradients at the crack tip, the mesh pattern used in this region can potentially influence the
results. To quantify this possible inaccuracy, stress intensity values, predicted for the same
crack geometry using differing mesh configurations, were compared. These configurations
were obtained by varying meshing control parameters, such as element divisions along
lines near the crack tip. The results in Table 5.2 indicate that, with approximate variations
of 2% in G and 1% in , the mesh used has a minor effect on the results, which is within
the acceptable range of error. These are all consistent with the results for the regular mesh,
as given in the second row of Table 5.1.
Table 5.2: Comparison of mechanical energy release rate values, calculated via displacement
correlation, and predicted deflection angles, from the (KII)min criterion, for a straight crack with
several different mesh pattern variations. Mesh patterns are described in terms of the element size at
the crack tip, r. 1 m is equivalent to approximately 0.1% of crack length.

Mesh
1
2
3
4

Description
Coarser (r = 8 m)
Finer (r = 1 m)
Asymmetric (rleft = 5 m, rright = 2 m)
Finer in front (rfront = 2 m, rrear = 5 m)

G [J/m2]
45.0
46.7
44.6
46.4

k [deg]
34.6
35.4
34.9
35.1

5-14

Regular mesh
Coarser mesh
Finer mesh

y [mm]

Asymmetric
Finer-in-front

5
4
3
2
1

Effect of Alternate mesh pattern

0

x [mm]

Figure 5.6: Effects of crack-tip meshing pattern on predicted crack propagation path. It is apparent
this has a negligible influence on predictions within the range of variation considered.

Figure 5.7(a) shows that the size of the test kink influences the KII value calculated for a
particular kink angle, thereby affecting the predicted optimum kink angle. For an initially
straight crack, the test kink size has a systematic influence on the prediction of deflection
angle, with larger test kink sizes resulting in larger deflections (Figure 5.7(b)). This is in
agreement with the findings of Becker et al. [2000] for graded materials, and of Karihaloo
et al. [1980] for homogeneous materials. After propagation of several increments, the
predicted angle for subsequent growth shows some variation with test kink size but this
appears to be less systematic (Figure 5.7(b)). In fact, any differences in predicted deflection
angle are averaged out after several increments of propagation, as shown in Figure 5.8(a),
in agreement with the conclusions of Bergkvist and Guex [1979]. This indicates that the use
of larger test kink sizes does not, in this case, compromise the accuracy of crack path
predictions as significantly as previously suggested [Becker et al., 2000].

Furthermore, the effect of crack extension increment size on crack path also appears to be
negligible, certainly within the limited range of increment sizes investigated. These ranged
from 0.2 mm (2% of original crack length) to 1 mm (10% of original length), and as
5-15

shown in Figure 5.8(b), the crack paths displayed very little deviation and no systematic
divergence over the trajectories studied. Nevertheless, a crack increment size of 0.5mm was
used in all subsequent calculations.
5

3 10

a =10 um

(a)

1/2

Mode II SIF, K [Pa m ]

a =50 um
a =300 um
k

II

2 10

1 10

25

30

(b)

Predicted Deflection Angle, K [deg]

10

35

40

45

Deflection Angle, [deg]

Test-kink size, a k [m]

100

Straight crack
Curved crack
40

After
2 deflection &
propagation

36

1
Initial
straight
crack
32

0.001

0.01

Relative Test-kink size, a k/a0

Figure 5.7: Effect of test kink size, aK, on (KII)min angular predictions. a) Variation of KII with
kink angle for different test-kink sizes. b) Variation of predicted deflection angle with test-kink size,
for the (KII)min criterion. Calculated values are for a straight crack ( = 0.5, a0/h = 0.33) and a curved
crack ( ~ 0.02, a/h ~ 0.6) in a stepped gradient (Nsteps = 5, w=15 mm, E2/E1 = 100, n = 1).

5-16

8
7

(a)

6
5
4

a k

a k/a0

10 m
30 m
50 m
100 m
300 m

0.001
0.003
0.005
0.01
0.03

3
2

Effect of Test-kink Size

1
0

x [mm]

a
0.2 mm
0.35 mm
0.5 mm
0.7 mm
1.0 mm

(b)

6
5

a/a0
0.02
0.035
0.05
0.07
0.1

4
3
2
1

Effect of Increment Size

0

x [mm]

Figure 5.8: Effects of (a) test-kink size, aK, and (b) crack-extension increment size, a, on crack
propagation paths predicted with the (KII)min criterion, for stepped gradient (n = 1, = 0.5, E2/E1 =
100, w = 15 mm, Nsteps = 6). Values for KII were calculated from DCT, with default values: a = 0.5
mm in (a), aK = 50 m in (b). It is apparent that the choices of test-kink and increment sizes do not
influence predictions noticeably.

5-17

5.4

Implementation of the local symmetry criterion

An improved method for predicting deflection angle was developed during this
investigation. Investigating the local symmetry criterion for use in simulations of cracks in
graded materials, it was observed that the variation in Mode II SIF, kII, with test-kink
direction is approximately linear over a wide range of angles, as shown in Figure 5.9. This
linearity was utilised in an improved algorithm for finding the optimum deflection angle.
This section discusses the linearity of the kII variation and its implementation for
application to crack path predictions.
0.5

0.25

FE results

II

Mode II SIF, K [MPa m 1/2]

Linear regression

-0.25

-0.5
10

20

30

40

50

60

Test-kink angle, [deg]

Figure 5.9: Variation of Mode II kink-tip stress intensity factor, kII, with kink angle, , for a crack
in a graded specimen under four-point bend loading, as reported by Tilbrook et al. [2004].

5.4.1 Stress intensity factors for kinked cracks

Prior to kinking, as in Figure 5.3(a) on page 5.10, the stress field around the main crack is
dictated by stress intensity factors, KI and KII. For an infinitesimally small kink extending
at an angle, , from the tip of the existing crack, as in Figure 5.3(b), it may be assumed the
stress field acting on the kink is equivalent to that around the main crack in the absence of
the kink [Anderson, 1995]. Hence, SIFs at the kink-tip are:
k I () = C11 K I + C12 K II + A1 T rc

k II () = C 21 K I + C 22 K II + A 2 T rc (5.8a,b)
5-18

where Cij and Ai are angular functions given by Palaniswamy and Knauss [1978] and rc,
related to process zone size [Smith et al., 2001], is equal to the kink-length. The full
expression for kII() is:
rc
1
1

3
k II () = [sin + sin ]K I + [cos + 3 cos ]K II + sin( 2)
T
4
4
2
2
2
2
2

(5.9)

This is a first-order approximation that is only rigorously applicable to infinitesimally small

kinks, as emphasised by Karihaloo and co-workers [1979,1980]. However, it is remarkably
accurate [Howard, 1978]. Accordingly, it is deemed sufficient for the initial discussion
here. The effect of inaccuracies in these expressions will be addressed subsequently.
Kitagawa et al. [1975] and Lo [1978] showed that the effect of kink length, other than that
related to T-stress, is very minor for kinks less than 1% of the original crack length.

The trigonometric functions in Equation 5.9 may be expressed as Taylor series expansions,
around = K, the as-yet unknown optimum deflection angle given by the local symmetry
criterion. For example, the expansion for sin(/2) is:
1
1

1

sin = sin K + cos K ( K ) sin K ( K ) 2 cos K ( K ) 3 + ...

8 2
48
2
2
2 2
2

(5.10)
This approach may be extended to address the issue of T-stresses directly. The correct
deflection angle, = K, gives kII() = 0, ie. the solution of:
rc
1
1

3
k II () = [sin + sin ]K I + [cos + 3 cos ]K II + sin( 2)
T = 0 (5.11)
4
4
2
2
2
2
2
Solving this algebraically to find an expression for K directly may be possible but the
expression obtained will not be simple. Rather, the fact that this expression is zero will be
employed subsequently. A very similar analysis is possible based on expansion about M,
the deflection angle predicted with the MTS criterion, which is known, however this does
not intrinsically take T-stresses into account. This analysis was presented in a conference
paper [Tilbrook M & Hoffman M. Implementation of the local symmetry criterion for
crack-growth simulations. Proc. Int Conf Struct. Integr. Fract., Brisbane, 2004].

5-19

Using such series expansions, kII() may be expressed as a series of terms, kII(n)(), where n
is the order in (-K):
k II () = k (II0 ) ( K ) + k (II1) ( K ) + k (II2 ) ( K ) + k (II3) ( K ) + k (II4 ) ( K ) + ... (5.12)
The initial, zeroth-order, term is:

k (II0 ) () =

KI
4

K
3
sin
+ sin K
2
2

K II
+
4

K
3
cos
+ 3 cos K
2
2

rc

T
+ sin( 2 K )
2

(5.13)

This is equal to zero, due to the kII() = 0 criterion. The first-order term is:
K
k (II1) () = I
8

K
3
cos
+ 3 cos K
2
2

K II

8

K
3
sin
+ 9 sin K
2
2

rc

T ( K )
+ 2 cos( 2 K )
2

(5.14)

This is non-zero and is linear in ( - K). The second-order term is:

KI
k (II2 ) () =
32

K
3
sin
+ 9 sin K
2
2

K II

32

K
3
cos
+ 9 cos K
2
2

rc
4
T ( K ) 2
sin( 2 K )
2
3

(5.15)

As the derivatives of sine and cosine are cyclic, a dominant part of this expression is
proportional to kII(K), and hence approaches zero as approaches M. Similarly the fourthorder term, and all subsequent even terms, will undergo partial cancellation, reducing the
magnitude of the symmetric component of the kII(K) variation. For angles within 30 (~0.5
rad) of K, the increase in power of ( - K) leads to significant diminishing of the higher
order terms. The third term, for example, has a dominant part that is proportional to, but
much smaller than, the first term:
KI
k (II3) () =
196

K
3
cos
+ 27 cos K
2
2

3
K II K
sin
+ 81sin K
+
2
196 2

rc
2
T ( K ) 3
cos(2 K )
2
3

(5.16)

The remaining part will also be small, and will partially cancel with the remainder from the
second term. Consequently, the dominant linear behaviour may be explained by the partial
cancellation of the even terms, resulting from the condition that kII(0)() is close to zero.
Accordingly, the linear term (Equation 5.14) may be justifiably used as an approximation
for kII(). Cotterell and Rice noted the predominant linearity of kII() for a deviating crack

5-20

under Mode I loading only, around = 0. They did not extend their analysis to higher
mode-mixities. This is equivalent to the findings in the present study, for the case of KII =
0, so that K = 0.
It has been shown that crack-tip stress and displacement fields have the same form in
homogeneous and graded materials, under linear-elastic conditions [Eischen, 1987a] and in
the presence of small scale yielding [Jin and Noda, 1994]. As long as the kink is small
compared with the scale of the compositional (and material property) variation, this
analysis will be as valid for a kinking crack in a graded material, as for one in a
homogeneous material. It should be noted, however, that the presence of a material gradient
can often be accompanied by a certain degree of anisotropy, so that the toughness depends
on the direction of the crack relative to the gradient direction. In this case, the local
symmetry criterion would not be valid, as it could be more energetically favourable for the
crack to extend under mixed-mode loading as in the case of interface cracks discussed in
Section 2.5.2 [Hutchinson & Suo, 1992].

To summarise, the aim of this analysis was to establish the reason for the observed linearity
in the kII() relationship. With a relatively simple analysis, it was demonstrated that the
even-order terms are similar to the zeroth-order term, due to the cyclicity of the derivatives
of the trigonometric functions which comprise the expression for kII(). As the zeroth order
term, by the assumption of local symmetry, goes to zero at the optimum deflection angle,
these even-order terms are small around this angle, so the odd terms dominate, and kII() is
predominantly linear. This provides a reasonable explanation for the linearity of kII(),
despite being based on an approximate expression. If a more rigorous approach were to be
used, based on a more complete expression for kII(), it is likely that the same underlying
principles would apply: the zeroth-order term would still go to zero by definition; the evenorder terms would be similar to the zeroth-order term, and hence be small. This would
result in very similar findings. Accordingly, a more rigorous analysis does not appear to be
justified in the present study.

5-21

5.4.2 Finite element implementation

SIFs were calculated from nodal crack-opening displacement values near the crack tip, both
before (K) and after (k) kinking. In this notation, KI and KII are the far-field stress
intensity factors acting on the main crack before kinking occurs, whereas kI and kII are the
local stress intensity factors acting on the kink after kinking has occurred. These were
obtained from nodes within 20 m of the crack tip, equivalent to ~0.2% of initial crack
length. A small test kink, aK, was extended from the crack tip at varying angles, and kII
was calculated from nodal displacements. Kink length was aK = 40 m, giving a relative
kink length of aK/ao = 0.004, where ao was initial crack length. Nodal displacements were
obtained from the half of the kink closest to the crack tip. Parallel T-stress before kinking,
T, was determined from stresses directly ahead of the crack tip using the expression [RaviChandar & Yang, 1999]:
T = xx yy

(5.17)

This was evaluated at a number of points approaching the crack tip, to obtain an average
value. T-stresses were also calculated from nodal displacements of 3 crack-face nodes, A,B
and C, situated at distances of 0, L/4 and L from the crack tip respectively, where the value
of L was chosen as 20 m, ie. ~0.2% of initial crack length. The following expression was
used [Tan & Wang, 2003]:
T =

2E '
( u A 4u B + 2u C )
L

(5.18)

A small test kink, of length aK = 40 m, was extended from the crack tip, and kII was
calculated from nodal displacements along the faces of the kink, as in Equation 5.6. This
was conducted for a number of different values of kink angle, , in order to evaluate the
variation of kII with .
Initial upper and lower bounds for the optimum kink angle, up and lo, were determined
from a wide interval around M, the angle predicted by the MTS criterion. The deflection
angle predicted by the MTS criterion, M, was calculated directly from KI and KII using
Equation 2.37 on page 2.35. The initial upper and lower bounds for the optimum kink
angle, up and lo, were determined by taking a wide interval around M.
5-22

The original algorithm was based on successive 30% reductions of the angular range, until
a sufficiently small range was attained. The values of kII were calculated for each of up and
lo, then the angle values were modified according to:
if kII(up) < kII(lo),

lo lo + 0.3(up lo)

if kII(up) > kII(lo),

up up - 0.3(up lo)

if kII(up) = kII(lo),

up up - 0.15(up lo), lo lo + 0.15(up lo)

(5.19)

The use of a 30% rather than 50% reduction in angular range per iteration provides a
tolerance for the scatter in kII values which is likely to occur. The results shown in Figure
5.9 on page 5.16 were obtained using this approach.

The linearity of the angular kII variation was exploited in a simple algorithm for finding the
optimum kink angle more efficiently. The initial angular bounds, 1 = lo and 2 = up, were
estimated from M as in the previous method. The optimum angle was estimated directly,
from these angles and their corresponding kII values, by interpolation:
g = 1 + ( 2 1 )

k (II1)
k (II2 ) k (II1)

(5.20)

The estimate may be refined by calculating kII for angular bounds closer to the estimated
optimum, ie. = g with around 4, then repeating the interpolation. With this
approach, the optimum angle is estimated within 0.1 in 4 iterations, compared with 12 or
more iterations required for an equivalent estimate with the original approach.

To assess the accuracy of the approach, the error in estimated optimum kink angle was
estimated from the intrinsic errors associated with the assumption of linearity. The values
of kII for the initial angular bounds will be near the linear values, as in Figure 5.10, with
each differing by a small error, x:
k (II1) = m(1 k ) + 1

k (II2 ) = m( 2 k ) + 2

(5.21)

where m is the linear gradient from (1,kII(1)) to (2,kII(2)). The estimated kink angle, from
(5.20), is:
g =

1 + 2
k ( 2 ) + k (II1)
( 2 1 ) II( 2 )
2
k II k (II1)

(5.22)

5-23

Combining (5.21) and (5.22), the error in g, the difference between estimated and actual
optimum kink angle, is:
1 + 2 2 1 k (II2 ) + k (II1)
g = g k =
+
2m
2m k (II2 ) k (II1)

(5.23)

If the angular bounds were distributed fairly symmetrically about the optimum angle, then
kII(2) and kII(1) would be similar in magnitude though opposite in sign, as would be the errors
1 and 2., due to the dominant odd-terms. Accordingly, both terms in Equation 5.23 will
be small, especially so in the refinement step, when the angular bounds are closer to, and

Mode II SIF, kII, at kink-tip

more symmetrical about, the optimum angle.

Linear
Approximations
Including T-stress
Neglecting
T-stress

k
g

2
Kink angle,

Interpolation
between
kII(1) & kII(2)

Figure 5.10: Schematic showing variation of Mode II SIF, kII, with kink angle, . Discrepancies, 1
and 2, between actual kII values () and linear approximation values, for angles 1 and 2, resulting
in error, g, in predicted optimum kink angle.

5.4.3 Crack growth simulations

The application of the improved algorithm to crack-growth simulations is demonstrated
here for two cracked specimen configurations: an asymmetrically-notched homogeneous
beam under three-point bend loading, and an inhomogeneous beam, with longitudinal
compositional variation, containing a straight crack under four-point bend loading, as in
Figure 5.1. The geometry and meshing for each configuration were in concord with
descriptions given in Section 5.1.
5-24

Values of SIFs and T-stress for the initially unkinked cracks were calculated from nodal
stress and displacement values near the crack tip. For the homogeneous specimen, KI = 1.0
MPam; KII = 0.089 MPam; T = 11 MPa; whilst for the graded specimen, KI = 1.62
MPam; KII = 0.76 MPam; T = 16 MPa. These values were used in the expressions in
Equations 5.9 and 5.13 to 5.16 to calculate the variation of kII() and the expansion terms.
These are plotted in Figure 5.11 for the graded specimen, showing that the linear term is
0.3

1/2

Mode II SIF Expansion terms, k [MPa m ]

extremely dominant.

0.2

II

k
0.1

(2)
II
(3)
II

k
0

(0)

kII

-0.1

-0.2

(1)
II

II

-0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

Test-kink angle, [rad]

Figure 5.11: Terms from Taylor expansion of kII(), compared with full expression.

The results for each configuration are plotted in Figure 5.12, which compares the FEM
results with the predicted variation (Equation 5.9) and the linear approximation (Equation
5.14), with and without the inclusion of T-stresses. The high degree of linearity of the kII-
relationship is apparent. Good agreement may be observed for both specimen
configurations. Predictions are shown with and without T-stress included. These
correspond to the assumptions rc = 0 and rc = aK respectively. The predictions with Tstress included show much better agreement with the FE results, illustrating the notable
influence of T-stress on deflection behaviour. The very close agreement between the full
expression and the FE calculation results indicates that the use of the expression in
5-25

Equation 5.9 on page 5.17, although it is strictly an approximation, is valid.

(a) 4 105
FE Results
Exact Expression KII
Linear Approximation

1/2

Mode II SIF, k [MPa m ]

3 10

II

2 10

Neglecting T-stress

1 10

Including T-stress
0

-1 10

-2 10

0.2

0.4

0.6

0.8

Test-kink angle, [rad]

4 10

2 10

FE Results
Exact Expression
Linear Approximation

Neglecting T-stress

II

1/2

Mode II SIF, K [MPa m ]

(b) 6 105

-2 10

-4 10

-6 10

Including T-stress

0.4

0.5

0.6

0.7

0.8

0.9

1.1

Test-kink angle, [rad]

Figure 5.12. Variation of Mode II stress intensity factor, kII, with kink angle, , for initially straight
crack in (a) an asymmetrically notched homogeneous specimen under three-point bend loading, and
(b) in a notched graded specimen under four-point bend loading. Comparison of predictions, from
exact expression and linear approximation, with results from finite element model.

5-26

0.06

Discrepancies between:

1/2

Mode II SIF, k [MPa m ]

FE results & linear approx.;

0.04

FE results & full expr.;

Full expr. & linear approx.

II

0.02

Second-order k
-0.02

Third-order k
Fourth-order k

-0.04

Fifth-order k
-0.06

0.4

0.5

(2)
II

(3)

II
(4)
II

(5)
II

0.6

0.7

0.8

0.9

1.1

Test-kink angle, [rad]

Figure 5.13. Non-linear component of kII variation. Plot showing discrepancy between linear
approximation and exact expression, along with second and third order terms from series expansion,
for crack in graded specimen.

The non-linear component of the kII() variation for the graded specimen is illustrated in
Figure 5.13. The relative discrepancy between the linear approximation and the exact
expression is plotted against , and compared with the second and third order terms
(Equations 5.15 and 5.16), along with higher-order terms. It may be observed that the
relative discrepancy is small, and the higher order terms have a very minor influence, in the
case investigated. This indicates that the use of the linear approximation in estimating
optimum deflection angle is valid, though the use of refinement step is recommended.

Crack propagation paths were simulated for both specimen configurations, utilising the
improved algorithm. For comparison, simulations were also conducted using the original
algorithm with the (kII)min criterion and the Gmax criterion, and using the MTS criterion
which does not require a test-kink. Predicted crack paths are shown in Figure 5.14 for both
specimen configurations. Clearly there is excellent agreement between results from all
5-27

approaches, with the only exception being a divergence of the MTS predictions for both
specimens later in the crack propagation, due to T-stresses which influence crack path but
are not taken into account in MTS predictions.
14
MTS
(k )

II min

12

max

Impr. (k )

II min

10
Homogeneous
Specimen

y [mm]

Graded
Specimen

0
0

6x [mm] 8

Figure 5.14. Predicted propagation paths for cracks in homogeneous and graded specimens.
Predictions obtained using the improved algorithm are compared with those obtained using the
more rigorous algorithm, with the (kII)min and Gmax criteria.

5.4.4 Further remarks

The variation of Mode II stress intensity factor, kII, at the tip of a kinked crack with the
deflection angle of the kink is close to linear, to a very close approximation, over a wide
range of deflection angles, and is predominantly antisymmetric about kII = 0. This was
explained in terms of the expression for Mode II SIF for a small kink, which can be
expanded as a series. The linear term, and subsequent odd terms, dominated due to partial
cancellation of the even terms. This has been utilised in a novel algorithm for prediction of
optimum deflection angles, which increased the efficiency of crack path simulations

5-28

significantly without compromising accuracy. Linear interpolation had been used

previously to predict deflection angle, by Becker et al. [2000], though this was over a much
smaller angular range, around 5.

It may be of interest in the future to extend this type of analysis to crack-deflection under
small scale-yielding conditions. Using the expressions for crack-tip stress-fields [Rice &
Rosengren, 1968, Hutchinson, 1968], an expression for the kink-tip mode-mixity could
conceivably be determined. Although stress intensity factors are not always applicable in
the presence of yielding, the local symmetry criterion may be equally well applied to cracktip stresses or displacements, as in the crack-tip opening displacement of Sutton et al
[2000]. Pettit and co-workers [1999] extended the analysis of Cotterell and Rice [1980] for
a slightly curved crack to materials with a crack-tip process-zone. They used a cohesive
zone within a boundary layer and investigated the influence of T-stresses and process-zone
size on crack path. Their model suggested that the presence of a process zone tended to
increase the sensitivity of crack path perturbation to T-stress. Rubinstein used a similar
approach to demonstrate that deflection behaviour will be influenced by the size of the
process-zone and by the nature of the cohesive tractions [2003]. Suresh and Shih [1986],
based on the earlier work of Shih [1974], analysed the stress fields around the tip of kinked
and forked cracks in the presence of plastic yielding, assuming a Ramsberg-Osgood
relation. They showed that the plastic near-tip stress and strain fields are dictated
predominantly by the plastic mode-mixity, and the strain-hardening exponent. Assuming
the optimum deflection angle would be that which results in Mode I loading, the plastic
mode-mixity should be fairly close to the elastic mode-mixity according to their results.
Their analysis could conceivably be furthered to branched cracks under mixed-mode
loading.

Crack deflection under mixed-mode loading in elastic-plastic materials has also been
investigated, with details given in Chapter 8. A result from this is presented in Figure 5.15,
which shows that the kII() variation is also highly linear for the case of small-scale
yielding. This suggests that the local symmetry criterion, and the algorithm presented here,
may be applicable for simulations of crack propagation in materials when plastic yielding is
5-29

likely to occur, certainly for the case of small-scale yielding. When small-scale yielding
conditions prevail, the crack-tip stresses are still dominated by the stress intensity factors,
so that deflection criteria based on SIFs would still be applicable. This explains the linearity
observed in Figure 5.15, but has not been shown rigorously. As discussed earlier, this could
be an interesting area for further work.
1

0.5

FE results

II

1/2

Mode II SIF, K [MPa m ]

Linear regression

-0.5

-1
-45

-40

-35

-30

-25

-20

Test-kink angle, [deg]

-15

-10

Figure 5.15: Observed linearity in kII() relationship for simulation of a crack in a layered coppertungsten composite specimen. The crack was situated in a more tungsten-rich section so that,
although some yielding occurred at the crack tip, the use of stress intensity factors was still valid.
Negative deflection angles are due to residual stresses.

5-30

5.5

Experimental specimens: Simulations & data analysis

In addition to investigating various aspects of crack propagation, the finite element

platform was also used in conjunction with experiments. Firstly, it provided crack path and
load profile predictions for comparison with the experimental results, and secondly, it was
used to calculate the values of crack-tip stress intensity factor for cracks in graded
specimens.

For a priori simulations, propagation direction was predicted using the local symmetry
(KII = 0) criterion, and critical loads were predicted from estimated values of effective
toughness. These were obtained from homogeneous specimen results. Predictions are
compared with experimental results in Chapter 10.

For analysis of experimental data, the crack path coordinates and load values, measured as
in Section 4.4, were used, along with assumed spatial property variation, to calculate stress
intensity factors and fracture toughness. Measured load values were applied to specimens,
with known specimen dimensions and crack geometry equivalent to the measured crack
path. The resultant SIF value was taken as the fracture toughness at that point, in the
monotonic loading cases. In the cyclic loading cases, resultant SIF values were used to
calculate fatigue crack growth resistance values. SIF calculations relied on assumptions of
spatial property variation and inherited any inaccuracy in these assumptions. As discussed
in Section 2.2.6 on page 2-18 and Section 4.2 on page 4-14, the averaging of composite
properties over a finite representative volume element could result in an effective
smoothing of sharp changes in material properties at step interfaces. Accordingly, analyses
were conducted with smoothed steps in addition to analyses with sharp steps, and results
were compared.

Data analysis via FE simulations was conducted for the specimens tested in the present
study and also for alumina-epoxy and alumina-aluminium specimens under monotonic
loading, investigated by Lyndal Rutgers [2004] and Robert Moon [2004]. Examples of
these analyses are included in Chapter 10.
5-31

6 Simulation Results: Effect of Gradient

Elastic gradient is expected to be a key influence on crack propagation path and stress
concentration in graded alumina-epoxy composites. This is investigated in this chapter in
results from finite element simulations examining effects of elastic property gradient and
crack dimensions. As discussed in Chapter 2, the elastic property gradient is the primary
influence on cracks in graded specimens and this has been investigated previously for a
range of crack and gradient configurations. However, calculations for cracks normal to the
gradient in graded joints, stress intensity factor profiles for propagating cracks in graded
specimens and the comparison of stepped and continuous gradients have not been
addressed previously in the literature. These issues are addressed in this chapter, whilst
effects of crack shape, crack-bridging and plastic deformation are addressed in Chapters 7
and 8.
6.1

Straight Cracks

A range of straight cracks in various gradients was analysed to elucidate the key influences
on stress intensity factor and mode-mixity at the crack tip. The results are shown in Figures
6.1 to 6.8. By considering the normalised stress intensity factor magnitude, |K|/P, the
effects of material gradient and crack geometry may be examined separately from applied
load. In the following discussion, the effect of crack position and the differences between
stepped and continuous gradients are examined first. The effects of gradient steepness and
scale and crack length are then addressed.

6.1.1 Effect of Crack Position

Several effects were observed in the variation of stress intensity factor and mode mixity
with crack position across the graded region, as illustrated in Figure 6.1. Firstly, for stepped
gradients, a cyclic variation was observed as cracks situated near or at interfaces between
steps exhibited higher mode-mixity and stress intensity than those situated within steps.
This is due to the fact that the gradient effect, the increase in both stress intensity factor and
mode-mixity associated with the presence of a material stiffness gradient, is higher in
regions in which the stiffness varies most sharply. It is worth highlighting in advance that
most of the results for cracks in stepped gradients, in Figures 6.3 to 6.8, pertain to a crack
6-1

situated in the middle of a step. Consequently, the mode-mixity and stress intensity are
generally lower than those for a similar crack in a continuous gradient. An implication of
this is that the presence of these interfaces in a stepped gradient represents a structural
disadvantage due to their stress concentrating effect. Conversely, the higher mode-mixity in
these regions could cause crack deflection toward regions in which the stress concentration
is lower.
w

30

10000

Stepped
25

Continuous

8000

20

15

6000

10

-3/2

Normalised SIF , K/P [m

4000

K/P Stepped

Crack-Tip Mode-Mixity, * [deg]

35

K/P Continuous
0
0

0.2

0.4

0.6

0.8

Relative Crack Position,

Figure 6.1: Effect of relative crack position within gradient region, , on crack-tip mode-mixity,
*, and normalised stress intensity factor, K/P, for straight cracks (n = 1, E2/E1 = 100, a/h = 0.27, w
= 15 mm), in continuous () and stepped (Nsteps = 5) () gradients.

Secondly, there is a general decrease in both mode-mixity and stress intensity for cracks
closer to material 2 (ie. higher values of , as in Figure 5.1(b)). This is attributed to the
decrease in relative material stiffness gradient: although the absolute gradient,

dE

/dx, is

constant across the graded region, the relative gradient, (dE/dx)/E, decreases as x and E
increase. Accordingly, it appears that the gradient effect is dependent on the relative
stiffness gradient in the vicinity of the crack tip. Also, there is a significant drop-off in the
stress intensity factor for cracks situated very close to material 1, < 0.2, however this only
occurs for cracks in continuous gradients. This drop-off in stress intensity is not mirrored
by the mode-mixity, which continues to increase as the crack is moved toward the interface
6-2

with material 1. This suggests that, although stress concentration and mode-mixity are both
related to the elastic property gradient, they are related in different ways, so that neither can
be reliably inferred from the other. The effect of position for continuous gradients is
comparable with that observed by Marur and Tippur [2000] and Gu and Asaro [1997a,b] in
Figures 2.39 to 2.41 (pp. 2-89 to 2-91).

6.1.2 Stepped & continuous gradients

Figure 6.2 compares stepped gradients with cracks situated in the middle of steps with
those situated at interfaces between steps. Cracks in stepped gradients experienced a much
more pronounced gradient effect near to the interfaces between steps. The figure
demonstrates that, as the number of steps in a stepped gradient is increased, the stress
intensity and mode-mixity values increase for cracks situated within steps and decrease for
those situated at interfaces between steps. This indicates an increase in the number of steps
reduces the magnitude of fluctuations across the stepped gradient and fracture parameters
approach those for cracks within a continuous gradient.
6500

Step-Interface
6000

K/P Step-Interface
25

5500

K/P Continuous
5000
20
4500

Continuous

15

Mid-Step
0

4000

Normalised SIF , K/P [m-3/2]

Crack-Tip Mode-Mixity, * [deg]

30

K/P Mid-Step
8

10

12

3500
14

Number of gradient steps, N steps

Figure 6.2: Effect of number of steps on crack-tip mode-mixity, *, and normalised stress intensity
factor, K/P, for straight cracks (n = 1, = 0.5, E2/E1 = 100, w = 15 mm), in stepped gradients at
points situated in the middle of (), and at interfaces between (), steps.

6-3

It is understood that this is a result of the cracks in stepped gradients being situated in the
middle of steps, as noted previously. For most positions, cracks in continuous gradients
exhibit higher mode-mixity and stress intensity factor magnitude than equivalent cracks in
stepped gradients. This may be noted in Figures 6.3 to 6.6.

6.1.3 Effects of gradient steepness & shape

The effect of varying the stiffness mismatch across the material gradient is shown in Figure
6.3. An increase in stiffness mismatch, E2/E1, which leads to a steeper material gradient,
results in an increased stress intensity factor for a given applied load, and an increase in the
mode-mixity. A similar effect is observed in Figure 6.4, with a decrease in gradient width,
w, which also increases the gradient steepness. The effect is less consistent in the latter case
due to the interference between gradient geometry and finite specimen geometry,
particularly for larger values of w. This is consistent with the findings of Erdogan [1995]
and others [Marur & Tippur, 2000, Gu & Asaro, 1997] reviewed in Section 2.8 (p. 2-89).
6000
5500

Continuous

5000

15

Stepped

4500
4000

10

K/P Stepped
K/P Continuous

3500
3000

Normalised SIF , K/P [m-3/2]

Crack-Tip Mode-Mixity, * [deg]

20

2500
0

10

100

2000

Stiffness Mismatch, E 2/E 1

Figure 6.3: Effect of material stiffness mismatch, E2/E1, on crack-tip mode-mixity, *, and
normalised stress intensity factor, K/P, for straight cracks (n = 1, = 0.5, a/h = 0.27, w = 15 mm), in
continuous () and stepped (Nsteps = 5) () gradients.

6-4

The effect of gradient steepness is in concord with the general findings of Marur and
Tippur [2000] and Gu and Asaro [1997a,b] in Figures 2.39 to 2.41 (pp. 2-89 to 2-91). The
alumina-epoxy system has a Youngs modulus mismatch around 100, as compared with 5
for alumina-aluminium. Figure 6.3 shows this would result in much higher values of modemixity for alumina-epoxy, so it is more appropriate for experimental investigation of crack
deflection in FGMs with the present specimen configuration, as discussed in Section 3.3.
12000

30

10000

25
8000

Continuous

20

Stepped
6000

15

10

K/P Continuous

4000

Normalised SIF , K/P [m-3/2]

Crack-Tip Mode-Mixity, * [deg]

35

K/P Stepped
5
0

10

15

20

25

30

35

2000
40

Gradient Width, w [mm]

Figure 6.4: Effect of gradient region width, w, on crack-tip mode-mixity, *, and normalised stress
intensity factor, K/P, for straight cracks (n = 1, = 0.5, E2/E1 = 100, a/h = 0.27), in continuous ()
and stepped (Nsteps = 5) () gradients.

The effect of gradient steepness is further illustrated in Figure 6.5, which shows the
variation of mode-mixity and SIF magnitude with position for a 3mm wide gradient.
Comparing this with the equivalent variation for a 15 mm wide gradient, in Figure 6.1,
shows that the change in gradient steepness has a significant qualitative effect on the
variation of |K|/P with position. The variation of mode-mixity with position is fairly similar
for the two cases. The significant drop-off in |K|/P as material 1 is approached, in Figure
6.5, is similar to that observed in Figure 6.1 for the continuous gradient case only. This
6-5

suggests that there is an underlying effect in the case of the stepped gradient also in Figure
6.1, but that this is obscured by the effect of the steps. For the stepped gradient case in
Figure 6.5, the effect of step interfaces on fluctuations in the |K|/P profile are much more
significant towards the left side of the gradient, whereas the relative fluctuations in modemixity are fairly similar across the gradient. The large variability in |K|/P introduces a
degree of uncertainty that compromises structural reliability.

14000

K/P Stepped
40

12000

35

10000

Stepped
8000

30

6000

-3/2

Normalised SIF , K/P [m

25

Crack-Tip Mode-Mixity, * [deg]

45

K/P Continuous
20

0.2

0.4

Continuous
0.6

0.8

Relative Crack Position,

4000

Figure 6.5: Effect of relative crack position in a narrower gradient, within gradient region, , on
crack-tip mode-mixity, *, and normalised stress intensity factor, K/P, for straight cracks (n = 1,
E2/E1 = 100, a/h = 0.27, w = 3 mm), in continuous () and stepwise (Nsteps = 5) () gradients.

The shape of the material stiffness gradient profile was also observed to affect SIFs, as
shown in Figure 6.6, in agreement with theoretical predictions in the literature, discussed in
Section 2.8.2. More compliant gradients, being those richer in material 1, exhibited higher
mode-mixities. For cracks situated at = 0.5, the stress intensity factor was maximum for
gradients with a profile exponent around 1.5. This may be understood by considering the
limiting cases, n 0 and n , which are equivalent to the entire gradient region being
comprised of material 2 or material 1, respectively.
6-6

In each case, the crack is situated a distance of w/2 away from a bimaterial interface, so the
crack is at a maximum distance from the region of material property variation, and the
stress intensity factor is a minimum. The stress intensity factor is thus maximised when the
distribution of relative property gradient around the crack is approximately symmetrical.
For a crack situated in the middle, this is the case for a gradient profile exponent of n = 1.5.
For cracks situated at other positions, the value of n for which the stress intensity factor
would be maximised would be different. This is illustrated in Figure 6.7, which shows the
effects of gradient profile for cracks at several different positions. Figure 6.7(a) shows that,
for a crack located closer to the compliant side, the exponent resulting in maximum stress
concentration is around n = 0.5. On the other hand, for a crack located closer to the stiffer
side, Figure 6.7(b) shows that maximum stress concentration occurs for an exponent around
n = 7.
5500

K/P Continuous
K/P Stepped

5000

20
4500

4000

Stepped
15

Continuous

10
0

0.5

1.5

2.5

-3/2

3500

Normalised SIF , K/P [m ]

Crack-Tip Mode-Mixity, * [deg]

25

3000
3.5

Gradient Profile Exponent, n

Figure 6.6: Effect of material stiffness gradient profile (described by exponent, n) on crack-tip
mode-mixity, *, and normalised stress intensity factor, K/P, for straight cracks ( = 0.5, E2/E1 =
100, a/h = 0.27, w = 15 mm), in continuous () and stepped (Nsteps = 5) () gradients.

6-7

1 10

9000

Continuous

Stepped 8000

25

7000
6000

20

K/P Continuous

5000
4000

15

K/P Stepped

Normalised SIF , K/P [m-3/2]

(a)

Crack-Tip Mode-Mixity, * [deg]

30

3000
10
0

0.5

1.5

2.5

2000
3.5

Gradient Profile Exponent, n

9000

(b)

Continuous
8000
25

Stepped
7000

20

6000

5000

15

K/P Continuous
K/P Stepped

4000

-3/2

10

10

3000

5
0

Normalised SIF , K/P [m

Crack-Tip Mode-Mixity, * [deg]

30

2000
12

Gradient Profile Exponent, n

Figure 6.7: Effect of material stiffness gradient profile (described by exponent, n) on crack-tip
mode-mixity, *, and normalised stress intensity factor, K/P, for straight cracks (E2/E1 = 100, a/h =
0.27, w = 15 mm), in continuous () and stepwise (Nsteps = 5) () gradients, located (a) toward
material 1 ( = 0.17) and (b) towards material 2 ( = 0.83).

6-8

6.1.4 Effect of crack length

As the length of the crack increases, the stress at the crack tip under a given load increases,
as shown in Figure 6.8. This is also the case for a homogeneous material, and little
difference is observed between stepped and continuous gradients. It is interesting to note
that the mode mixity is a maximum for a crack that extends halfway across the specimen,
ie. a/h = 0.5. This could be explained as follows. The mode-mixity at the crack tip is
understood to result from the mismatch between the deformations that occurs on each side
of the crack. For a very short crack, for instance, the mutual constraint that acts between the
two sides is spread over a wide region. On the other hand, for a very long crack, which
extends most of the way across the specimen, the material on each side of the crack is
almost completely free to deform with only slight constraint from the opposing side. The
mode-mixity is dependent on the amount of shear constraint acting across the sample,
which is proportional to the remaining ligament length, as well as depending on the
allowable asymmetric deformation, which would be proportional to the crack length.
50000

25

20

Normalised SIF , K/P [m-3/2]

Crack-Tip Mode-Mixity, * [deg]

Continuous
Stepped

40000

30000

15

K/P Continuous

10

20000

K/P Stepped

10000

K/P Homogeneous
0
0

0
0.2

0.4

0.6

0.8

Relative Crack Length, a/h

Figure 6.8: Effect of crack length, a/h, normalised by specimen height on crack-tip mode-mixity,
*, and normalised stress intensity factor, K/P, for straight cracks (n = 1, = 0.5, E2/E1 = 100, w =
15 mm), in continuous () and stepwise (Nsteps = 5) () gradients.

6-9

6.1.5 Generalised gradient effect

Much of the data in Figures 6.1 to 6.7 suggests that, for a given crack length, the variation
of mode-mixity, , and of stress concentration, |K|/P, with each gradient parameter is very
similar. This implies that there may be a general gradient effect, as mentioned previously,
related to both mode-mixity and stress concentration. The data for continuous cracks in
Figure 6.5, and to a lesser extent in Figure 6.1, indicates that there is not a fixed
relationship between mode-mixity and stress concentration. For cracks situated close to
Material 1 in each of these Figures, an increase in mode-mixity is observed along with a
decrease in stress concentration. Figure 6.9 shows data from Figures 6.1, 6.2, 6.4 and 6.6,
plotted as normalised SIF, |K|/P, versus mode-mixity, . It is observed that there is a
general trend relating mode-mixity and stress concentration, but that there is significant
scatter around this trend. This indicates that there is not an underlying gradient effect that
dictates both mode-mixity and stress concentration, but that each of these is independently
related to the gradient and crack geometry.

Normalised SIF, |K|/P [m

-3/2

0.012

0.01

Shape, n, stepped
Shape, n, continuous
Position, , stepped
Position, , continuous
Number of steps, N
s

Width, w, stepped
Width, w, continuous
0.008

0.006

0.004

0.002
0.1

0.2

0.3

0.4

0.5

0.6

Mode-mixity, [deg]
Figure 6.9: Relation between mode-mixity, *, and normalised stress intensity factor, K/P, for
straight cracks (E2/E1 = 100, a/h = 0.27), in continuous () and stepwise () gradients (Default
parameter values: n = 1, = 0.5, w = 15 mm, Nsteps = 5). Data from Figures 6.1, 6.2, 6.4 and 6.6.

6-10

6.2

Propagating cracks

Crack propagation was simulated for a range of different initial cracks, utilising the
automatic remeshing algorithm and deflection criteria described in Section 5.2.3. Cracks in
continuous gradients and stepwise gradients, with 3, 5 and 11 steps, were considered.
Gradient profile and crack position were varied, with values for n of 0.5, 1 and 2, and
values for of 0.17, 0.5 and 0.83, considered. Initial crack length was maintained constant
such that a0/h = 0.35. Crack propagation paths, shown in Figures 6.10(a) to 6.13(a), exhibit
deflection towards the compliant side, with deflection angles of 35 to 55. Corresponding
normalised stress intensity factor profiles are shown in Figures 6.10(b) to 6.13(b). These
have been given as the SIF normalised by the SIF for a crack, of the same vertical length, in
a homogeneous specimen of the same dimensions. This allows the effects of elastic
gradient and crack shape to be examined independently of the dominant influence of crack
length. These results are presented and compared in further detail in Appendix D.

The effect of gradient continuity is shown in Figure 6.10 for the linear (n = 1) gradient case.
It is observed that the effect on crack path is very minor. The shape of the SIF profile is
generally similar for gradients with differing continuity, although the presence of steps
leads to fluctuations in the normalised SIF across the gradient. Furthermore, there is a
deviation between the SIF profiles for stepped and continuous gradients as the compliant
material side is approached. The SIF decreases for the continuous gradient whereas it
increases for the stepped gradient, similar to the results in Figures 6.1 and 6.5.

The effect of gradient profile is more significant. This is shown in Figures 6.11 and 6.12,
for cracks in stepped and continuous gradients respectively. Cracks positioned at = 0.5
and = 0.83 in gradients with n = 2 experienced higher initial deflection. These generally
deflected more throughout propagation, with the exception of those in continuous gradients
(Figure 6.12(a)), where the crack in the n=2 gradient straightened up and crossed over the
cracks in gradients with lower n values. Gradients with n = 0.5 generally exhibited the least
total deflection for cracks positioned at = 0.5 and = 0.83. For cracks positioned at =
0.17 however, those in gradients with n = 0.5 deflected most and those with n = 2 deflected
6-11

least. This is expected as the relative gradient, (dE/dx)/E, is highest on the left of the gradient
for n = 0.5 and highest on the right of the gradient for n = 2. Furthermore, gradient profile
has a significant effect on SIF profile, with n = 0.5 resulting in much higher stress intensity
factors than gradients with n = 2.

The starting position of the crack does not appear to influence the average propagation
angle significantly, although cracks initially situated to the right of the gradient (ie. =
0.83) will obviously have to propagate further before reaching the interface with Material 1.
There are similar observations for SIF profiles. After the crack has deflected and begun
propagating, the relative stress concentration values for cracks starting at different positions
in the same type of gradient become very similar. This suggests that the effects of crack
shape and initial crack position are secondary to the effects of the gradient. It is possible
that for alternative crack/gradient configurations these will have a greater influence.

Figure 6.10 shows that the effect of gradient continuity, or number of steps, is not as
significant. For n = 1, the crack paths are all fairly similar: the crack in the gradient with 3
steps and that in the continuous gradient are almost identical. This might be expected from
the results in Figures 6.1 and 6.2: although mode-mixity varies across a stepped gradient,
on average it is very similar to that for an equivalent continuous gradient. However, for n =
2 as in Figure 6.13, there is some significant divergence. This may be attributed to the fact
that the relative gradient is most widely distributed in this case. This means a crack tip
located in the middle of a step and one at an interface between steps will tend to experience
a more significant difference in mode-mixity values. This increases the probability that
cracks in gradients of differing continuity will bifurcate.

Comparing the SIF profiles in Figure 6.13(b) with those in Figure 6.10(b), the significant
influence of gradient profile on stress concentration is apparent. The n = 2 gradient profile
results in notably less stress concentration in general, and in particular less stress
concentration on the left-hand side of the gradient. This is important as the left-hand side
(low ) is likely to be more vulnerable, as cracks are deflected to that side and also due to
the toughness distribution. This issue is addressed further in the following section.
6-12

12

Propagation Paths: n = 1

(a)

y [mm]

10

3 steps
5 steps
8

11 steps
Continuous

10

12

x [mm]

(b)
6

5 steps
11 steps
3 steps
Continuous

gr

Normalised SIFs, K /K

hom

SIF Profiles: n = 1

10

12

x [mm]

Figure 6.10: (a) Crack propagation paths and (b) corresponding SIF profiles for initially straight
cracks in continuous and stepped gradients (3,5 and 11 steps) with gradient profile given by n = 1
(E2/E1 = 100, a0/h = 0.35, w = 15 mm). The effects of gradient continuity and initial crack position
( = 0.17, 0.5, 0.83) are illustrated.

6-13

12

Propagation Paths: 5 Steps

(a)

y [mm]

10

n=1
n = 0.5
n=2

10

12

x [mm]

(b)

SIF Profiles: 5 Steps

n=1
n = 0.5
n=2

gr

Normalised SIF, K /K

hom

10

12

x [mm]

Figure 6.11: (a) Crack propagation paths and (b) corresponding SIF profiles for initially straight
cracks in continuous and stepped gradients (3,5 and 11 steps) with gradient profile given by n = 1
(E2/E1 = 100, a0/h = 0.35, w = 15 mm). The effects of gradient continuity and initial crack position
( = 0.17, 0.5, 0.83) are illustrated.

6-14

12

Propagation Paths: Continuous

(a)

y [mm]

10

n=1
n = 0.5
n=2

10

12

x [mm]

(b)

n=1
n = 0.5
n=2

gr

Normalised SIF, K /K

hom

SIF Profiles: Continuous

10

12

x [mm]

Figure 6.12: (a) Crack propagation paths and (b) corresponding SIF profiles for initially straight
cracks in continuous and stepped gradients (3,5 and 11 steps) with gradient profile given by n = 1
(E2/E1 = 100, a0/h = 0.35, w = 15 mm). The effects of gradient continuity and initial crack position
( = 0.17, 0.5, 0.83) are illustrated.

6-15

12

(a)

Propagation Paths: n = 2

y [mm]

10

3 steps

5 steps
11 steps
Continuous

10

12

x [mm]

3.5

(b)

SIF Profiles: n = 2

2.5

gr

Normalised SIF, K /K

hom

2
1.5

5 steps
11 steps
3 steps
Continuous

1
0.5
0

10

12

x [mm]

Figure 6.13: (a) Crack propagation paths and (b) corresponding SIF profiles for initially straight
cracks in continuous and stepped gradients (3,5 and 11 steps) with gradient profile given by n = 2
(E2/E1 = 100, a0/h = 0.35, w = 15 mm). The effects of gradient continuity and initial crack position
( = 0.17, 0.5, 0.83) are illustrated.

6-16

Gradient profile and crack-tip position are the most significant influences on stress
concentration. Values for |K|/P for identical cracks in gradients with different gradient
profile can vary widely. For example, cracks in a continuous gradient, as in Figure 6.12(a),
with crack tip located around = 0.15, vary by a factor of more than three, from |K|/P = 1.3
for the n = 2 gradient, to |K|/P = 4.3 for the n = 0.5 gradient. Similar trends are observed
between stepped and continuous gradients, though cracks tend to experience a steady
increase then decrease in |K|/P with crack propagation for continuous gradients, whereas
|K|/P continues to increase towards the interface with material 1 for cracks in stepped
gradients. Otherwise, the presence of steps only has a perturbing effect on the SIF profile,
causing fluctuations about the trend observed for a continuous gradient. Gradients with n =
2 are clearly advantageous from a stress minimisation perspective.

6.3

Toughness influences

Whilst it is interesting and instructive to consider the variation in stress concentration

across the gradient, the final structural integrity will depend on both stress concentration
and toughness. Accordingly, toughness variation has been incorporated into the analysis of
graded materials, and used to calculate critical loads. A linear rule-of-mixtures for effective
toughness was assumed, with toughness determined from relative position, = x/a, as:
Kceff() = Kc(1) + (Kc(2) Kc(1))n
where Kc(1) and Kc(2) are toughness values for Materials 1 and 2 respectively, and n is the
exponent defining gradient profile. The effects of crack extension toughening are addressed
separately in Chapter 8.

The variation of critical load for crack propagation with crack position for a straight crack
is shown in Figure 6.14, for the case of alumina-epoxy composites. In this case, the
toughness varied from 4 MPam on the alumina side ( = 1) down to 1 MPam on the
epoxy side ( = 0). Critical loads vary significantly, by a factor of 6, across the gradient.
This is because, moving towards the left of the gradient, the stress concentration increases
whilst toughness decreases. Given that crack deflection is predicted toward that side, this
implies that this type of gradient is fairly vulnerable to failure. Furthermore, as the crack
6-17

increases in length, the SIF magnitude will increase. The stress-concentrating effect of the
steps is observed as local decreases in the critical load for the stepped gradient.

350

Stepped Gradient
Continuous Gradient

250

Critical Load, P [N]

300

200
150
Crack will tend to
deflect to left

100
50
0
0

0.2

0.4

0.6

0.8

Relative Crack Position,

Figure 6.14: Effect of position, , on critical load, assuming a linear toughness variation across the
gradient, for straight cracks (E2/E1 = 100, a/h = 0.27, w = 15 mm), in continuous () and stepwise
(Nsteps = 5) () gradients. The toughness is assumed to decrease linearly (or in steps) from 4
MPam, at = 1 to 1 MPam at = 0.

In a case where the more compliant phase is tougher, ie. Kc(1) = 12 MPam, the profile is
changed significantly, as shown in Figure 6.15. Moving left across the gradient, the critical
load increases slightly, as compared with the strong decrease observed in Figure 6.14. This
increase in critical load could lead to crack arrest.

6-18

Crack will tend to

deflect to left

Critical Load, P [N]

500

400

300

Stepped Gradient
Continuous Gradient
200
0

0.2

0.4

0.6

0.8

Relative Crack Position,

Figure 6.15: Effect of position, , on critical load, for straight cracks (E2/E1 = 100, a/h = 0.27, w =
15 mm), in continuous () and stepwise (Nsteps = 5) () gradients. The toughness is assumed to
decrease linearly (or in steps) from 4 MPam, at = 1 to 1 MPam at = 0.

The results in Figures 6.14 and 6.15 suggest that the presence of steps, rather than a
continuous gradient, could be beneficial. Although they tend to decrease the critical load
when the crack tip is near step-interfaces, the subsequent increase in critical load as the
crack moves past the interface is likely to cause crack arrest.

6-19

1200

n=1
n = 0.5
n=2

800

Critical Load, P [N]

1000

Crack moves toward

compliant side

600

400

200

0
0

10

12

Crack-tip position across gradient, x [mm]

Figure 6.16: Critical load profiles for initially straight cracks in stepped gradients (E2/E1 = 100, a0/h
= 0.35, w = 15 mm, Nsteps = 5). The effects of property gradient profile and initial crack position (
= 0.17, 0.5, 0.83) are illustrated. The toughness is assumed to decrease linearly (or in steps) from 4
MPam, at x = 15 to 1 MPam at x = 0.

This analysis is extended to propagating cracks in Figures 6.16 and 6.17. For the aluminaepoxy gradient in Figure 6.16, the critical load drops off dramatically toward the compliant,
epoxy-rich side. This is a combination of the higher stress concentrations toward the
compliant side as seen in Figure 6.10(b), the increase in SIF with crack length and the
decrease in toughness with increasing epoxy content. This indicates that once a crack
initiates, it is likely to fast-fracture. Figure 6.17 shows critical load profiles for the more
structurally robust, albeit hypothetical, material combination, as was considered in Figure
6.15. Here a general decrease in critical load is observed as the crack traverses the gradient,
however it is much less steep than in Figure 6.16. Furthermore, the effect of steps is such
that the crack could arrest at a number of points across the gradient. This could be
beneficial in terms of the reliability of the structure.
6-20

4000

Crack moves toward

compliant side

Critical Load, P C [N]

3000

n=1
n = 0.5
n=2

2000

1000

10

12

Crack-tip position across gradient, x [mm]

Figure 6.17: Critical load profiles for initially straight cracks in stepped gradients (E2/E1 = 100, a0/h
= 0.35, w = 15 mm, Nsteps = 5). The effects of property gradient profile and initial crack position (
= 0.17, 0.5, 0.83) are illustrated. The toughness is assumed to decrease linearly (or in steps) from 4
MPam, at x = 15 to 1 MPam at x = 0.

The toughness distribution is likely to influence the optimum gradient profile and width.
Considering the case where Material 1 is less tough, as the crack is likely to deflect into this
region, failure is inevitable. The optimal design strategy would involve reducing the overall
stress concentration by using a higher gradient profile exponent, ie. n = 2, and increasing
gradient width. It must be stressed that this is an unlikely situation, despite being the case
for alumina-epoxy FGMs in the present study. Normally, the less rigid material is chosen
for its higher toughness, as in Figures 6.15 and 6.17. In that situation, the design strategy
would involve balancing deflection into the tougher material with stress concentration to
find an optimal width. From Figure 6.17 it appears that a higher gradient profile exponent
would be beneficial for reducing critical load in the left-hand side of the gradient. The issue
of gradient optimisation is addressed in more depth in Section 11.3.
6-21

Considering the process of crack deflection in terms of the test-kink discussed in Chapter 5,
the variation in toughness experienced by the kink could have an influence on propagation
path. Figure 6.18 shows the variation of mechanical energy release rate, G, and fracture
toughness, Gc, with kink angle. It shows that the applied mechanical energy release rate
under a given load varies much more with deflection angle than the critical value does. This
indicates that the spatial variation in toughness is unlikely to have a major influence on
crack path in a continuously graded specimen. In the limit of an infinitely small test-kink,
the effect of variation in Gc will go to zero. If crack propagation direction is determined by
the maximum stress some distance away from the crack-tip. In this case, the use of a
deflection criterion based on stress distribution rather than on mechanical energy release
rate would be expedient, and higher-order terms in the stress field may need to be taken
into account. This accords with the findings of Becker et al. [2002]. Furthermore, FGMs
are likely to exhibit microstructural heterogeneity, so that crack path at small scales is

Normalised Mechanical Energy Release Rate, G()/G(0)

dominated by local microstructural features rather than macroscopic toughness gradient.

1.1

Applied G

1.05

Critical G - Case 2
1

Critical G - Case 1
0

10

20

30

40

50

Deflection Angle, [deg]

Figure 6.18: Variation of applied and critical mechanical energy release rate with kink angle.
Critical mechanical energy release rate, ie. fracture toughness Gc, was calculated for two cases: (1)
the case considered in Figures 6.14 and 6.16, where Material 1 is less tough than Material 2, and (2)
the case considered in Figures 6.15 and 6.17, where Material 1 is more tough than Material 2.

6-22

When G is more strongly dependent on crack propagation direction, it may have a more
significant influence on propagation path. This would be the case in anisotropic materials or
at bimaterial interfaces, as discussed in Section 2.5.2 on page 2-62. The possibility of
toughness anisotropy at step-interfaces influencing crack path in step-graded specimens is
addressed in Chapter 10.

This section has only considered the spatial distribution of intrinsic toughness, and has
ignored R-curve effects. These may be incorporated into the analysis by one of several
approaches, depending on the relative length scales of the R-curve and the material
property gradient. If the R-curve is very short compared to the material gradient, then only
the plateau toughness needs to be considered. Given plateau toughness as a function of
position, the above analysis could be used. If the length of the R-curve is comparable with
the scale of the material gradient, spatial variation of extrinsic toughening mechanisms
along the length of the bridging zone becomes a possibility. The contribution of bridging to
effective toughness then becomes dependent upon the position of all the points in the
bridging zone, rather than just the crack tip. This could be modelled explicitly or an
approximation could be used. These approaches are discussed in more detail in Chapter 8.
6.4

Prior to crack initiation

While understanding how a crack will behave in a gradient is important, it is also useful to
gain some insight into where cracks are likely to initiate, and how the gradient profile is
likely to influence this. This was achieved by conducting analyses for continuously graded
and step-graded beam specimens without cracks under four-point bend loading. These were
essentially identical to the simulations in Section 6.1, except for the absence of cracks. The
component of stress acting along the lower face of the beam, which is most likely to cause
crack initiation, was calculated at numerous points along the beam, in and around the
graded region. These stress profiles are plotted in Figures 6.18 to 6.21.

Figure 6.19 shows that gradient profile has a strong effect on stress profile, particularly on
the position of maximum stress concentration. Higher values for gradient profile exponent
lead to stress concentration being maximised further towards the more rigid material, as
was the case for the cracks in Figures 6.6 and 6.7.
6-23

Stress concentration factor, /P [Pa/N]

110

Epoxy

Graded Region

105
100

n=1

Alumina

n=2

n = 0.5

95
90
85
80

n=4
75
70

-2

10

12

Position, x [mm]
Figure 6.19: Stress profile along the lower face of a continuously graded beam, with gradient width
10 mm. Effect of gradient profile is shown.

Stress concentration factor, /P [Pa/N]

50000

w = 2 mm
w = 5 mm

45000

w = 7.5 mm
40000

w = 10 mm
35000

w = 15 mm
w = 20 mm

30000

25000

Graded Region

Epoxy

0.5

Alumina

Relative position across gradient,

Figure 6.20: Stress profile along the lower face of a continuously graded beam, with gradient
profile given by n = 1. Effect of gradient width is shown.

6-24

Stress concentration factor, /P [Pa/N]

50000

n=2

n=4

n=1
45000

40000

n = 0.5
n = 0.2
35000

30000

Alumina

Graded Region

Epoxy

0.5

Relative position across gradient,

Figure 6.21: Stress profile along the lower face of a continuously graded beam, with gradient width
3 mm. Effect of gradient profile is shown.

The effect of gradient width is illustrated in Figure 6.20. This shows that a reduction in
gradient width, which will increase the steepness, leads to an increase in stress
concentration, and also has an influence on the stress concentration profile. This is similar
to the results in Figure 6.5. The result indicates that the stress concentration at a given point
is related not only to the local gradient at that point, but also to the proximity of the point to
all the material property variation. This is further illustrated in Figure 6.21 for 3 mm wide
gradients with various gradient profiles, which may be compared with the results in Figure
6.19. Figure 6.22 shows that the presence of steps also has an influence on stress
concentration profile, with stress spikes at step-interfaces. Although there are decreases in
stress at some of these points as well as increases, the uncertainty introduced by the step
interfaces will compromise the reliability of the structure. This effect is also noted at the
interface between the gradient and material 1 in Figures 6.19 and 6.20.

It should be noted that the location of crack initiation will depend on the spatial distribution
of initiation susceptibility, as well as on the stress distribution across the specimen. There is
6-25

also likely to be a stochastic aspect to the initiation process. Depending on the bulk material
response to cyclic loading, the most likely point of initiation may differ between high cycle
and low cycle fatigue conditions. Also, the results here pertain specifically to the bend
geometry examined though similar analyses could be conducted for alternative geometries
and load configurations.

Stress concentration factor, /P [Pa/N]

20000

n=4
18000

n=2
16000

n = 0.5

14000

n=1
n = 0.2

12000

10000

Alumina

Graded Region

Epoxy

0.5

Relative position across gradient,

Figure 6.22: Stress profile along the lower face of a step-graded beam, with gradient width 10 mm.
Effect of gradient profile is shown.

6.5

Summary

This chapter examined, through FE simulations, the influence of elastic property gradient
and toughness gradient on stresses in uncracked specimens, mode-mixity and stress
concentration for cracked specimens and crack propagation paths.

Gradient profile, continuity and width will all influence the stress profile of the uncracked
specimen. This, combined with the distribution of initiation susceptibility, will dictate the
likely location for crack initiation. Higher profile exponent and lower gradient width lead to
6-26

higher stresses on right-hand (stiffer) side. Wider gradients lead to lower stresses overall.
After initiation, crack position and gradient parameters will determine the crack path and
SIF profile. An increase in gradient width reduces stress concentration and mode-mixity.
For lower values of gradient profile exponent, stress concentration and mode-mixity are
greater on the left side of the graded region, whereas for higher values, they are relatively
greater on the right side of the graded region, but lower overall. The presence of steps
causes significant fluctuations in the SIF magnitude profile, but has a less pronounced
effect on crack path.

The toughness distribution will depend primarily on the toughness of the constituent
materials, and also on gradient profile and continuity. To a reasonable approximation, the
toughness distribution may be estimated directly from compositional distribution, so a
stepped gradient is expected to have a stepped toughness profile. For a higher gradient
profile exponent, the toughness across most of the graded region will be closer to that of the
material on the left side, whereas for a lower profile exponent, the toughness will generally
be closer to the toughness of the material on the right.

Together, the SIF profile and toughness distribution will dictate the critical load profile.
While higher values of gradient profile exponent may result in lower stress concentration, it
could also lead to lower overall toughness if the more compliant material is less tough, as in
alumina-epoxy FGMs. For metal-ceramic FGMs, where the more compliant metal is more
also more tough, higher values of gradient profile exponent would be beneficial, as stress
concentration would be reduced whilst toughness would generally be increased.
Implications of these findings for design of optimal FGM structures are discussed in
Section 11.3.

In this chapter, simulation of crack propagation resulted in cracks with a variety of shapes.
Effects of crack shape were taken into account by the FE model; however, these effects
were not examined separately from the influence of material gradient. In the following
chapter, this issue is investigated in more detail for cracks of various shapes in
homogeneous and graded materials.
6-27

7 Simulation Results: Effect of Crack Shape

It was shown in the previous chapter that elastic mismatch leads to asymmetric crack-tip
stresses and, consequently, crack deflection. This results in non-self similar crack
propagation, where the shape of the crack changes as it extends. The aim of this chapter is
to examine the importance of crack shape.

As discussed in Section 2.3.6 (p. 2-38), the crack shape can influence crack-tip stresses,
critical loads for crack extension and subsequent propagation direction. This influence is
investigated here, via a simple analytical model, based on the Griffith concept, for cracks in
homogeneous materials. This model is verified through finite element simulations. It is
demonstrated that accurate predictions of mechanical energy release rate and crack
deflection angle may be obtained from a small number of crack-shape parameters.

The concept is then extended to curved cracks in functionally graded materials (FGMs).
This is particularly relevant, as the spatial variation in composition and properties that
characterises FGMs often leads to crack deflection and curvature, as discussed in Sections
2.6 and 2.8. It was mentioned in Section 2.8.4 (p. 2-96) that analytical and computational
models of fracture in FGMs have focussed almost exclusively on straight cracks. If it can
be demonstrated that straight cracks give an adequate approximation of curved cracks in
graded materials, then the existing solutions for straight cracks provide a sufficient
foundation for fracture analysis of FGMs. On the other hand, if straight cracks do not
adequately approximate curved cracks in FGMs, then the development of solutions for nonstraight cracks in graded materials is a priority. The question, of whether existing solutions
for straight cracks in graded materials may be validly applied to curved cracks, or whether
solutions should be developed for other crack shapes in graded materials, is addressed here
through FE simulations of crack propagation using a number of crack-shape
approximations.
7.1

Quantifying crack shape

Although a crack shape may be quite complicated, several researchers have suggested that
the stress intensity factor may be approximated by a function of several simple crack-shape
7-1

parameters, or indeed by approximating the curved crack to an equivalent straight crack. As

discussed in Section 2.3.6 (p. 2-39), the relevance of these types of approximation has been
demonstrated for particular crack-shapes and loading configurations, although reasons were
not given for why this is so [Kitagawa et al., 1975, Leevers et al., 1976, Alpa et al., 1980,
Noda et al., 1993, 1994]. Approximation to an equivalent straight crack, or other simplified
crack-shape, simplifies calculation appreciably and may often be sufficiently accurate. In
fact, this approximation is made when treating any macroscopically straight crack as being
perfectly straight, despite the fairly ubiquitous presence of crack deflection at a microscopic
level. The validity of the approximation in this case is generally not questioned. For
macroscopically deflecting cracks, this approximation is clearly useful for simply
estimating fracture parameters, and thus warrants further investigation.

da

Edge
Figure 7.1: Example of deviating crack showing key parameters: crack length normal to edge, a;
relative transverse deviation d; crack-tip angle, ; and loading angle, .

It is proposed that the shape of a crack may be described by several parameters, from which
an approximate SIF value may be predicted. For an edge crack, a common configuration in
experimental fracture mechanics, the proposed parameters, shown in Figure 7.1, are:

crack length perpendicular to the edge, a;

transverse deviation parallel to the edge, d;

crack-tip orientation relative to loading direction, ;

7-2

The geometry of the deviating crack is shown in Figure 7.1, along with the key crack shape
parameters describing crack and loading configuration. The applied loading may have
tensile and shear components, and is described by:

applied principal stress, which would exist in the absence of the crack, ,

loading direction relative to the edge, .

It is assumed for simplicity that the effects of specimen geometry may be ignored and that
loading is in the form of far-field stress.
7.2

Analytical model

A simplified approach given by Ashby and Jones [1996], for calculating the mechanical
energy release rate, based on the assumption of an elliptical region of stress relaxation
around the crack and the increase in this region with crack extension, was presented in
Section 2.3.2 (p. 2-27). For a straight crack of length, 2a, as in Figure 7.2(a), the rate of
energy release, G, associated with a crack extension, A = 2a, was determined as:
dU rel
dU rel dA
G=
=

dA
da da

d k 2 a 2 t 1 k 2 a

=
=
da 2E ' 2 t
2E '

(7.1)

where E is the effective Youngs modulus, and k describes the aspect ratio of the
relaxation volume. It can be shown that k is 2, for a through-thickness straight crack in an
infinite medium.

(a)
ka
Relaxed
material

Loading
angle

(b)

d a

2a

aeff

Transverse
Deviation

kaeff

Relaxed
material

Figure 7.2: Stress relaxation volume around (a) a straight internal crack under far-field tensile
loading and (b) a curved edge-crack under far-field mixed-mode loading

7-3

Although this is a very simplistic approach, the form of the expression for mechanical
energy release rate is correct. The reason for this is that the stress distribution, though it is
actually more complicated than assumed, still scales with the square of length. It is believed
this approach may be extended to more complicated crack configurations, whilst
maintaining the intuitive aspect which makes it appealing. For a curved edge-crack as in
Figure 7.1, the relaxation volume is taken as an elliptical section with the major axis of the
ellipse along the loading direction and the end of the semi-minor axis coincident with the
crack tip, as shown in Figure 7.2(b). In the case of an edge crack, it is assumed that the
centre of the ellipse lies on the edge of the specimen, although not necessarily at the crack
mouth. The key assumption is thus that the shape of the relaxation volume does not depend
on the shape of the crack.

(a)

(b)
aeffc

Loading
direction

Figure 7.3: Extension, c, from crack tip (a) along crack-tip direction, , and associated increase in
effective crack length, aeff, and (b) along optimal kinking direction, k, so that the effective crack
length increase is maximised, ie. aeff = c.

It should be noted that this ellipse is regarded as virtual. It is not being suggested that the
stress is totally relaxed inside the shape and completely unaffected outside. Rather, it is
representative of the degree of stress relaxation which occurs in the proximity of the crack.

7-4

Relaxation volume may be scaled with the square of crack length, hence the derivative will
scale linearly with the crack length.

7.2.1 Mechanical energy release rate

The stress relaxation volume is determined from the effective crack length, aeff, in the
direction normal to the loading axis, as shown in Figure 7.2(b):
a eff = a cos + d a sin

(7.2)

Hence the volume of the elliptical prism of relaxed material, assuming unit thickness in the
z-direction, is:
2
Vrel = 12 a eff

(7.3)

where is an unknown geometrical constant, defining the aspect ratio of the ellipse. This
leads to an energy release of:
U rel

2
2 2
=
Vrel =
a eff (cos + d sin ) 2
2E '
4E '

(7.4)

Assuming crack growth occurs along the current crack-tip orientation, as in Figure 7.3(a),
an advance of the crack tip, c, leads to an increase in effective crack length, aeff, of:
a eff = c cos( )

(7.5)

Therefore the relaxation volume increases to:

Vrel + Vrel =

[a (cos + d sin ) + c cos( )]2

2

(7.6)

The increase in relaxation volume, Vrel, is:

Vrel =

2a (cos + d sin )c cos( ) + (c) 2 cos 2 ( )

2

(7.7)

which results in an increase in energy release of dU:

U rel =

2
2a (cos + d sin )c cos( ) + (c) 2 cos 2 ( )
4E '

(7.8)

The increase in crack area, A, is equivalent to the crack extension, c, due to the
assumption of unit thickness. The mechanical energy release rate is calculated from Urel
and A:
U rel 2
=
[2a (cos + d sin ) cos( )]
c 0 A
4E '

G = lim

(7.9)
7-5

To estimate the value of , the simple case of a straight crack under tensile loading is
considered. In this case, parameters d, and are zero, and:
G=

2 a
2E '

(7.10)

This may be compared with the known expression for an edge crack [Broek, 1991]:
G=

K 2 1.12 2 2 a
=
E'
E'

(7.11)

and the geometrical constant is thus determined as = 2(1.12)2 2.5.

7.2.2 Deflection angle

Whilst a number of different criteria have been used to predict deflection angle, the focus of
the current analysis is strain energy release rate. Accordingly, the maximum energy release
rate criterion is most appropriate. Considering a kink extending at angle from the tip of a
particular curved crack, as in Figure 7.3(b), the mechanical energy release rate is estimated
from Equation 7.9 as:
G () =

2
[a (cos + d sin ) cos( + )]
2E '

(7.12)

This is maximised for:

dG ()
=0
d

(7.13)

2
[a (cos + d sin ) sin( + )] = 0
2E '

ie.

ie.

sin( + ) = 0

(7.14)

This yields the optimum value of the kink angle, :

k =

(7.15)

This implies that mechanical energy release rate will be maximised for crack growth
perpendicular to the loading direction, which is generally understood to be the case.

This simple analytical model may not be validly applied to FGMs, as the elastic property
gradient, leads to a spatial variation in strain energy density. It was assumed, however, that
curved cracks in FGMs, by analogy with those in homogeneous materials, may also be
7-6

approximated by simple-shaped cracks. This was investigated with FE simulations, using

several different crack-shape approximations.
7.3

FE Simulations

These were conducted in two stages. In Stage I, stationary cracks with a range of shapes in
homogeneous specimens were modelled to obtain values for mechanical energy release rate
and deflection angle. In Stage II, propagation of cracks under mixed-mode loading in
homogeneous and graded specimens was simulated, and crack paths and SIF profiles were
obtained.

The general specimen configuration for Stage I, an edge-cracked size plate with stresses
applied to edges of plate, and a representative mesh, are shown in Figure 7.4(a). Deflection
angle was predicted using the maximum energy release rate criterion for consistency with
the analytical model.

(a)

bspec

(b)

r
A

Figure 7.4: Finite element modelling of curved cracks (a) Example of finite element mesh, showing
specimen width and boundary conditions. (b) Generic crack shapes used in FE validations: A
Concave, B Piecewise Linear, C Convex, showing shape parameter, w, used to vary crack
shape.

7-7

To investigate a range of crack shapes, a degree of variability within the constraints of the
set parameter scheme was introduced by varying the curvature of the cracks, whilst keeping
the defined parameters constant. Three different general crack shapes were considered, A,
B and C, as shown in Figure 7.4(b). By using each of these for each set of values of a, d,
and , the sensitivity to curvature could be examined. It should be noted that, in the
geometrical scheme used, the radius of curvature, r, is proportional to a and d, so that when
d is varied, the relative curvature, ie. r/a, varies also.

For the propagation simulations, propagation direction was predicted using either the
maximum energy release rate (MERR) criterion, with mechanical energy release rate
calculated from SIFs (Equation 2.28), or the maximum tangential stress (MTS) criterion,
with SIFs calculated from displacement correlation (Equation 5.6). The MERR criterion is
more reliable as it takes T-stresses into account but the MTS criterion is generally easier to
implement in a finite-element model, as it does not require a test-kink. Cracks were
extended by c = 500 um at each increment, with the new crack-tip position calculated
from the crack-tip position and deflection angle.

When the entire crack shape was used, the finite extension per increment resulted in a path
made up of numerous small line segments. As the propagation direction did not change
abruptly, these formed a smooth path. The effect of using approximated crack shapes on
predictions of crack-path and stress concentration was examined. When simplified crackshapes, as shown in Figure 7.5(a), were used, the crack was extended in the same way;
however, after each crack extension, a new approximated crack-shape was defined and
meshed. This approximated crack shape was used for calculation of the propagation
direction for the subsequent increment. This is illustrated schematically in Figure 7.5(d).
The crack-shape geometry and surrounding mesh was deleted and redefined between each
increment of propagation. The crack paths presented here are essentially plots of the
variation of crack-tip location, as calculated using a variety of crack-shape approximations.
Results obtained from simulations that used the entire crack-path shape, and employed the
MERR criterion, were regarded as the reference, or correct, results for each
configuration.
7-8

(b)
Actual

60 mm

Material 2

Material 1

3PB Load point

Material 2

Piecewise-Linear

(c)

Diagonal

Notch Tip

120 mm

30 mm

(a)

4PB Load
points

Interface Regions

Straight

(d)

Straight

Diagonal

Piecewise

Figure 7.5. Crack-path predictions. (a) Illustration of crack-shapes approximations used. (b)
Schematic of specimen and loading geometry. (c) Representative FE mesh for crack growth
simulations. (d) Examples of the sequence of crack shapes used for calculations at each increment

It should be emphasised that the use of approximated crack shapes in this manner did not
improve simulation efficiency, except for simplifying the mesh pattern around the crack
which could slightly reduce the number of elements and hence the computation time.
Rather, the efficiency of analytical solutions would be improved notably if a simplified
crack shape could be used. Establishing the validity of this was a principal aim of this
study.

This approach was applied to several different material configurations. The general
specimen geometry, an edge-cracked beam in bending, is depicted in Figure 7.5(b) and a
representative example of the finite element mesh is shown in Figure 7.5(c). Specimens
containing the following material interface configurations were examined:
(1) no interface (asymmetrically-notched homogeneous specimen), with the notch
situated 36mm from the load-line (centre of specimen);
7-9

(2) step interface (bimaterial specimen), with crack initially situated 8 mm from the
interface;
(3) continuously-graded interface, with linear spatial variation in material properties
across a 10 mm wide gradient region, and crack initially situated 6 mm from the
interface between the gradient region and Material 1;
(4) continuously-graded interface, with parabolic spatial variation in material properties
across a 10 mm wide gradient region, and crack initially situated 8 mm from the
interface between the gradient region and Material 1;
(5) discontinuously-graded interface, with stepped-linear (3 steps) spatial variation in
material properties across a 10 mm wide gradient region, and crack initially situated 8
mm from the interface between the gradient region and Material 1.
Four-point bend loading was used for all configurations except the homogeneous specimen.
This was subjected to symmetric three-point loading in order to obtain mixed-mode
loading. The positions of the loading points for three-point bend (3PB) and four-point bend
(4PB) configurations are shown in Figure 7.5(b), along with specimen dimensions. The
assumed properties of Material 1 were E1 = 3.4, 1 = 0.25. Properties of Material 2 were
assumed as E2 = 340 and 2 = 0.25 for all cases except the homogeneous specimen, which
Materials 1 and 2 were identical. The large disparity in elastic properties was assumed so
that deflection angles would be high, thereby amplifying differences which may arise
between predictions from different crack-shape approximations.

7.4

Results: Curved cracks in homogeneous materials

The predictions from this analytical model are shown in Figures 7.6 to 7.10, along with
results obtained from finite element analysis. The FE results are shown as dots or triangles,
with a degree of scatter between results obtained for different crack shape types, A, B and
C, whilst analytical model results are shown as solid lines. Mechanical energy release rate
values are normalised with respect to effective Youngs modulus, E, and applied stress, ,
so that the effects of crack geometry may be examined directly. However, as applied stress
and effective Youngs modulus were held constant for all cases examined, the absolute and
normalised values for mechanical energy release rate are directly proportional.
7-10

In Figures 7.6(a) and 7.6(b), the effect of variation in crack length on the fracture
parameters is displayed for cracks with loading angles of 0 and 30 respectively. Crack
length values are normalised with respect to specimen width, bspec (clarify the use of
a/bspec). Mechanical energy release rate increases with crack length, as expected from basic
fracture theory. For short cracks, G is directly proportional to a, in accordance with the
ideal relation between energy release rate and flaw size, however for longer cracks,
specimen geometry becomes an influencing factor. Due to the finite geometry of the FEA
model used, a specimen geometry effect is observed for longer crack lengths, as predicted
by basic fracture theory [Broek, 1991]. To account for this, the model may be modified by
considering the increase in stress due to the finite width of the remaining ligament, h-a.
Accordingly, there is an increase in effective local stress inversely proportional to the
remaining ligament size:
eff =

ha

(7.16)

Incorporating this specimen geometry effect improves the agreement between predictions
and results. For cracks significantly smaller than the sample, length does not appear to
influence deflection angle. The model clearly captures the increase in G with crack length.
This might be expected as the model was developed from the simple straight crack model
which related crack length to G. In both cases, the stress fields will generally scale with
crack length in two dimensions, so G should be proportional to crack length.

In Figure 7.6(a), there is a systematic difference between deflection angles predicted with
the three different crack shapes. Predictions from crack-shape A are generally highest,
whilst those from crack-shape C are lowest, with approximately 10% variation between
predictions. This trend is not reflected in the predicted values of G, which are essentially
independent of crack shape, nor in the deflection angles in Figure 7.6(b) for the inclined
loading case. This indicates that, although curvature can have a systematic influence on
deflection angle, this is fairly minor.

7-11

- Analytical

50

Propagation Direction, k [deg]

(b)

40

- FEA

A
B

30

G - Analytical 20
- corrected

G - FEA

G - Analytical

10

0
0

0.05

0.1

0.15

Relative Crack Length, a/h

0.2
spec

30

25

- FEA

Propagation Direction, k [deg]

Normalised Energy Release Rate, GE/2 [m] Normalised Energy Release Rate, GE/2 [m]

(a)

20
5

15

- Analytical

10

G - Analytical
- corrected 0

G - FEA

2
1
0

-5

G - Analytical

-10
0

0.05

0.1

0.15

Relative Crack Length, a/h

0.2
spec

Figure 7.6: Effect of crack length, a, normalised by specimen width, bspec, on mechanical energy
release rate and propagation direction. Comparison of analytical and FE predictions for cracks (d =
0.2, = 45.8) under (a) tensile ( = 0) and (b) inclined ( = 30) loading. An increase in energy
release rate with crack length is observed, whilst deflection angle remains fairly constant.

7-12

1.6

- FEA

50

k - Analytical

1.4
40
1.2

G - Analytical
30

1
0.8

20

G - FEA

0.6

10
0.4
0.2

Propagation Direction, [deg]

(b)

0
0

0.1

0.2

0.3

0.4

0.5

Relative Transverse Deviation, d

30
1.6

G - FEA
1.4

25

G - Analytical

1.2

20

- FEA
k

Propagation Direction, [deg]

2
Normalised Energy Release Rate, GE/2 [m] Normalised Energy Release Rate, GE/ [m]

(a)

0.6

15

- Analytical

0.8

0.1

0.2

0.3

0.4

0.5

10

Relative Transverse Deviation, d

Figure 7.7: Effect of transverse deviation, d, on mechanical energy release rate and propagation
direction. Comparison of analytical and FE predictions for cracks under (a) tensile ( = 0) and (b)
inclined ( = 30) loading (a/bspec = 0.06, = 45.8). The influence of transverse deviation on
deflection angle, for cracks under inclined loading (b), is not captured by the analytical model.

7-13

Normalised Energy Release Rate, GE/2 [m]

1.6
50

G - FEA
1.4

Propagation Direction, k [deg]

40
30

1.2

G - Analytical

20
1

- FEA

10

- Analytical
k

0.8

-10

0.6

-20
0.4

10

20

30

40

50

60

70

Crack-tip Orientation, [deg]

1.6

80

80

(b)

G - FEA

1.4

60

1.2
1

G - Analytical

40

k - FEA

0.8

20

0.6

Propagation Direction, [deg]

Normalised Energy Release Rate, GE/ [m]

(a)

- Analytical

0.4

0
0.2

10

20

30

40

50

60

70

Crack-tip Orientation, [deg]

80

Figure 7.8: Effect of crack-tip orientation on mechanical energy release rate and propagation
direction. Comparison of analytical and FE predictions for cracks (a/bspec = 0.06, d = 0.2) under (a)
tensile ( = 0) and (b) inclined ( = 30) loading. Note that G is maximised and deflection angle
0 for crack-tip orientation equal to loading angle, , ie. crack tip normal to loading direction.

7-14

The effect of variation in relative deviation, d, is illustrated in Figures 7.7(a) and 7.7(b) for
cracks with loading angles of = 0 and = 30 respectively. There is a systematic
influence of d on the deflection angle in Figure 7.7(b). Although the difference is larger for
larger values of d, the influence of d seems to be greater for smaller d values, as the slope
of the variation of with d is greater in that region. Variation in d leads to a variation in the
curvature at the crack tip, so the influence of d is likely to be attributable to curvature.
The effect of variation in crack-tip angle, , is illustrated in Figures 7.8(a) and 7.8(b) for
cracks with loading angles of = 0 and = 30 respectively. Crack-tip orientation is seen
to influence fracture parameters significantly. Increasing the crack-tip angle leads to an
increase in deflection angle, and a decrease in mechanical energy release rate. The effect of
variation in loading angle is illustrated in Figure 7.9. Very close agreement is observed
between the model and FE predictions for both energy release rate and deflection angle.

The effect of variation in crack shape is illustrated in Figures 7.10(a) and 7.10(b) for cracks
with loading angles of = 0 and = 30 respectively. In these cases, the key shape
parameters remain constant whilst the shape of the crack is varied. This was achieved by
varying the radius of curvature in the circular arc section of the crack shape. Crack
curvature has only a minor influence on fracture parameters, as anticipated in the original
hypothesis. This implies that a wide range of cracks may be simulated reasonably
accurately by analogy with simplified crack shapes. Crack curvature does seem to have a
slight effect, especially when the radius of curvature near the crack tip is small. This is in
agreement with the findings of Kitagawa et al. [1975].

To provide further validation, the simple analytical model was used to predict mechanical
energy release rate and deflection angle values for several types of curved crack previously
investigated by Noda and co-workers [1993, 1994], and mentioned in Section 2.3.6 (p. 239). Values of shape parameters, a, d, and r, were determined for the required crack
shapes. Results are compared with those of Noda et al. [1994] in Figures 7.11 for circular
arc-shaped surface cracks with crack-mouth normal to the surface, equivalent to = .

7-15

Good agreement is observed except in the case of very high angles. Further results are

60

1.5

G - FEA

50

1.25

40

G - Analytical
1

30

- FEA

0.75

20

10

- Analytical

0.5

0.25

Propagation Direction, k [deg]

Normalised Energy Release Rate, GE/2 [m]

given in Appendix E.

-10

16

24

32

40

48

Loading Direction, [deg]

56

-20
64

Figure 7.9: Effect of loading direction, , on mechanical energy release rate and propagation
direction. Comparison of analytical and FE predictions for cracks (a/bspec = 0.06, d = 0.2, = 45.8)
under inclined loading. Note that G is maximised, and deflection angle 0, for loading direction is
equal to crack-tip orientation, , ie. loading normal to crack-tip orientation direction.

7-16

Normalised Energy Release Rate, GE/2 [m]

70

1.2

G - Analytical

G - FEA

Propagation Direction, k [deg]

(b)

60

0.8

- Analytical

50

0.6

40

- FEA

0.4

30
0.2

20

-1

Shape Parameter, ln(w)

30

1.6

G - FEA
1.4

Propagation Direction, k [deg]

Normalised Energy Release Rate, GE/2 [m]

(a)

25

1.2

k - FEA

G - Analytical

20

15

- Analytical

0.8

10
0.6
5
0.4
0.2

0
-1.6

-0.8

0.8

1.6

2.4

3.2

4.8

Curvature, ln(r/a)

Figure 7.10: Effect of variation in radius of curvature at crack tip on mechanical energy release rate
and propagation direction. Comparison of analytical and FE predictions for cracks under (a) tensile
( = 0) and (b) inclined ( = 30) loading (a/bspec = 0.06, d = 0.2, = 45.8). Except for a slight
influence on deflection angle observed in (a), the effect of curvature is negligible, as assumed in the
analytical model.

7-17

Normalised Energy Release Rate, GE/ [m]

Noda et al
Noda et al - Equiv Straight
Present Analytical Model

70

0.4

60

50

0.3

40

0.2

30
20

0.1

Predicted Deflection Angle, [deg]

0.5

10

0
0

10

20

30

40

50

60

Crack Arc Angle, [deg]

70

Figure 7.11: Mechanical energy release rate predictions, for surface crack with circular arc shape
and crack-mouth normal to surface, compared with results of Noda et al. [1994] for arc crack and
equivalent straight crack.

7.5

Results: Curved cracks in graded materials

Crack paths and SIF magnitude profiles obtained using each of the approximated crack
shapes are compared with those obtained using the entire crack-path shape, in Figures 7.12
to 7.16. SIF magnitudes have been normalised with respect to the SIF for a straight crack of
the same length in a homogeneous specimen of equivalent dimensions under four-point
bend loading, as calculated with FE. This allows the influence of material property
variation to be examined independent of the effects of crack length, which would otherwise
dominate the variation of SIF magnitude. The normalised SIF magnitude profiles are
plotted against crack-tip position in the horizontal direction so that variation across the
interface region may be compared between the different crack-shape approximations. The
results in the previous chapter showed that crack-tip position is a dominant influence on
SIF magnitude and mode-mixity. Plotting in this way also enables direct comparison with
the corresponding crack path figure.
7-18

Predictions from the MERR and MTS criteria differed, even when the exact crack shape
was used, so paths predicted with each criteria are included in the plots. Only SIF
magnitudes calculated in simulations using the MERR criterion are presented, however, as
the values from simulations with the MTS criterion were almost identical, differing only
due to slight differences in propagation path.

Figure 7.12 shows results for the asymmetrically-notched homogeneous specimen. Path
predictions and SIF magnitudes obtained using crack-shape approximations show very
good agreement with those obtained using the entire crack-path. In the homogeneous
material specimen in Figure 7.12(b), the use of an approximated crack shape appears to
have almost no effect at all on SIF magnitude predictions. Results for the bimaterial
specimen are presented in Figure 7.13. Deviation between results from approximated
crack-shapes and the entire crack-path is more significant in this case. Figures 7.14 and
7.15 show the results for the two continuously-graded specimens with differing gradient
steepness. Results for the discontinuously-graded (step-graded) specimen, which contained
three steps, are shown in Figure 7.16.

The presence of material inhomogeneity, in Figures 7.13 to 7.16, seems to cause greater
deviations between results from approximated crack shapes and those obtained using the
entire crack-path. In inhomogeneous specimens, the results predicted using straight and
diagonal approximations deviated significantly from the results for exact crack-shape.
These approximations tended to result in less deflection. This could be attributed to a slight
skewing of the crack-tip stress field, caused by the difference in crack-tip orientation of the
approximated and exact crack shapes. The piecewise-linear approximation gave good
predictions for all the configurations examined.
The crack paths show a cumulative deviation effect. In the first increment of crack growth,
before any deflection has occurred, the exact and approximated crack shapes are identical,
then they become progressively more different as the crack propagates, as was shown
schematically in Figure 7.5(d). Concordant with this, differences in the crack propagation
direction, due to the use of a crack-shape approximation, are initially very small then these
7-19

increase as the crack extends. Furthermore, the differences accumulate, and the differences
between crack paths lead to further divergence.

For cracks in the specimens with some material inhomogeneity, the SIF magnitude varied
significantly across the interface region. In Figure 7.13(b), it increases sharply as the
bimaterial interface is approached. A similar effect is observed as the crack approaches the
two interfaces in the step-graded specimen in Figure 7.16(b). A more continuous variation
in SIF magnitude is observed for the continuously-graded specimens, though these still
exhibit notable variation with crack-tip position. The contrast between the results for linear
property variation in Figure 7.14(b) and those for parabolic property variation in Figure
7.15(b) highlights the importance of material property distribution in influencing crack-tip
stresses.
For the crack approaching the bimaterial interface in Figure 7.13(b), SIF magnitudes varied
significantly between simulations using different crack-shape approximations. For a crack
with tip situated 1 mm from the bimaterial interface, for instance, the relative SIF
magnitude calculated for using exact crack-shape is 1.8 times that calculated using the
straight crack approximation. This means that using the straight crack approximation would
lead to an overestimate of critical load by a factor of 1.8. Similar comparisons may be made
for cracks in continuously-graded and step-graded specimens. These also show significant
differences between SIF magnitudes obtained from different crack-shape approximations.

7-20

14

(a)

13
12
11
10

y [mm]

9
8
7

Crack-tip
position A

Exact MERR
Exact MTS
Straight MERR
Straight MTS
Diag. MERR
Diag. MTS
P/wise MERR
P/wise MTS

5
4
3
2
1

3/2
3

(b)

Normalised SIF Magnitude, |K|/P [x 10 m ]

x [mm]

21
18
15
12
9
6

Exact
Straight
Diagonal
Piecewise

3
0

Crack-tip position, x [mm]

Figure 7.12. Crack propagation trajectories predicted using exact and approximated crack-shapes
for an asymmetrically notched homogeneous beam under three-point loading. Crack-tip position A
(x = 3mm, y = 8.1mm) is referred to in the text.

7-21

10

(a)

Bimaterial
Exact MERR
Interface
Exact MTS
Straight MERR
Straight MTS
Diagonal MERR
Diagonal MTS
Piecewise MERR
Piecewise MTS

y [mm]

1
0

x [mm]

12
Exact
Straight
Diagonal
Piecewise

hom

Normalised SIF Magnitude, |K|/K

(b)

10

Bimaterial
Interface

Compare
SIF values
6

Crack-tip position, x [mm]

Figure 7.13. Crack propagation trajectories predicted using exact and approximated crack-shapes
for a bimaterial specimen under four-point loading.

7-22

Gradient/Material 1
Exact MERR
Interface
Exact MTS
Straight MERR
Straight MTS
Diagonal MERR
Diag. MTS
P/wise MERR
P/wise MTS

(a)
8

y [mm]

Crack-tip
position B

Normalised SIF Magnitude, |K|/K

(b)

hom

x [mm]

Exact
Straight
Diagonal
Piecewise

Gradient/Material 1
Interface
3

Crack-tip position, x [mm]

Figure 7.14. Crack propagation trajectories predicted using exact and approximated crack-shapes
for graded specimens with linear variation in Youngs modulus, under four-point loading. Crack-tip
position B (x = 3mm, y = 2.25mm) is referred to in the text.

7-23

(a)

12
11
10
9
8

y [mm]

Gradient/Material 1
Interface

Exact MERR
Exact MTS
Straight MERR
Straight MTS
Diag. MERR
Diag. MTS
P/wise MERR
P/wise MTS

7
6
5
4
3
2
1
0

(b)

x [mm]

Normalised SIF Magnitude, |K|/K

hom

Exact
Straight
Diagonal
Piecewise

1
Gradient/Material 1
Interface
0

Crack-tip position, x [mm]

Figure 7.15. Crack propagation trajectories predicted using exact and approximated crack-shapes
for graded specimens with parabolic variation in Youngs modulus, under four-point loading.

7-24

(a)

10

Exact MERR
Exact MTS
Straight MERR
Straight MTS
Diag. MERR
Diag. MTS
P/wise MERR
P/wise MTS

y [mm]

Step
Interface
5

Gradient/Material 1
Interface

x [mm]

Normalised SIF Magnitude, |K|/K

hom

(b)

Step
Interface

Exact
Straight
Diagonal
Piecewise

5
4
3
2
1
0

Gradient/Material 1
Interface
0

Crack-tip position, x [mm]

Figure 7.16. Crack propagation trajectories predicted using exact and approximated crack-shapes
for a graded specimen, with a stepped linear variation in Youngs modulus, under four-point bend
loading.

7-25

The distinction between homogeneous and inhomogeneous specimens, in terms of the

effect of using a crack shape approximation on SIF magnitude values, is further illustrated
by considering cracks, which have traversed a horizontal distance of 3mm, along the path
predicted using the entire crack shape. The crack-tip positions are denoted in Figures
7.12(a) and 7.14(a) for homogeneous and linear continuously graded specimens
respectively. For the homogeneous specimen, the relative SIF magnitude values obtained
using the exact shape and using the diagonal crack-shape approximation were 1.112 and
1.098 respectively, which differ by less than 2%. For the linear continuously-graded
specimen, however, relative SIF magnitude values differed by almost 20%, with 3.788 for
the exact shape and 3.126 for the diagonal crack-shape approximation. This provides
further evidence that crack-shape is a more important influence in graded or layered
materials, than in homogeneous materials.
7.6

Discussion

Initially addressing the results for curved cracks in homogeneous materials in Figures 7.6 to
7.11, there was generally good agreement between results from the simple analytical model
and those from the FE simulations, for both mechanical energy release rate and deflection
angle values. Trends in the variation of these with each of the shape parameters, observed
in FE results, were reflected in the model predictions. The only exceptions were the slight
influences of transverse deviation and shape parameter on deflection angle. The model is
based on the assumption that the region over which stresses relax may be described by
some length scale that is proportional to the square root of this relaxation region. Each
parameter of the crack shape will contribute to this length scale. For the case of an edgecrack in a homogeneous material, this assumption is apparently valid. The model appears to
encapsulate the contribution of each parameter to the size scale of the stress relaxation
region. This is essentially a scaling argument; the stress field scales with crack size, so the
relaxation volume and energy release rate will also. The validity of this is compromised
when the crack is growing in a non-self-similar material, for instance near a hole, near an
interface or in a compositionally graded region.

7-26

The influence of curvature, which is not included in the analytical model, is minor
compared with the other parameters, which are taken into account. This appears to be the
main source of disparities between predictions from the model and those from FEA. This
was observed primarily in Figures 7.6 and 7.9, in which the curvature was varied, either
directly or indirectly, by varying the radius of curvature. This could be attributed to the
crack shape near the crack tip influencing the flow of stresses around the tip. This influence
of crack shape is also observed in earlier figures where predictions using the different crack
shapes, A, B and C, defined in Figure 7.4 lead to slightly different results. In general the
convex crack shapes (A) lead to higher values of deflection angle and the concave shapes
(C) lead to lower values. This effect was not observed in all data sets and did not extend to
mechanical energy release rates. It is therefore concluded that curvature at the crack tip
does have an effect, mainly on predicted deflection angle, but that this is limited, and is
secondary to the influences of other crack-shape parameters.

Although the crack shapes and loading configurations considered were somewhat idealised,
it is conceivable that other more realistic situations may lend themselves to analysis with
this model. One example might be a crack in a laminated composite material that, due to
toughness anisotropy, is constrained to propagate in mixed-mode [Hutchinson & Suo,
1992, Selvarathinam, 1998], as discussed in Section 2.5.4 (p. 2-62). Another may be a
crack that, under variable-direction fatigue loading, has propagated in several modes,
resulting in a non-straight crack path [Plank & Kuhn, 1999].

There is also a certain intrinsic value in this concept. It demonstrates that a simple
physically-based model can provide useful results, without requiring large amounts of
complicated mathematics or computation. Furthermore, the model provides a more detailed
explanation, of why curved cracks may be approximated by straight cracks, than had
previously been suggested. This concept could certainly be applied to other material
configurations, for instance cracks in graded materials or near interfaces. The analytical
model proposed in the present study is interesting for its simplicity and remarkable
accuracy, at least in the cases examined. More importantly, the findings in this study

7-27

provide strong justification for the use of approximated crack shapes, which will certainly
be required in various crack propagation models.

Focussing now on the results for propagating cracks, in Figures 7.12 to 7.16, the piecewiselinear approximation gave good predictions for all the configurations examined. It is noted
however that the crack paths in the configurations considered did not deviate significantly
from a piece-wise linear shape. In other configurations, this may not be the case.

Comparing Figures 7.12 and 7.13, while cracks in bimaterial and step-gradient specimens
are both propagating through homogeneous material regions, there is a significant degree of
deviation between results from different crack-shapes, which is not observed for the
homogeneous specimen. This implies that the presence of material inhomogeneity, even if
remote from the crack, leads to a more significant dependence on crack-shape, and a higher
sensitivity to the use of crack-shape approximations. In the homogeneous specimen, the
crack deflection occurs due to a geometric effect, which does not vary when the crack
propagates. In the other specimens, crack deflection and curvature results from material
property asymmetry. This appears to depend strongly on the crack-tip orientation as well as
position. This could be attributed to the cause of deflection being material property
mismatch, which acts locally and can result in high T-stresses, so that a different crack-tip
orientation will result in a different interaction with the local material gradient.
The inclusion of the MTS criterion predictions for propagation paths, which generally
differed from the MERR criterion predictions, served to demonstrate that the choice of
deflection criterion can influence crack path predictions as significantly, if not more so,
than the use of an approximated crack path.

While crack path predictions can be compared directly with real cracks, these predictions
are, from a structural integrity perspective, secondary to predictions of critical load. The
accuracy of critical load predictions is dependent upon the accuracy of mechanical energy
release rate, or SIF magnitude, calculations. The SIF magnitude profiles examined suggest
that such calculations are quite sensitive to the use of crack-shape approximations. The
influence of crack shape on SIF magnitude in graded materials appears to be fundamentally
7-28

different to that in homogeneous materials. This is understood to be due to the influence of

material property distribution around the crack on stress concentration and mode-mixity.
The use of an approximated crack-shape results in a significantly different distribution of
properties around the crack, so crack-tip stress fields will be altered.

A key motivation for this study was to determine whether the existing solutions for straight
cracks in graded materials could be relevantly applied to deflected or curved cracks in
graded materials. It appears that reasonable estimates may be obtained using solutions for
straight or diagonal (inclined straight) crack shapes, though the deflection angle will tend to
be underestimated slightly. The superiority of predictions from piecewise linear crack shape
approximations indicates that the development of analytical solutions for branched or
piecewise linear cracks in graded materials is a useful area for further work. Such solutions
will almost certainly be more easily obtained than solutions for exact crack shapes and will
offer a significant improvement in accuracy over straight crack solutions.

It may be inferred that crack shape does have a significant effect on the predicted optimum
propagation direction for a particular crack. Using approximated crack shapes to predict
propagation can lead to large cumulative errors, due to the effect of path deviation on
subsequent propagation. While solutions for straight cracks may be adequate for crudely
approximating curved cracks in FGMs, piecewise-linear shapes provide a much better
approximation.
7.7

Summary

An analytical model has been developed to predict the mechanical energy release rate,
mode-mixity, stress intensity factors and crack deflection angle for deviating cracks under
mixed-mode loading. It is based on a highly simplified and intuitive approach, considering
stress relaxation around the crack. Predictions from the model were compared with those
obtained from finite element analysis. Then crack propagation in homogeneous and graded
materials was simulated using finite element analysis. Several conclusions were reached:
1. Curved cracks in homogeneous materials may be reasonably approximated by
equivalent straight cracks for the calculation of mechanical energy release rates and
deflection angles.
7-29

2. The approximation of non-straight cracks by equivalent straight cracks may be

understood in terms of the volume of material around the crack from which strain
energy will be released during crack propagation. Estimates from this approach are
reasonably accurate, to within 10% for energy release rate values and within 5 for
deflection angle predictions.
3. Crack length and crack-tip orientation are the key factors influencing mechanical
energy release rate, and their influence was predicted by the simple model.
4. Curvature at the crack-tip also appears to influence mechanical energy release rate and
optimal propagation direction, which was not accounted for by the analytical model.
This led to some disparity in predicting the effect of variation in transverse deviation or
curvature on results.
5. Cracks in graded materials may also be reasonably approximated by equivalent straight
cracks, although systematic divergence occurs between simulations using the entire
crack shape and those using approximated crack shapes.
6. Piecewise linear crack shapes provide a significantly better approximation than
diagonal or straight crack shapes. Accordingly, analytical solutions for piecewise linear
cracks in graded materials would be very useful, and should be a focus of further work
in this area.
7. When a crack-shape approximation is used, predictions of crack path tend to be better
than predictions of SIF magnitude. This is attributed to the sensitivity of SIF magnitude
to material property distribution around the crack and crack-tip orientation.

It may be instructive to reconsider the results in Chapter 6 in light of these findings. While
those simulations did take crack curvature into account implicitly it was not separated from
the effects of material gradient. Had crack shape not been taken into account, the findings
in the present chapter suggest that SIF magnitude and propagation path predictions would
have become quite inaccurate. It is deemed that, for the present study, the influence of
crack shape should be taken into account; this was done for simulations in Chapters 6 and
8, and simulations of experimental specimens in Chapter 10.

7-30

8 Simulation Results: Non-linear Effects

The two preceding chapters examined the influences of material stiffness gradient and
crack and specimen geometry, all within the linear elastic regime. While this is very useful,
and may often be sufficient for analysing certain FGM systems, the question arises as to
how non-linear behaviour is likely to influence crack growth and structural integrity. In this
case, the term non-linear refers to any effect that results in a non-linear, or nonproportional, relationship between applied load and crack-tip stress intensity factor. In this
chapter, results are presented from the finite element simulations incorporating crackbridging, plasticity and thermal residual stresses, each of which introduces a degree of nonlinearity into the problem. Thermal stresses are regarded as a non-linear effect as they result
in a mode-mixity that varies with applied load, and non-proportionality between SIFs and
applied load [Chapa-Cabrera & Reimanis, 2002a].

The inclusion of crack-bridging into FE simulations is described in Section 8.1. Its

influence effective toughness is addressed for several types of composite, and then its effect
on crack propagation path is investigated for alumina-epoxy graded composites. The
incorporation of thermal stresses and plastic yielding into the simulations is described in
Section 8.5, then effects of these on crack path and crack growth resistance are investigated
for layered copper-tungsten composite specimens. The results here provide some insight
into FGMs beyond the alumina-epoxy system investigated in the present study.
8.1

Simulation of crack-bridging

The inclusion of bridging into the computational simulations presented a significant

challenge, due to the requirements of self-consistent solution. Continuous bridging traction
distributions were approximated by a series of point loads, so that each force was an
average of the traction p(x) over a small length of crack:
f y = p I ( u y (x )) x

f x = p II (u x (x ), u y (x )) x

(8.1a,b)

The application of point loads is, strictly speaking, aphysical as it leads to stress and strain
singularities. However, in the FE formulation, these singularities are averaged out over the
elements neighbouring the node at which the point force was applied. Distributed surface
loads were not used as they cannot be applied parallel to the surface in ANSYS.
8-1

Bridging traction values were defined iteratively, from crack opening displacement values
at the traction application points. The total extension of the bridges was taken as the total
magnitude of crack-opening displacement. For the nth iteration, at the kth point, this is:
(d) n,k = ((u y ) n -1,k ) 2 + ((u x ) n -1,k ) 2

(8.2)

where (uy)n-1,k and (ux)n-1,k are mode I and II COD values at point k for the (n-1)th iteration
and d is a measure of total deformation of the bridge. The forces, (fy)n,k and (fx)n,k, at point k
in the nth iteration in the y and x directions respectively (ie. mode I and mode II closure
tractions) acting in each direction was then given by:
(f y ) n, k = p I ((u y ) n -1, k )

(f x ) n, k = p II ((u x ) n -1, k , (u y ) n -1, k )

=0

=0

for (d ) n,k -1 d crit

for (d ) n, k -1 > d crit

(8.3)

where pI and pII are functions representing the deformation relations in modes I and II for
the bridging ligaments and dcrit is the failure limit of the bridge. For tractions acting on
deflected or curved cracks, the COD components were transformed from global coordinates (x,y) to local co-ordinates (z,w):
u w,k = u y,k cos k u x,k sin k

u z, k = u y,k sin k + u x, k cos k

(8.4)

where k is the angle of inclination from the global coordinate system of the crack
trajectory at point k. The closure tractions were then determined in the local coordinates:
(f w ) n, k = p I ((u w ) n -1,k )

(f z ) n,k = p II (( u z ) n -1, k , ( u w ) n -1, k )

(8.5)

These were then transformed back into the global coordinate system:
f y,k = f w, k cos k + f z,k sin k

f x, k = f w, k sin k + f z, k cos k

(8.6)

After reaching a stable solution for COD profile, the mechanical energy release rate, G, was
determined then the new load, PN+1, was calculated from the old load, PN:
G
PN +1 = PN C
GN

/2

(8.7)

where 1 provided a degree of damping to reduce overshoot in the search algorithm.

Due to the non-linear response of the bridging zone to applied loading, several iterations
were required to reach a sufficiently accurate solution.

8-2

An initial set of guess values for the crack-opening displacement (COD) profile was used,
obtained from the profile for a non-bridged crack with the same intrinsic toughness. The
final COD profile is equal to this guess profile at the crack tip, then becomes greater
moving away from the crack tip. After defining and meshing the structure, a loading
sequence was used to attain the initial guess values for bridging tractions and applied load:
Step 1: An initial load was applied and iterated, as in Equation 8.7 with = 1, until the
crack-tip energy release rate was equal to the fracture toughness. COD values were
determined and used to calculate bridging tractions for:
Step 2: These tractions were applied negatively so they acted to open rather than close the
crack. Solving in the absence of external applied load yielded a value for the MERR
due to bridging, Gbr.
Step3: Bridging tractions were removed, then a load was applied and iterated until the
crack-tip energy release rate was equal to Gbr.
Step 4: The sum of the loads found in Steps 1 and 3 was used as the guess load, whilst the
COD profile used in Step 2 was used as the guess solution.
Step 5: The model was iterated until the crack profile was stable and the crack-tip energy
release rate was equal to the toughness.

The key drawback to the implementation of bridging in the finite element model was the
large amount of computation time required to reach equilibrium solutions. This was due to
the nested-loop structure of the programs and was unavoidable, although improvements in
convergence rate were achieved by varying parameters within the program.
8.2

Crack-bridging & effective toughness

The model was used to calculate the contribution of crack bridging to effective toughness
under Mode I loading. A set of idealised bridging traction relations, corresponding to
linear-hardening, constant traction and linear-softening behaviours, was simulated for edge
cracks in homogeneous beams under tensile loading and corresponding R-curves were
determined. The tractions predicted R-curves for each case are shown in Figure 8.1. The
shape of the traction relation does not influence the final effective toughness (R-curve
plateau), as would be expected from Equation 2.46. It does have a strong influence,
8-3

however, on the initial gradient, dKR/da. This could have implications for materials with
short cracks, as the initial steepness of the R-curve determines whether a short crack will
arrest.

Relative bridging traction, p(u)/p

max

(a)

Linear softening
Linear hardening

0.8

0.6

0.4

Constant traction

0.2

0
0

0.2

0.4

0.6

0.8

1/2

1.4

max

Linear softening

(b)

Effective Toughness, K [MPa m ]

Relative crack opening displacement, u/u

1.2

Linear hardening
Constant traction

0.8
0

Crack extension, a [mm]

Figure 8.1. Effect of bridging traction relation shape on R-curve. Idealised bridging tractions are
shown in (a) and resulting R-curve predictions in (b).

8-4

The model was then applied to a number of specific materials for comparison with
experimentally measured R-curves and crack-opening profiles. An example is shown in
Figure 8.2, for alumina as studied by Rdel, Kelly and Lawn [1990]. They measured Rcurves and observed crack-opening displacement profiles with in situ electron microscopy
during fracture testing. Their results were simulated reasonably well with the model.
Crack-opening displacement, u [nm]

1600
1400

Expt
FEM

1200
1000
800
600
400
200
0
0

500

1000

1500

Distance from crack-tip, x' [ m]

2000

[J/m ]

60

Effective Toughness, G

Eff

50

40

30
Expt
Model

20

10

0
0

500

1000

1500

2000

Crack Extension, c [m]

Figure 8.2. Application of FE model to alumina, based on experimental results of Rdel et al.
[1990]. Reasonable agreement between simulation and experimental results is observed.

8-5

Homogeneous and graded alumina-aluminium composites were also simulated, as shown in

Figures 8.3 and 8.4 respectively. The homogeneous composite was an aluminium fibre
reinforced, alumina matrix composite specimen with 13% aluminium content, as
investigated by Hoffman et al. [1997]. They provided a bridging traction relation which
was used for the simulation. The graded specimen was the n = 1 case presented in Figure
2.37 on page 2-87, investigated by Chung et al. [2000]. Good agreement between
simulation and experimental results is observed in each case.

1/2

[MPa m ]

Experiment
FE Simulation

Fracture Toughness, K

eff

Crack extension, a [mm]

Figure 8.3. Application of FE model to alumina-aluminium, based on experimental results of

Hoffman et al. [1997].

8-6

[MPa

Fracture Toughness, K

eff
c

1/2

5
4

Experiment
FE Simulation

3
2

Crack Extension, a [mm]

Figure 8.4. Simulation of R-curve for crack propagation parallel to the gradient in graded aluminaaluminium interpenetrating-network composite, based on experimental results of Chung et al.
[1990].

8.3

Crack-bridging & mode-mixity

As discussed in Section 2.3.5 (p. 2-37) and Section 2.4.3 (p. 2-52), bridging of a crack
under mixed-mode loading can act differently in each mode, and thereby alter the modemixity at the crack-tip. In this section, a simple analytical model is proposed and results are
presented to illustrate the possible effects of bridging on crack-tip mode-mixity in
alumina/epoxy composites.

The crack tip was assumed as loaded by far-field applied SIFs, KI and KII, as in Figure 8.5.
These SIFs would depend upon applied load, specimen and crack geometry and material
gradient. The crack propagates through the alumina phase more easily, hence ligaments of
epoxy remain intact behind the crack tip, providing closure tractions and partially shielding
the crack tip from applied stresses. As crack opening occurs in both mode I and mode II,
bridges are treated as deforming in both tension and shear, thus providing closure tractions
in both modes. Due to the different natures of the tractions in each mode, the effective
mode-mixity at the crack tip can be altered by the bridging tractions from to eff. After
8-7

developing the model for the bridging traction relations for each mode, the effect on modemixity is calculated analytically and using FEA.

8.3.1 Bridging tractions

Several approximations were made, regarding the bridges, as shown in Figure 8.5:
1. Ligaments are rectangular with width, w, and unit depth, and are distributed uniformly
along the crack direction, separated by an interligament distance, dB.
2. Delamination between epoxy and alumina occurs along the ligament for a distance, L,
which is assumed to be proportional to bridge width, ie. L = w.
3. Bridges extend linear-elastically up to point of failure, which occurs at extension v0,
due to deformation in tension and shear.
4. Having treated bridges discretely to determine the closure tractions, they are then
smeared resulting in a continuous traction distribution behind the crack tip, as shown
in Figure 8.6.

K II

KI

Bridge

L
px

py

2v(x)

Crack edge
2u(x)
Delamination

KI

K II

Figure 8.5. Crack tip showing single bridging ligament and associated closure tractions.

Consequently, the traction distributions are given by:

p y ( v) =

Ef w v Ef v
=
,
L dB
dB

p x (u) =

Ef w u
Ef w
=
tan eff
dB v
2(1 + f ) d B

(8.8a,b)
8-8

where v and u are crack-opening displacements in Modes I and II, Ef and f are Youngs
modulus and Poissons ratio of the bridging phase. Interligament distance is related to
bridge width and bridging phase volume fraction VB by dB = w/VB.

KI

KII
Bridging Zone

2v 0

Crack edge

Figure 8.6. Schematic of bridging zone with continuous traction distribution.

Maximum values of crack-opening displacement (COD) in each mode are given by:
v m = v 0 cos eff

u m = v 0 sin eff

(8.9a,b)

where vo is the maximum total extension of the bridges before failure.

8.3.2 Analytical solution

The additional contributions to crack-growth resistance, due to bridging tractions in modes
I and II respectively, are:

E f VB v 0
cos 2 eff
w

(8.10a)

E f VB v 0 sin 2 eff
R II () = 2 p x (u)du =
(1 + f ) cos eff
0

(8.10b)

vm

R I () = 2 p y (v)dv =
0

um

To overcome this additional resistance to crack propagation, the applied SIFs must be
increased, as compared with the case of no bridging.

K I + K II = (R 0 + R I + R II )E C
2

(8.11)

8-9

where R0 and EC are the intrinsic fracture toughness and local Youngs modulus of the
composite, and KI and KII are related to the total applied SIF magnitude, KA, by:

K I = K A cos

K II = K A sin

(8.12)

Hence KA is given by:

K A = (R 0 + R I + R II )E C

(8.13)

For crack propagation, the effective crack-tip SIF may be related to effective mechanical
energy release rate, G, by:
( K I K SI ) 2 = E C (G I R I )

( K II K SII ) 2 = E C (G II R II )

(8.14)

where KSI and KSII are reductions in SIF due to shielding.

GI and GII are components of applied mechanical energy release rate due to loading in
modes I and II respectively, given by:

GI =

KI

G II =

EC

K II

EC

(8.15)

Hence, for i = I, II:

K 2
( K i K Si ) 2 = E C i R i

EC

Giving:
K i K Si = K i R i E C
2

(8.16)

From this, the effective mode-mixity may be calculated:

tan eff

K II K SII
=
=
K I K SI

K II R II E C
2

K I R IEC
2

(8.17)

Combining Equations 8.10, 8.12, 8.13 and 8.17 yields an expression relating , eff and
known constants. Thus eff is determined as a function of for particular values of
constants w, , EC, R0 and VB.

From the MTS criterion, as in Equation 2.37 on page 2.33, the crack deflection is predicted
for both a non-bridged crack:

8-10

 2 2 1 + 8 tan 2
= 2 tan 1
,
8 tan

(8.18a)

and a bridged crack:

2 2 1 + 8 tan 2
eff
= 2 tan
8 tan eff

(8.18b)

The effect of bridging on the crack deflection angle is expressed as:

= B

(8.19)

Variation in EC does not affect the effective mode mixity or kink angle, whilst variation in
R0 has only a minor effect. Consequently, inaccuracy in these values is not a major source
of error.

The model is demonstrated by application to alumina-epoxy stepwise graded composites. A

specimen geometry comprised of a crack situated in an intermediate composite region
between epoxy and alumina was assumed. The composition of the central composite
region was taken as 30vol% epoxy and 70vol% alumina, ie. VB = 0.3. Pertinent material
properties are, for alumina: E = 390 GPa, = 0.27, R0 = 40 J/m2; for epoxy: E = 3.4 GPa,
= 0.35; and for composite region: E = 100 GPa, = 0.3, R0 = 25 J/m2. The bridges were
assumed to have a width of w = 25 m, hence the distance between bridges was calculated
as 83 m. Failure extension and delamination distance factor were assumed to be v0 = 3
m and = 2. By varying each of the parameters Ef, VB, w, v0 and in turn, whilst the
others were maintained at their nominal values, the effect of each parameter on the
effective mode mixity and kink angle was ascertained.
Figure 8.7 shows the dependence of effective mode mixity, eff, and predicted kink angle,
, on applied mode mixity, , for nominal values of bridging and local material constants.
Figures 8.8 to 8.12 illustrate the dependence of kinking angle on ligament stiffness,
ligament width, ligament failure extension, bridging phase volume fraction and
delamination distance. Figure 8.8 demonstrates that increasing ligament width results in an
increased bridging effect. This is attributable to an increase in delamination distance, so
8-11

mode I closure tractions are reduced relative to mode II tractions resulting in a decreased
mode-mixity.

60

60

without bridging
50

50

Eff

Effective Mode-Mixity,

eff

40

without bridging

with bridging 40

30

30

20

20

10

10

0
0

10

20

30

eff

Crack Deflection Angle, [deg]

[deg]

with bridging

40

50

Applied Mode-Mixity,

[deg]

App

0
60

Figure 8.7. Effect of bridging on effective mode mixity and kink angle.

Decrease in Kink Angle, (deg)

35
w = 100 m
30

w = 50 m
w = 30 m

25
20

w = 10 m

15
10

w = 3 m

5
0

10

20

30

40

50

60

Mode Mixity, (deg)

Figure 8.8. Effect of varying bridging ligament width on decrease in kink angle.

As shown in Figure 8.9, increasing bridging ligament stiffness leads to an increase in effect
of bridging on crack deflection, as expected due to increased closure tractions. It is
8-12

expected that increasing bridging phase volume fraction causes an increase in bridging
effect on crack deflection, and this is verified in Figure 8.10.
30

E = 34 GPa

Kink Angle Decrease, (deg)

E = 10.8 GPa
f

25

E = 3.4 GPa
f

E = 1 GPa
f

20
E = 0.34 GPa
f

15
10
5
0

10

20

30

40

50

Mode Mixity, (deg)

60

Figure 8.9. Effect of varying bridging ligament stiffness on decrease in kink angle.

Decrease in Kink Angle, (deg)

30
V = 0.40
b

25

V = 0.25
b

V = 0.10
b

20

V = 0.03

15

10

V = 0.01
b

5
0

10

20

30

40

50

60

Mode Mixity, (deg)

Figure 8.10. Effect of varying bridging phase volume fraction on decrease in kink angle.

In Figure 8.11, the effect of increasing failure extension of the bridges is a decrease in the
effect on crack deflection. This may be explained in terms of the work done on extending
bridges in each mode, which increases more significantly in mode I for an increased failure
8-13

strain. Figure 8.12 shows that increased delamination distance leads to an increase in
bridging effect. This is attributable to an increased dominance of mode II closure tractions
over mode I tractions as tensile strain is decreased while shear strain remains constant.

Decrease in Kink Angle, (deg)

30
v = 1 m

v = 2 m

v = 3 m

25

v = 5 m
0

20

v = 10 m

15

10
5
0

10

20

30

40

60

50

Mode Mixity, (deg)

Figure 8.11. Effect of varying bridging ligament critical extension on decrease in kink angle.

Decrease in Kink Angle, (deg)

40

= 10

35

=5

30
25

=2

20
15

= 0.5

10
5
0

10

20

30

40

50

60

Mode Mixity, (deg)

Figure 8.12. Effect of varying delamination distance factor on decrease in kink angle.

8-14

8.3.3 Computational solution

The effect on mode-mixity was examined using finite element analysis also. The traction
relations were calculated with the default values of Ef, VB, w, v0 and as above. This
resulted in bridging tractions acting in Mode II that were relatively higher than those acting
in Mode I.

This led to a disproportionate suppression of mode II crack opening

displacements, shown in Figure 8.13, and a reduction in crack-tip mode-mixity, which is

illustrated in Figure 8.14.

Crack Face Displacements u, v [m]

With Bridging
Without Bridging
1.5

Mode I - Crack-Opening
1

Mode II - Crack-Sliding

0.5

0
0

0.5

1.5

Distance from Crack-tip, x [mm]

Figure 8.13. Crack-opening displacement profiles for a bridged and unbridged cracks under mixedmode loading. The high suppression of the mode II opening may be observed.

This change in mode-mixity could have implications for crack-path development in

materials with crack-bridging under mixed-mode loading. The result shown, however, is
highly dependent on the assumed tractions in each mode. As these are generally not wellunderstood, it is difficult to extrapolate the findings, obtained here for a hypothetical set of
traction relations, to real materials. In summary, this model shows how bridging can have
an effect on crack-tip mode-mixity, and illustrates the effects of various bridging
parameters on the change in crack-tip stresses. This issue is addressed further in the
following section.
8-15

Figure 8.14. Principal stress field (1) around the crack tip, (a) without and (b) with bridging
tractions acting behind the crack tip.

8.4

Crack-bridging & propagation path

Considerations regarding the effects of bridging on crack-tip stresses, and subsequently on

crack propagation, are extended in this section to a variety of possible mixed-mode
bridging behaviours.

A straight bridged crack under mixed-mode loading, as in the previous section, and
illustrated here in Figure 8.15(a), is considered initially. Equation 8.17 gives the effective
mode-mixity, which is understood to dictate initial deflection angle, is given by Equation
8-16

(8.17). It is now assumed that the shielding components of the mode I and II SIFs are given
by functions of the applied SIFs, and are smaller than the applied SIF in each mode, ie.
KSI = fI(KI)

|KSI| < |KI|

KSII = fII(KII)

|KSII| < |KII|

(8.20)

The crack is assumed to extend in the direction dictated by the crack-tip mode-mixity, as
shown schematically in Figure 8.15(b). The bridging-zone then extends along two sections
of the crack. The crack, having changed direction relative to the external loading, now
experiences new applied SIFs, KI and KII. It should be noted that the notation used here
for SIFs differs from that used in Section 5.4.

(a)

KI

KII

(b)

KI
KII

Bridges

Figure 8.15: Crack with bridging zone under mixed-mode loading (a) before and (b) after
deflection. After deflection, bridging occurs along both the deflected and undeflected sections.

Depending on the relative magnitudes of the shielding components in Modes I and II, the
possible outcomes, in terms of crack-tip mode-mixity, are as follows:
1. More shielding in Mode II, so eff < appl.
This would lead to under-deflection, as compared with the case of no bridging. After
deflection, as in Figure 8.13(b), the crack will still experience mixed-mode loading, ie. KII
will be positive, as the crack is still inclined relative to the external loading direction. The
new crack-tip mode-mixity will be smaller than the new applied mode-mixity, but will be
non-zero so further deflection will occur. Eventually the crack will approach the direction it
would have propagated in the absence of bridging, ie. the far-field propagation direction.
2. Equal shielding in Modes I and II, so eff = appl.
This would lead to the same deflection as in the case of no bridging. The new mode II SIF,
KII, will then be zero, so that further deflection does not occur. It is possible there could be
8-17

interactions between the shielding effects acting in the deflected section and those acting in
the initial straight section. This could lead to mixed-mode loading, but this would be a
second-order effect that would decrease as the crack propagates further.
3. More shielding in Mode I, so eff > appl.
This would lead to over-deflection, as compared with the case of no bridging. The crack
will now be inclined from the applied loading such that KII is negative. This would tend to
cause deflection back towards the far-field propagation direction. Along the length of the
bridging zone, sections will experience different mode-mixities. Their relative response to
this will depend on the details of the bridging traction relations and crack-opening profiles
in each mode. Assuming the mixed-mode loading in the other direction is also amplified,
the crack will be over-deflected back. The eventual behaviour could be one of two types:
I. The mode-mixity amplification is very severe, ie. mode I shielding extremely high and
mode II shielding extremely low. The crack direction would then be inherently unstable,
such that each kink is of greater magnitude than the preceding one, in the opposite
direction to the previous one. This is unlikely, for several reasons:
(1) Cracks do not proceed in discrete kinks, except possibly at the microstructural
scale. At this scale, crack path would be determined predominantly by
microstructure rather than external mechanical loading.
(2) High mode-mixities are liable to lead to alternate crack-tip propagation
mechanisms ie. mode II crack propagation.
(3) The development of the bridging zone behind the crack tip, over the length of
several kinked sections, will lead to a distribution of bridging zone orientations.
This will probably lead to an averaging out of bridging tractions in each mode,
making mode-mixity amplification and deflection overcompensation less likely.
(4) Even if extreme amplification of mode-mixity continues, the most that a crack
will kink under extreme mixed-mode conditions, ie. eff = 90, is 73.
II. For less severe amplification of mode-mixity, therefore less overcompensation of
deflection, such that kinks are of diminishing magnitude, the crack direction will
oscillate then stabilise. For the reasons above, especially (1) and (3), this stabilisation is
likely to occur over a relatively small distance of crack propagation and the final
propagation direction is likely to be equal to the far-field propagation direction.
8-18

Attention is turned now to the case of an initially straight crack under mixed-mode loading
without a bridging zone, as in Figure 8.16(a). After initial deflection, it will extend along
the far-field propagation direction, and neither the bridges nor the crack tip will
experience mixed-mode loading, as in Figure 8.16(b). Assuming that as it propagates
further, it begins to experience a change in stress field due to specimen geometry or
material gradient, mixed-mode loading will occur at the crack tip, as in Figure 8.16(c). This
situation is more likely to occur in graded materials, such as those simulated in Chapter 6,
than the situation described in Figure 8.15.

(a) KI

KII

(b)

KII= 0

(c)
KI

KI
KII

Bridges
Figure 8.16: (a) Straight crack under mixed-mode loading. This deflects (b) and develops a
bridging zone under mode I loading, then propagates further into a region of mixed-mode loading.

However, this situation is fairly similar, in terms of the interaction between bridges and
crack deflection, to that considered previously in Figure 8.15. The possible outcomes will
be similar also. If the bridges cause a suppression of Mode II then the crack will deflect
more slowly under mixed-mode loading, than it would without bridges. It is likely though
to eventually follow the same final propagation direction. If the bridges cause a suppression
of Mode I, ie. an amplification of mode-mixity, then the deflection is likely to be
overcompensatory and less stable, though final propagation direction will be fairly similar
to far-field. This suggests that, even if bridging has an influence on crack-tip mode-mixity,
the effect on total crack propagation should be minor.

The current analysis has not yet considered the possibility of instability in crack paths. It
was shown in Chapters 6 and 7 that interactions between crack shape, material property
distribution and specimen geometry could lead to divergence of crack trajectories. A small
difference between two crack paths can lead to greater deviation as the cracks propagate,
8-19

due to the differences in stress field that each crack tip will experience at different
positions. This provides a possible mechanism for systematic deviation of bridged cracks
away from the trajectories that would be followed in the absence of bridging.

This was further investigated through finite element simulations of crack propagation,
incorporating selective suppression of SIF in one or both modes. Cracks in stepped linearprofile gradients (Ns = 5, n = 1, w = 15mm, ao = 10mm, Erat = 100) were simulated in the
same manner as the cracks in graded specimens presented in Section 6.2 on pages 6.13 to
6.16, with the exception of the deflection criterion. Rather than calculating propagation
direction with a test-kink and the local symmetry criterion, the MTS criterion was used
along with modified SIF values. These were given by:
K eff
= K I (1 I l bz )
I

K eff
II = K II (1 II l bz )

(8.21)

where I and II are the factors by which each mode SIF was suppressed and lbz is the
bridging zone length, which was equal to crack extension up to a maximum length, lbzmax.
Values for I and II and lbzmax for each case considered are shown in Table 8.1. While this
approach is admittedly somewhat artificial, it provided a simple method for examining how
crack-bridging could influence crack path.
Table 8.1: Cases considered for simulations of crack propagation with crack-bridging effects.

Case
I
II
max
lbz [mm]

1
0.1
0.7
1

2
0.1
0.7
5

3
0
0
n/a

4
0.4
0.4
5

5
0.7
0.1
1

6
0.7
0.1
5

The results are shown in Figure 8.17. Proportionally equal suppression, as in case 4,
resulted in an essentially identical crack path to case 3 where bridging was not included.
This agrees with the argument for point 2 above (Page 8.17). After some propagation, both
these cracks appeared to become slightly unstable resulting in wandering, rather than
smooth, crack paths. This was probably due to T-stresses as discussed in Section 6.2.
Higher suppression of Mode II tractions, as in Cases 1 and 2, led to slightly less deflection
and a more stable crack path. Higher suppression of Mode I tractions tended to result in

8-20

more overall deflection and more path instability. Again, these findings are consistent with
the preceding discussion.
Mode II Suppression, l

y [mm]

20

BZ

= 1 mm

Mode II Suppression, l = 5 mm
BZ
No bridging
Equal suppression, l = 5 mm
BZ

18

Mode I Suppression, l
Mode I Suppression, l

BZ
BZ

= 1 mm
= 5 mm

16

14

12

10

10

x [mm]

Figure 8.17. Effect of bridging on crack propagation path. Suppression of mode II SIF tends to
reduce crack deflection, whilst suppression of mode I SIF tends to increase crack deflection.

The deviation between the different bridging cases is understood to be due to the build-up
effect, as discussed above, whereby slight differences in crack path early in propagation led
to larger deviations later due to the interactions between crack shape and stress fields.
Further simulations with smaller increments verified that these trends were fairly
insensitive to crack extension increment size. This provides strong evidence that the effect
is real, ie. that bridging can cause divergence of crack trajectory, rather than being an
artefact of the simulation process.

The key outcome from this analysis is that, while the presence of bridging may have an
effect on propagation path, the overall propagation direction is dictated by the elastic
gradient, and bridging will be a second-order effect.

8-21

8.5

Incorporating thermal & plastic mismatch

To further investigate the applicability of the modelling scheme, it was applied to

simulation of layered copper-tungsten specimens with cracks parallel to, and very close to,
the interface with the more ductile region. This configuration was chosen as it is also being
investigated experimentally by Keith Rozenburg and Ivar Reimanis at Colorado School of
Mines [Rozenburg et al., 2005], and follows from earlier work carried out there by Jesus
Chapa-Cabrera [2000, 2001, 2002b], which was reviewed in Section 2.8.3 (p. 2-93).
Pertinent literature on cracks near interfaces, and plasticity effects in particular, was
reviewed in Section 2.5 (pp. 2-63 to 2-69). This configuration involves mismatch in
thermal and plastic, as well as elastic, properties. As discussed in Chapter 2 and highlighted
in Section 3.3 (p. 3-4), mismatch in these properties will occur in many FGM systems, but
does not influence alumina-epoxy composites.

The specimen configuration investigated, shown in Figure 8.18, consists of a symmetrical

layered composite. It is understood that the crack-tip stresses will be influenced by the
mismatch of elastic, plastic and thermal properties. Nonlinear elastic-plastic stress-strain
relations were defined and a range of thermal residual stress conditions and loading cases
were simulated.
P/2

Sin = 10mm

P/2

w 0-2000 m
Material 1
More
Brittle

Material 1
More
Brittle

h = 8mm
ao

Sout = 20mm

Material 2
More Ductile

x < 1mm

Figure 8.18. Copper-tungsten layered composite specimen configuration, showing dimensions and
material compositional distribution.

8-22

Plastic deformation behaviour was assumed to be isotropic and was defined in terms of
multi-linear stress-strain curves. This was achieved within the formulation for nonlinear
material behaviour defined in ANSYS. The nonlinear material properties used were those
obtained experimentally by compression testing of homogeneous composite specimens
[Chapa-Cabrera & Reimanis, 2003], which are shown in Figure 8.19. It was assumed that
bulk flow behaviour is similar in tension and compression.

Composition

Youngs
Modulus, E
[GPa]

Poisson
Ratio,

Thermal
Expansion
[K-1]

Yield Stress
(0.2% offset)
o [MPa]

Fracture
toughness, Kc
[MPam]

SSY length
2.5(Kc/o)2
[m]

Table 8.2: Measured elastic, thermal and fracture properties for copper-tungsten composites.

60%W 40%Cu
40%W 60%Cu
20%W 80%Cu
100%Cu

232
208
163
124

0.286
0.314
0.330
0.328

8.9
11.2
13.9
17.5

688
567
259
49

16
17*
18.7
-

1.35
2.38
13.5
*estimated.

1000

20%Cu/80%W
40%Cu/60%W

Stress, [MPa]

800

60%Cu/40%W

600

80%Cu/20%W
400

200

100% Cu
0

0.05

0.1

0.15

0.2

Strain,
Figure 8.19: Non-linear deformation behaviour of Cu-W composites, for a range of compositions,
obtained from experimental results of Chapa-Cabrera [2003]. Compression and tension behaviour
assumed equivalent. (cf. Figure 2.8 on page 2-23)

8-23

Two different cases were considered. For Case 1, the moderate plastic mismatch case, the
standard configuration consisted of a crack of length ao = 3.2 mm situated a distance xo =
400 m from the interface, as shown in Figure 8.16. Material 1 was generally assigned the
properties of the 60%W composite, as in Table 8.2, while Material 2 was generally
assigned the properties of the 20%W composite. To account for the fact that, in a layered
composite, there is no defined interface, a single intermediate layer of elements
approximately 50 um wide situated on the interface was assigned the properties of the
40%W composite. In Case 2, the high plastic mismatch case, the ductile layer was reduced
in thickness, to 100 m, and its ductility was increased by setting the properties of Material
2 to those of pure copper. There was no intermediate layer and the straightness of the
interface was ensured, by increasing the mesh density in that region. Yielding was not
allowed in Material 1. Macroscopically homogeneous notched beams, comprised of a
single composition throughout, were also simulated to enable examination of material
response without the added complication of the interface and the attendant elastic, plastic
and thermal property mismatches. In some simulations of layered specimens, one or more
of these mismatches was artificially switched off or amplified to enable comparison of the
relative influences of each, and to examine interactions between the different effects.

Many ductile/brittle interface systems contain residual stresses as a result of differential

contraction on cooling from elevated processing temperatures. Accordingly, residual
stresses were considered in the present simulations. As thermal contraction can lead to
plastic deformation, the FE model was solved after cooling, then loads were applied and the
model was re-solved. Plastic strain fields before loading were similar to those reported by
Cao et al. [1988]. The effect of notching on redistribution of stresses and strains was also
examined and shown to have a negligible influence on crack-tip stresses after loading.

Quarter-point nodes were used at the crack tip for elastic fracture analysis and regular
midside-node elements were used for elastic-plastic analysis. Youngs modulus and thermal
expansion coefficient values were assigned to elements, based on their positions. In
simulations involving thermal residual stresses and plasticity, definition of property
8-24

variation via nodal temperatures was not feasible. In this case, several different materials
with different properties were defined and the material number for each element was
defined, based on the position of the element (NSEL,,LOC,X), using MPCHG. While freemeshing results in a somewhat irregular mesh-pattern, this is necessary for meshing around
arbitrary crack-shapes and is understood not to compromise the accuracy of results, as the
element aspect ratios were all within the range required by the shape functions used in
ANSYS. Due to the irregularity of the mesh pattern resulting from the free-meshing, this
interface was not precisely straight, however this should not have a significant influence on
results. Simulations with increased mesh density in this region, which resulted in much
straighter interfaces, yielded essentially identical results.

Mode I and II SIFs, KI and KII, were calculated by displacement correlation for nodes near
the crack tip. From these, the total stress intensity factor magnitude, |K|, was calculated as
|K| = (KI2+KII2). The validity of the stress intensity factor as a descriptor of crack-tip
stresses is demonstrated and discussed later in this section. J-integrals were also calculated,
by integration along a path outside the region of plastic yielding. Values of |K| obtained
from J-integrals, |K| = (JE), and from load-line displacements enabled verification of SIF
values.

Prediction of critical load for crack extension and of deflection angle, which is dependent
on failure load [Chapa-Cabrera & Reimanis, 2002a,b], required several iterations to
determine the load which results in a critical SIF at the crack tip. As plasticity results in a
non-linear relationship between applied load and crack-tip SIF, a non-linear search
algorithm was required. The one used was similar to that in Equation 8.7:
Kn

Pn +1 = Pn

K
c

(8.22)

where 1 introduced a degree of damping to prevent possible overshoot. This enabled

the correct load, to within 3%, to be obtained within 5 iterations.

Several deflection criteria were considered: the maximum tangential stress (MTS), local
symmetry (KII = 0) and maximum energy release rate (MERR) criteria. The mode-mixity,
8-25

, was calculated from crack-opening displacements as in [Chapa-Cabrera & Reimanis,

2002] and [Sutton et al., 2000]. For the MTS criterion, the deflection angle, , was
calculated directly from . For the MERR and KII = 0 criteria, a test-kink was extended
from the crack tip at several different angles, to enable determination of the optimum angle
for crack extension. Mechanical energy release rate values were obtained from stress
intensity factor magnitude, |K|, and Youngs modulus, E, according to G = |K|2/E. KII
values were obtained from correlation of nodal displacements near the crack tip. The search
algorithm detailed in Section 5.5 (p. 5.21-5.22) reduced the number of test-kink angles
required. Under small-scale yielding conditions, crack-tip stresses, hence mode-mixity,
should still be dictated by the stress intensity factors, so that deflection criteria based on
SIFs are applicable [Shih, 1974, Suresh & Shih, 1986]. As discussed in Section 5.4 (p.
5.29), process zone effects and T-stresses can have an influence on crack-tip mode-mixity,
hence influencing crack-path or stability.

After crack extension, the residual plastic strain field, which results from the stress
concentration that follows the crack-tip trajectory, may influence the crack-tip stresses.
Accordingly, this should be taken into account during simulations. In the present study, this
was achieved with the use of coupling and decoupling of crack-face nodes, to obtain
equivalent plastic strain fields after remeshing. After crack extension and associated
remeshing, nodes along the increment of crack extension were coupled, then loads were
applied and the model was solved to obtain the plastic strain field before extension. The
coupled nodes were then decoupled and the model solved again to obtain the SIFs after
extension. This method can become time-consuming after several increments of crackextension, but it is fairly straightforward. The alternative would be to map residual
displacements onto the new mesh after remeshing around the new crack path. The approach
used enables the effect of prior plastic strains on crack propagation to be examined without
knowing the crack path previously. Finding the critical load in this case can become
computationally demanding, as each iteration requires the application of prior loadings and
subsequent nodal releases, to obtain the prior plastic strain field.

8-26

To use the stress intensity factor as a fracture parameter for nonlinear materials, it must be
demonstrated that it uniquely and accurately describes the stress field around the crack tip,
outside of the plastic deformation region. Small scale yielding conditions must be
demonstrated, along with the accuracy of calculated SIF values. To demonstrate this for the
elastic-plastic composite materials under investigation, homogeneous notched beams under
four-point bending were analysed under elastic and elastic-plastic deformation conditions.
The stress intensity factors were calculated using the displacement correlation technique,
and were compared with values calculated from the J-integral. These results are shown in
Table 8.3 for elastic and elastic-plastic analysis of three different compositions, illustrating
the implications of a range of yielding behaviours.

The values of K calculated from displacement correlation and from the J-integral show
good agreement for the elastic-only cases. When plasticity is incorporated, the value from
J-integral does not change significantly, for the first two compositions, whilst the values
from crack-face nodal displacement correlation are quite different. This is due to the higher
crack-opening displacements, which result from plasticity near the crack tip, and is an
artefact of the calculation method. For the third composition, the yielding is so high that the
J-integral values deviate also. These differences in yielding behaviour are concordant with
the values of small-scale yielding length in Table 8.2. These indicate that the first two
compositions fulfil the criteria for small-scale yielding:
a, h, b 2.5(Kc/o)2

(8.23)

where a, h and b are the crack length, specimen height and thickness respectively. The third
composition does not. The presence of plastic deformation at the crack tip changes the
relationship between stress intensity factor and crack-opening displacements so that the
expression used for linear elastic materials, as given in Equation 5.6, is no longer valid. The
relations between stress intensity factors and crack-opening displacements were established
from simulations of homogeneous specimens, as in Table 8.3. This is shown in Figure 8.20,
which demonstrates that there is a clear relationship between SIF values calculated from the
J-integral and those from crack-opening displacements. This validates the calculation of
SIFs from crack-opening displacements for layered specimens in the presence of yielding.

8-27

60%W 40%Cu

1635

40%W 60%Cu

1791

20%W 80%Cu

1943

No Yielding
Yielding allowed
No Yielding
Yielding allowed
No Yielding
Yielding allowed

16.0
21.9
17.5
26.6
19.0
82.5

1003
1022
1483
1526
2092
3936

K from Jintegral
[MPa m1/2]

Load,
P [N]

J-Integral
[J/m2]

Composition

K from
DCT [MPa
m1/2]

Table 8.3: Calculated stress intensity factor and J-integral values for a crack in macroscopically
homogeneous four-point bend specimens of different compositions under critical applied loading.

16.2
16.3
17.9
18.1
19.5
26.9

SIF from displacements, K

DCT

1/2

[MPa m ]

80
70

a/h = 0.4, h/S = 0.4

a/h = 0.6, h/S = 0.5

60
50
40
30
20
10
0

10

15

20

25

30

1/2

SIF from J-integral, K [MPa m ]

J

Figure 8.20: Relationship between stress intensity factors and crack-opening displacements. This
was obtained by comparing SIF values from J-integral and displacement correlation for cracks in
homogeneous specimens, then used to calculate SIFs from nodal displacements for cracks in
layered specimens. These results indicate the relationship is independent of specimen geometry.

8-28

8.6

Results: Effect of thermal stresses

The effects of thermal residual stress on crack deflection are illustrated in Figures 8.21 and
8.22. Residual stresses cause significant deflection away from the interface, with deflection
angles becoming more negative as the interface is approached. For the T = 300K case in
Figure 8.21, the effects of plasticity are dominated by those of residual stress, so that
deflection angles are all negative. Thermal mismatch has a less significant effect on critical
load than on deflection angle. This may be understood in terms of the contribution to SIFs.
The thermal component of the SIF is predominantly Mode II. Accordingly, it will rotate the
stress field, and change the mode-mixity significantly, but will not influence the magnitude
of the total SIF.
80

20

70

Deflection Angle, M [deg]

0
60

K/P (T = 300)

(T = 100)

-20

50
40

-40

K/P
T = 100

(T = 300)

30
20

-60
10

K/P (T = 300)
-80
0

500

1000

1500

Normalised SIF Magnitude, |K|/P [m -3/2 ]

(T = 0)

0
2000

Applied Load, P [N]

Figure 8.21: Effect of residual stress on normalised stress intensity factor magnitude and crack-tip
mode-mixity for a range of applied loads, for a straight crack near an interface in a layered Cu-W
composite specimen under 4PB loading. Crack length was 3.2 mm, and crack position was 150
microns from the interface. Plastic yielding was allowed in both materials 1 and 2.

Predicted propagation paths for various combinations of thermal mismatch, plasticity and
initial notch position are shown in Figure 8.22. Residual stress clearly has a major
determining effect on crack path, causing deflection away from the interface. The influence
8-29

of plasticity may be observed by comparing the crack path predictions obtained from
elastic-plastic analysis with those obtained from elastic-only analysis. It is apparent that
plasticity tends to mitigate the influence of residual stresses to an extent, however this
effect is not large. Cracks situated nearer to the interface tend to experience more
deflection, in agreement with the results in Figure 8.21.

Distance from base of specimen, y [mm]

5.5

Interface
4.5

4
T = 0K Pl
T = 0K El
T = 100K Pl
T = 100K El
T = 300K Pl
T = 100K El

3.5

Notch
Tip
-0.5

0.5

Distance from interface, x [mm]

Figure 8.22. Effect of residual stress on crack propagation paths for cracks situated near the
interface in Cu-W composite structures. Initial crack length was 3.2 mm and initial crack position
was 200 microns from interface. Paths were predicted using FEM with either elastic or elasticplastic (moderate mismatch) deformation analysis, and a range of assumed cooling temperature
values, T, was considered.

These findings generally agree with those of Cao et al. [1988], who used FEA to show that
for an edge-crack in the brittle material, near the ductile layer, residual stresses will lead to
mixed-mode loading at the crack tip, which increases closer to the interface. They
presented experimental sandwich specimen results showing that a crack near the interface
with the ductile layer would deflect away from it due to thermal stresses. The predicted
paths in the case of no thermal stresses are similar to those observed by McNaney et al.
8-30

[1994] for alumina/aluminium sandwich specimens. It may be concluded from the present
findings that residual stress, through its influence on propagation path, has a profound
effect on structural reliability of interfaces, which is further affected by plastic mismatch.
8.7

Results: Effects of plasticity

The interaction of property mismatch effects was observed, exemplified by the evolution of
plastic strain shown in Figure 8.23. This shows plastic strain fields near the crack, which is
situated 200m from the interface, for the moderate plastic mismatch case. For low applied
load, the residual stress field dominates, resulting in large negative mode-mixity. Increasing
the applied load resulted in plasticity being induced in the neighbouring ductile region, as
well as more plasticity and less mode-mixity at the crack tip. At higher loads still, plasticity
in the copper-rich region increased significantly, causing the crack-tip stress field to rotate
around further. It is observed in these images that the interface region is not precisely
straight. This is a result of the free-meshing approach and the method used to define
material properties.

To illustrate the relative influences of plasticity near the crack and further away in the more
ductile layer, results were compared from several analyses in which plastic yielding was
enabled in different sections of the structure. The cases considered were: (1) yielding
allowed throughout, (2) yielding allowed in material 2 only, (3) in material 1 only and (4)
no yielding allowed. The validity of the assumption of small-scale yielding for a layered
structure is somewhat less clear than for a homogeneous material. This is particularly so
when small-scale yielding applies in the immediate vicinity of the crack tip but larger-scale
yielding may apply in the copper rich region of the specimen further away from the crack
tip. It is suggested that this plasticity may be separated from the crack-tip yielding, and may
be treated as far field, in that it influences the stress intensity factors at the crack tip,
without violating their dominance of the crack-tip stress fields. In this respect, the plastic
deformation in the copper-rich region will contribute, in combination with the elastic and
thermal mismatch, to the mixed-mode loading of the crack. The plasticity at the crack tip
experiences this mixed-mode loading within the confines of the K-dominant zone, though it
may modify the response of the crack to the mixed-mode loading. The values calculated for
8-31

normalised SIF magnitude, |K|/P, where K is calculated using the displacement correlation
technique (DCT), and mode-mixity, , across a range of loads are shown in Figure 8.24.
(a)

|K| = 4.6
= -22.8

(c)

(b)

|K| = 10.4
= -11.8

|K| = 24.0
= -2.9

Figure 8.23: Plastic strain fields at (a) 500N, (b) 1000N and (c) 1700N for a crack situated 200 um
from the interface.

Figure 8.24(a) demonstrates that the presence of yielding can influence crack-tip stresses.
Plasticity in Material 1, which occurs around the crack tip, appears to have a negligible
influence on SIF magnitude. Plasticity in Material 2, further away from the crack, has a
significant effect, however, which becomes more apparent for higher loads. The increase in
SIF magnitude results from a modification of the stress field due to the increase in
compliance of the Material 2 region with plastic deformation. This could have important
implications for structural integrity, as the additional stress concentration decreases the
8-32

critical load for a crack near an interface. The increase is similar to that predicted by
Sugimura et al. for cracks oriented normal to [1995a], or inclined to, an interface with a
more ductile region.
(a)

10

Yielding allowed in:

Entire specimen
Material 2 only
Material 1 only
No yielding

1000

1500

2000

2500

Applied Load, P [N]

10

(b)
5

Yielding allowed in:

Entire specimen
Material 2 only
Material 1 only
No yielding

-5

-1 0

-1 5

1000

1500

2000

2500

Applied Load, P [N]

Figure 8.24: Comparison of (a) normalised SIF magnitude, |K|/P, and (b) mode-mixity, , values
obtained across a range of loads for different elastic-plastic cases: (1) plasticity throughout, (2)
plasticity only in material 2, (3) plasticity only in material 1 and (4) elastic-only throughout.

8-33

Figure 8.24(b) shows that the effects on crack-tip mode-mixity are somewhat different.
Plasticity in Material 2 increases the mode-mixity significantly, due to the increased
compliance of the Material 2 region. Plasticity in Material 1 appears to lead to an
amplification of the mode-mixity, probably due to yielding at the crack tip leading to
disproportionate crack-opening displacements for the opening and shearing modes. The
plastic strain contour plots shown in Figure 8.23 support this conclusion.

Figure 8.25 show the development of plastic strains with crack extension for the high
plastic mismatch case. Plastic strain fields at critical loading are shown at different stages
of crack extension, for a crack situated 100 m from a 200 m wide ductile layer. After
several increments of crack extension (8.23(c)), there is a significant accumulation of
plastic strain in the ductile layer, extending a long way ahead of the crack tip
(a)

(b)

(c)

80%W
20%Cu
100%Cu

Figure 8.25: Plastic strain fields at critical loading (a) before extension and after crack extension of
(b) 100 m and (c) 200 m, for a crack situated 100 m from a 200 m wide ductile layer.

The plastic strains that develop adjacent to the crack tip can influence the non-linear
deformation during further crack extension, particularly for the high plasticity mismatch
case. For a given applied load, a 3.2 mm long crack positioned 100m from the interface
8-34

with a highly ductile region initially exhibited a higher SIF magnitude than an equivalent
crack in a specimen without yielding, 19.8 MPam1/2 as compared with 13.5 MPam1/2. The
crack was then extended, and the surrounding mesh modified. Prior plastic strains were
taken into account by the nodal release method described in Section 8.5 (p. 8-26). This
resulted in a significantly reduced SIF magnitude of 15.3 MPam1/2 for the same applied
load, whereas for a specimen without plastic yielding the SIF magnitude rose to 15.1
MPam1/2.

The reduction in SIF implies that, in the case of a crack closely parallel to a ductile layer, a
toughening effect can occur, due to the additional work associated with plastic deformation.
This is in contrast with the weakening effect due to the increased compliance. It should be
noted this toughening effect was very minor for the moderate plastic mismatch case. Using
the nodal release technique to take prior plastic strains into account in that case led to
reductions in SIF magnitude of less than 1%. In the following sub-sections, results are
presented and discussed for the moderate plastic mismatch and high plastic mismatch cases
respectively.

8.7.1 Moderate plastic mismatch

Figure 8.23 demonstrated that plastic deformation in the Material 2 region with increased
loading could influence crack-tip stresses. By increasing the compliance of Material 2, the
crack-tip stresses rotate around toward the interface as evidenced by the reduction in
negative mode-mixity. At low loads the residual stress is dominant, causing crack stresses
to be oriented away from the interface. As load increases, plastic deformation is induced in
the neighbouring copper-rich region. The resultant increase in compliance of this region
alters the mode-mixity, and increases the SIF magnitude.

Figure 8.26 shows the variation in (a) predicted deflection angle and (b) SIF magnitude
with applied load, illustrating the effects of plasticity and residual stress. Two different
yielding cases were considered, corresponding to the compliant central layer having the
properties of (i) the 20%W/80%Cu composite (o = 259 MPa) and (ii) pure Cu (o = 49

8-35

MPa). The contrast between the results for these highlights the important influence of
plasticity on crack-tip stresses.
70

Normalised SIF Magnitude, |K|/P [m-3/2]

(a)

= 49 MPa T = 0

60

o = 49 MPa T = 300

50

o = 259 MPa T = 0
= 259 MPa T = 300

40

30
20
10
0
0

200

400

600

800 1000 1200 1400 1600

Applied Load, P [N]

o = 49 MPa T = 0

60

(b)

= 49 MPa T = 300
o

= 259 MPa T = 0
o

= 259 MPa T = 300

o

20

Deflection Angle, [deg]

40

0
-20
-40
-60
-80

200

400

600

800 1000 1200 1400 1600

Applied Load, P [N]

Figure 8.26. Effect of applied load on (a) stress intensity factor magnitude and (b) predicted
deflection angle for a straight crack near an interface in a Cu-W layered composite specimen under
4PB loading. The effects of residual stress and increased plasticity are illustrated. Crack length was
3.2 mm, and crack position was 300 m from the interface. Two different yielding cases were
considered, corresponding to the compliant central layer having the properties of (i) the 20%W/
80%Cu composite (o = 259 MPa) and (ii) pure Cu (o = 49 MPa). (cf. Figure 2.43, p. 2-95)

8-36

At low loads, thermal mismatch is the dominant effect, when it is present. For the T =
300K cases, for example, high normalised SIF magnitudes and large negative deflection
angles result from the thermal stress contribution to crack-tip stresses, which is independent
of applied load. As load is applied, the crack-tip stresses are drawn back towards Mode I, as
they become more dominated by the applied loading and elastic mismatch. Further increase
in applied load brings the crack-tip stresses further around toward the interface, as plastic
deformation dominates the deflection behaviour. These results are similar to those obtained
by Chapa-Cabrera and Reimanis for copper-tungsten stepped graded specimens [2003], as
shown in Figure 2.43 in Section 2.8.3. The plastic deformation also causes an increase in
stress concentration at the crack tip, particularly when Material 2 is more ductile. This is a
result of anti-shielding due to the reduced compliance of the ductile region.

Figure 8.27 shows the effects of crack-tip position and crack-length on critical load for
crack propagation and predicted deflection angle. The critical load for fracture, Pc, such that
K = Kc = 16 MPam1/2, is shown normalized with that for an equivalent crack in a
homogeneous material (Material 1), Pchom, as it directly reflects the change in the overall
joint strength as measured experimentally. A sharp crack tip was assumed so that SIF for
crack initiation is the same as for propagation. Figure 8.27(a) shows that the deflection
angle generally decreases in magnitude with distance from the interface. The influences of
thermal mismatch and plastic mismatch on deflection angle both decrease with distance
from the interface, though the plastic mismatch influence seems to decrease more quickly.
Critical load generally increases with distance from the interface, which might be expected
as the stress concentrating effect of the interface declines.

Figure 8.27(b) illustrates the influence of crack length. Deflection angle appears to be fairly
unaffected, though critical load, relative to an equivalent crack in a homogeneous material,
increases with crack length. For longer cracks, the stress-field amplification effect of the
plastic mismatch becomes less dominant than the stress concentrating effect of finite
specimen geometry.

8-37

0.96

20

(a)

0.94

P (T = 0)

0.92

0.9

0.88

-5

0.86
c

-15

k (T = 300)

-20
0

200

400

600

800

0.84
0.82

(hom)
c

P (T = 300)

-10

Deflection Angle, [deg]

10

Normalised Critical Load, P /P

(T = 0)

15

0.8
1000

Distance from interface, x [m]

20

1
c

(x = 200 m)

0.95

P (x = 200 m) 0.9

10

(x = 600 m)

0.85

0
0.1

0.3

0.5

0.7

(hom)
c

Deflection Angle, [deg]

15

Normalised Critical Load, P /P

P (x = 600 m)

(b)

0.8
0.9

Normalised Crack Length, a/h

Figure 8.27: Effect of (a) crack-tip position and (b) crack-length on critical load, Pc, for crack
propagation (ie. for K = Kc) normalised by the critical load for a homogeneous specimen, Pc(hom),
and on predicted deflection angle, k, for a straight crack near an interface in a layered Cu-W
composite specimen under 4PB loading. Crack length in (a) was 3.2 mm.

8-38

Predicted propagation paths for three different yielding cases are shown in Figure 8.28.
This demonstrates that yielding in Material 2 has a significant effect on crack-path, in
agreement with the influence on mode-mixity and deflection angle shown in Figures
8.24(b) and 8.26(b). Yielding in Material 1 has only a minor influence on crack-path. This
suggests that the effects of near-tip plasticity on deflection angle are very minor. It should
be noted that the nodal release method was not used in these simulations so that prior
plastic strains were not taken into account. The fact that prior plastic strains had a
negligible influence in the moderate plasticity case, and the results of Li et al. [19], suggest
that this would not cause significant errors.

Yielding throughout
Yielding in Material 2
No yielding

Distance from base of specimen, y [mm]

5.2

4.8

4.4

3.6

3.2

0.4

0.8

Distance from interface, x [mm]

Figure 8.28. Effect of plasticity on crack propagation paths for cracks situated near the interface in
layered Cu-W composite structures. Initial crack length was 3.2 mm and initial crack positions of
200 m and 600 m from the interface were considered. Paths were predicted using finite element
analysis with either elastic or elastic-plastic deformation analysis.

8-39

8.7.2 High plastic mismatch

The effects of prior plastic strains were examined by coupling nodes along the crack faces,
then solving, and progressively decoupling them and solving until all crack-face nodes
were decoupled. In the high plastic mismatch case, thermal stresses were not considered
and deflection was always predicted toward the ductile layer. In the presence of residual
stress, crack-tip stresses would rotate away from the ductile layer. During this study it was
observed that there is a high dependence on mesh density, due to its influence on interface
roughness, and there also appears to be a significant dependence on crack-extension
increment length.

The effective toughness for a crack near an interface in the current configuration will result
from the anti-shielding and toughening effects. In this work, we calculate the effective
toughness by comparison with an equivalent crack in a homogeneous material:

Pcr

K eff
c = K c hom
Pcr

(8.24)

where Pcr is the critical load, such that the crack-tip stress intensity factor is equal to Kc,
and Pcrhom is the critical load for the same length crack in a homogeneous specimen with the
same dimensions. This takes into account the anti-shielding effect and any crack-extension
toughening effects that may occur.

In Figure 8.29, the increase in effective toughness with crack extension is illustrated. In this
case, the development of plastic strains under a constant load was considered. case, the
development of plastic strains under a constant load was considered. Strictly speaking, the
constant load case would only be relevant to fatigue. However, it enables the influence of
plastic strain build-up on stress intensity factor to be examined directly. SIF magnitude
values are normalised with respect to the initial value, to enable the influence of crack
extension to be examined. The SIF magnitude may be observed to decrease initially with
crack extension, in contrast with the case of no yielding where the SIF magnitude increases
due to increasing crack length. Two different ductile layer widths are compared. A sharper
increase in relative toughness is observed in the narrow gradient case, though a greater
overall increase is observed in the wide gradient case.
8-40

a = 0

Normalised SIF magnitude, |K|/|K|

Wide ductile region

Narrow ductile region

1.4

No yielding
1.2

0.8

0.6

0.2

0.4

0.6

0.8

Crack extension, a [mm]

Figure 8.29. Variation in crack-tip stress intensity factor with crack-extension, for a crack in the
brittle region (properties of 80%W no yielding allowed) close to a sharp interface with a highly
ductile region (properties of pure Cu) of width either 2mm or 200 m. SIF magnitude normalised
with respect to initial value, enabling influence of crack extension to be examined. Initially, SIF
magnitude decreases with crack extension, in contrast with the case of no yielding where SIF
magnitude always increases with crack length.

The increase in effective toughness with crack extension for cracks situated 100 m from
the interface with a ductile layer with high plasticity mismatch is shown in Figure 8.30(a).
In contrast to Figure 8.29, these results were obtained for critical load, ie. constant crack-tip
stress intensity factor. Effective toughness is calculated from the critical load for failure as
in Equation 8.25. The effect of variation in ductile layer width, w, is illustrated. Initially the
effective toughness is reduced due to the compliance mismatch, then the effective
toughness increases due to plastic strain build-up. These R-curves are comparable to those
predicted by Tvergaard and Hutchinson [1993,1994] for an interface crack between ductile
and brittle materials. The width of the ductile layer has a major influence on specimen
compliance, and therefore on stress intensity factor amplification (anti-shielding of the
crack tip). While a wider ductile region could allow for more plastic strain build-up, this is
8-41

not realised due to the much lower failure loads that result from the anti-shielding effect.
This is in concord with the conclusions of Tvergaard and Hutchinson [1994], Liu et al.
[1996], and Hasegawa et al. [2003].
In Figure 8.30(b), the effect of crack position on effective toughness, Kceff, for cracks at
various distances from the interface with a ductile layer of width, w = 100 m, with high
plasticity mismatch. Effective toughness is calculated from the critical load for failure,
corresponding to critical crack-tip SIF, as in Equation 8.25. Initial effective toughness
values, at a = 0 mm, before any prior plastic influences, are shown, along with effective
toughness values at a = 1 mm, after plastic-strains have built up. Effective toughening rate
values, dKceff/da, averaged over the first millimetre of crack extension, are also shown to
give a direct indication of the extent of crack-extension toughening resulting from plastic
strain build-up. The effects of variation in crack position, x, on initial toughness, crackextension toughening rate and final toughness (after crack extension of 1 mm), are
illustrated. Closer to the interface there is more anti-shielding effect, resulting in lower
initial toughness, but more crack-extension toughening occurs, as the ductile material is
closer to the crack tip, so that greater toughness values are eventually reached. Distance
from the interface leads to less extension toughening because (i) the stress field in the
plastic region will be less, as it is further from the crack tip, and (ii) the effect on the crack
tip will be less as the plastic strain build-up is further away.

Prior plastic strains were observed to have a negligible effect on crack-tip mode mixity.
This indicates that ignoring the influence of these in the propagation simulations for the
moderate plasticity case, as in Figure 8.28, was justified.

8-42

15

Effective Toughness K

eff

1/2

[MPa m ]

20

10

w = 0 m
w = 50 m
w = 85 m
w = 100 m

0.2

0.4

w = 140 m
w = 250 m
w = 800 m

0.6

0.8

Crack extension, a [mm]

10

1/2

eff

Initial Effective Toughness

4
Toughening rate

5
2

0
0

200

400

600

800

1/2

Effective Toughness, K

10

[MPa m /mm]

eff
/da
c

15

Crack-extension Toughening Rate,

Effective Toughness at a = 1 mm

dK

[MPa m ]

20

Distance from interface with ductile layer, x [mm]

Figure 8.30: (a) Increase in effective toughness, Kceff, with crack extension for cracks situated 100
m from the interface with a ductile layer of width, w, with high plasticity mismatch. The effect of
variation in ductile layer width, w, is illustrated. (b) Effect of crack position on effective toughness,
Kceff, at a = 0 and at a = 1mm and average toughening rate, dKceff/da, for cracks at various
distances from the interface with a ductile layer of width, w = 100 m, with high plasticity
mismatch. Effective toughness was calculated from the critical load for failure as in Equation 8.25.
Initially the effective toughness is reduced due to the compliance mismatch, then it increases due to
plastic strain build-up.

8-43

8.7.3 Discussion
As discussed in Section 2.5.3 (p. 2-65), it is apparent that plastic deformation has two
separate effects on the driving force for a crack situated close to an interface with a ductile
layer. These were each observed, along with their interaction, in the FE simulations in this
section.

Firstly, an increase in compliance associated with plasticity can increase the driving force
for crack growth. The presence of plasticity in the inner layer (Material 2) leads to an
enhancement of crack-tip stresses, effectively an anti-shielding or amplification effect, as
shown in Figures 8.26 and 8.27. This is similar to results reported by Sugimura et al. for
cracks approaching the interface [1995a,b], reviewed in Section 2.5.3 (p. 2-64). They
predicted amplification factors up to 2.6 and shielding factors less than 0.4 [1995a], and
they showed that negative (compressive) T-stress could increase the effect. This indicates
that values for SIF amplification of more than two predicted in the present study are
reasonable, given the presence of both elastic and plastic mismatches.

Secondly, the increased work associated with plastic deformation can decrease the crack
driving force as the crack extends. When the crack is very near the interface, and there is a
high plastic mismatch, a decrease of crack-tip stresses may occur with crack extension, as
predicted by Charalambides et al. [1990] and also discussed in Section 2.5.3 (p. 2-65). As
shown in Figures 8.29 and 8.30, when a crack is very near the interface, the SIF at a given
applied load decreases, due to the build-up of plastic strains. This corresponds to an
increase in the normalized critical load for crack propagation, or an increase in effective
toughness, Gc.
The relative magnitude of each effect will depend on the material properties of Materials 1
and 2, and on the crack and layer dimensions. However, each effect should have a different
dependence on the distance of the crack from the interface. It is seen in Figure 8.30(b) that
the increase in effective toughness with plastic strain build-up is strongly dependent on the
distance of the crack from the interface. On the other hand, a change in compliance should
not be as sensitive to crack location [Sugimura et al., 1995a], which is demonstrated in
8-44

Figure 8.27(a). The competition between these two effects will determine the concentration
of stresses at the crack tip.

The interaction between the two effects is interesting. Firstly, consider the case of high antishielding, or amplification, which leads to low failure loads. Assuming this leads to less
plastic deformation, there will be less extension toughening. This would limit the failure
load in later crack extension increments, resulting in less plastic deformation, and an
overall reduction in structural reliability. Conversely, a less significant anti-shielding effect
would lead to higher failure loads. This should result in more plastic deformation, which
will increase the effective toughness in subsequent crack extension increments. Higher
loads will then be required for further crack growth. As the crack grows, the load required
for crack extension will decrease. However, the stress field in the vicinity of the crack tip
would remain fairly constant, as would the plastic strains in the adjacent region of the
ductile layer. Comparing the constant load case, in Figure 8.29, with the constant crack-tip
SIF case, in Figure 8.30, the constant load case does not account for this feedback effect of
plastic crack-extension toughening. In any case, for fracture simulations, the constant SIF
approach is more applicable.

Distance from the interface leads to less extension toughening because (i) the stress field in
the plastic region will be less, as it is further from the crack tip, and (ii) the effect on the
crack tip will be less as the plastic strain build-up is further away. The width of the ductile
layer has a major influence on specimen compliance, and therefore on stress intensity factor
amplification (crack-tip anti-shielding). While a wider ductile region could allow for more
plastic strain build-up, this is not realised due to the much lower failure loads that result
from the anti-shielding effect. On the other hand, a very narrow ductile layer would limit
the extent to which toughening may occur.

An increase in the ductility of the inner layer leads to more stress amplification. While it
could also increase the possibility of crack-extension toughening, this is not realised due to
the low loads necessary for crack propagation. It could be predicted that higher crack-tip
toughness for the brittle material would lead to higher loads required for crack-extension,
8-45

so that more plastic deformation would occur in the ductile region. This would result in
more crack-extension toughening due to higher build-up of plastic strains, thereby leading
to a positive feedback situation. On the other hand, the antishielding effect could also be
influenced by the higher load. Increased yielding of the ductile layer would cause further
decreases in the effective toughness, due to the greater compliance mismatch. Thus, an
attempt to improve the toughness of the joint by utilizing a higher toughness outer layer
could be mitigated by the ensuing plastic dissipation in the ductile inner layer. This issue
may require further investigation in future.

8.8

Thermal & plastic mismatch: Further remarks

To summarise, apparently reasonable predictions may be obtained with finite element

modelling, for crack-tip stresses and critical loads for cracks near interfaces in the presence
of thermal residual stresses and plastic yielding. While the Irwin relation between crackopening displacements and stress intensity factors cannot be used when plastic deformation
occurs at the crack tip, an equivalent non-linear relationship may be obtained and used to
calculate stress intensity factors.

Deflection of cracks near interfaces can occur due to mismatch in any one or more of
elastic, plastic and thermal properties. The influence of residual stress tends to be more
complicated for materials with plasticity than for purely elastic materials. Plasticity tends to
mitigate the effects of residual stress on crack path. Accordingly, it is important to
characterise the residual stresses and plastic mismatch in an FGM system.

Several comments may be made regarding the comparison between the results in Figures
8.21 to 8.28 and those presented in Chapters 6 and 7 for simulations of crack growth in
graded alumina-epoxy specimens. Firstly, the deflection angles predicted for layered Cu-W
specimens in this chapter were low (-20 to +10) compared to those for alumina-epoxy
FGMs in Chapter 6 (up to +60). This suggests that the Cu-W system would be less
interesting, in terms of crack paths and the attendant variation in crack-tip environment
with crack propagation, for experimental investigation.
8-46

The elastic mismatch effect for Cu-W specimens, as illustrated in the T = 0K, elastic-only
case in Figure 8.22, was very minor despite the fairly significant mismatch in elastic
properties. It is less significant than in alumina-epoxy FGMs shown, for example, in
Figures 6.10 to 6.13 in Section 6.2. This comparison accords with the predictions in Figure
6.3 for the effect of variation in Youngs modulus ratio on crack-tip mode-mixity.

The deflection behaviour for layered Cu-W composites is predicted to be dominated by

thermal and plastic mismatch, rather than elastic mismatch. This is demonstrated by
comparing the various thermal and yielding cases in Figure 8.22. This contrasts with the
alumina/epoxy FGMs examined in Chapter 6 in which thermal and plastic mismatch effects
are understood to be negligible, as discussed in Section 3.3. This highlights the differences
in behaviour between different types of FGM. It was noted though, that both elastic
mismatch, as in Chapter 6, and plastic mismatch, in this chapter, influence crack path and
crack-tip stress concentration through compliance mismatch. Based on the findings in
Chapter 7 regarding crack path shape for cracks near bimaterial interfaces (ie. Figure 7.13
on page 7-23), it is important to have simulated the entire crack path in the layered Cu-W
composites, rather than rely on crack shape approximations. In addition, plastic mismatch
can result in increase in effective fracture resistance due to accumulation of plastic strains.
This effect is comparable to the development of a bridging zone, in terms of its overall
effect on crack growth.
8.9

Summary

The FE simulations in this chapter indicate that non-linear behaviour may be separated into
(i) that which occurs away from the crack, ie. remote plastic deformation and plastic
mismatch, and residual stresses, and (ii) that which occurs near the crack tip, ie. crack
bridging and near-tip ductile deformation.

Non-linear behaviour away from the crack will have similar types of effects on crack path
and stress concentration as elastic gradient. Thermal and plastic mismatch will combine
with elastic mismatch to define the field acting on the crack, which then determines crack
8-47

propagation direction and critical loads. It is understood that these influences may be
predicted with FE simulations.

Nonlinear behaviour near the crack-tip is understood to primarily influence effective

toughness and fatigue crack propagation resistance, with only a secondary effect on crack
path. Crack-bridging, for example, increases effective crack propagation resistance. While
it may have an effect on propagation path, the overall propagation direction is dictated by
the elastic gradient, and bridging will be a second-order effect. Similarly, accumulation of
plastic strains near the crack-tip can lead to an increase in effective toughness due to the
work absorbed during plastic deformation. Accordingly, it is important to consider the
effects of prior strain history on crack-tip stresses during simulations.

The influences of non-linear behaviour may be addressed in a similar framework to linearelastic behaviour. The results in this chapter provide some insight into FGMs beyond the
simple alumina-epoxy system investigated in the present study.

8-48

9 Experimental Results: Homogeneous Composites

The effect of material gradient on crack propagation in graded composites can best be
elucidated by comparison with cracks in non-graded composites. Furthermore, the
simulations of graded specimens rely on material properties, which cannot be obtained
from experiments on graded specimens. Accordingly, the mechanical and fatigue properties
of homogeneous alumina-epoxy composites were characterised over a range of
compositions, and are presented in this chapter.

The production and microstructural characterisation of these composites were described in

Sections 4.1 and 4.2, while mechanical and fatigue testing procedures were described in
Sections 4.3 and 4.4 respectively. The results in this chapter provide the relationships
between material composition, phase morphology and properties that underpin the
behaviour of graded specimens and are required for use in FE simulations of these
specimens. Also, the results for graded specimens can be compared with those of
homogeneous specimens to elucidate the influence of material gradient.
9.1

Elastic properties

A range of models for predicting effective elastic properties were discussed in Section 2.2,
and Youngs modulus values were predicted for 50%alumina-50%epoxy composite in
Table 2-1 (p. 2-19). The large difference in properties of the two constituents leads to
widely varying predictions from different models. This can be advantageous as the
increased variation, between different model predictions, enables improved identification of
the appropriate choice of model for composites with interpenetrating network structure.

Experimental results for Youngs modulus are shown plotted against composition in
Figures 9.1 and 9.2(a), in comparison with theoretical predictions. The results for Poissons
ratio are shown in Figure 9.2(b). The values obtained for the monolithic alumina and epoxy
samples are those shown in Table 1.1 on page 1.7, and are in agreement with results in the
literature [Ashby & Jones, 1996].

9-1

The isostress and isostrain bounds, given in Equation 2.3 on page 2-11, are shown in Figure
9.1 as curves A and B. While these bounds do contain the experimental results, they differ
widely, prohibiting their use for accurate predictions. The Hashin-Shtrikman bounds, given
in Equations 2.6 and 2.7 on page 2-12, are shown in Figure 9.1 as curves C and D. The
results also lie within these bounds; however, they provide only a slight improvement on
the isostress and isostrain bounds. The inadequacy of the Hashin-Shtrikman model is
probably attributable to the disparity in constituent properties, resulting in irregularities in
stress transfer at interfaces and stress distribution between phases. Results for the
powder/matrix composite samples are included in Figures 9.1 and 9.3. Compositions were
estimated from density measurements and microstructural analysis. This was difficult, due
to the disordered structure and residual porosity, so values are only approximate. The
properties are apparently dominated by the epoxy matrix phase, although the presence of
alumina particles does increase stiffness significantly.

Composite Young's Modulus, E* [GPa]

400
350
300

250

200
150
100

50

B
0
0

20

40

60

80

100

Composition [%vol Alumina]

Figure 9.1: Youngs Modulus Values. Experimental results for interpenetrating-network composite
() and powder composite () samples, and theoretical predictions: Isostress (A), Isostrain (B),
Hashin-Shtrikman upper (C) and lower (D) bounds.

9-2

400

Composite Young's Modulus, E* [GPa]

(a)

350
300
250

Tuchinskii Model:
Upper Bound
Lower Bound

200

EMA:
=1
=6

150
100

Wegner-Gibson
Model

50
0
0

20

40

60

80

100

Composition [%vol Alumina]

0.4

Composite Poisson's Ratio, * [GPa]

(b)
0.35
0.3
0.25
0.2
0.15
0.1

Experimental results
EMA prediction

0.05
0
0

20

40

60

80

100

Composition [%vol Alumina]

Figure 9.2 Effective mechanical property values for alumina-epoxy composites. (a) Youngs
Modulus: experimental results (), computational predictions () [Wegner & Gibson, 2001],
Tuchinskii model predictions [1983], and EMA with = 1 and = 6, (b) Poissons Ratio:
Experimental results () and EMA predictions ( = 6).

Predictions using the EMA, with shape factor () values between 1 and 6, show reasonable
agreement with measured values for Youngs modulus and Poissons ratio. The formulation
9-3

of the EMA was described in Sub-section 2.2.4 (p. 2-14), and is given in more detail in
Appendix A. Predictions from Tuchinskiis unit cell model [1983], as described in Section
2.2.3 (p. 2-14), and the finite-element model of Wegner and Gibson [2000a], described in
Section 2.2.5 (p. 2-18), are also included in Figure 9.2(a). These agree with experimental
results for alumina volume fractions above 0.8, however below this there is significant
disagreement. This may be attributed to the fact that the geometrically regular structures
assumed for these models, unit-cell (Fig 2-5, p.2-12) and stacked spheres (Fig 2.6, p.2-18)
respectively, differ significantly from the real structures (Fig 4.5, p.4.12). Accordingly, the
success of the EMA predictions is associated with its non-geometry specific nature.

The bounds predicted by Tuchinskiis unit-cell model [1983] are displayed in Figure 9.2(a).
Values predicted by this model are too high. This may be attributed to the assumption of a
simple regular cubic structure that, as shown in the microstructural images, was not the case
for these composites. The model of Wegner and Gibson [2000a] was based on stacked
spheres, which compares well with the microstructures for lower epoxy volume fractions,
shown in Figure 4.7(a,b) on page 4-10. Their model predicts a sharp drop-off in stiffness
with increasing interstitial phase volume fraction, which agrees with the experimental
results as shown in Figure 9.2(a). This is encouraging, although such a computational
model is time-consuming to produce and limited in applicability.
The EMA formulation allows different values of to be defined for each phase, resulting
in different shape tensors. This was not pursued in the current work, due to the uncertainty
in selecting shape factor values. Rather, a single shape factor was used for both phases. As
seen in Figure 9.2(a), the closest fit was obtained with the EMA by altering the inclusion
shape parameter, , to a value of 6. This is not physically unreasonable as microstructural
images indicated a fairly irregular internal geometry, particularly for the epoxy phase. Also,
from considering the results in Figure 9.2(b), the Poissons ratio of the alumina was
assumed to be 0.2, which is lower than values typically quoted in the literature [Ashby &
Jones, 1996]. This prediction is included in Figures 9.2 and 9.3, and in each of these cases
shows good agreement with measured results.

9-4

Figure 9.3 shows the strong correlation between Youngs modulus and density. Variation in
properties was observed between beam samples cut from the same plate. This was
attributed to compositional variation across the plates, which was identifiable in the
variation of measured density. These results are of particular importance from a processing
point-of-view and highlight the importance of ensuring sample homogeneity. It was
considered possible that samples may display anisotropic properties, as a result of the
production process. If samples were anisotropic, their flexural rigidities in different planes
would differ, which would lead to differing Youngs modulus results for in-plane and outof-plane measurements. As shown in Figure 9.3, these results did not differ significantly.
This was corroborated by microanalysis of samples in planes perpendicular (Figure 4.5 on
page 4.12), and parallel (Figure 4.8 on page 4.14), to the direction of foam-compression.
No anisotropy was observed.

Composite Young's Modulus, E* [GPa]

400
Expt results (Out-of-plane)
Expt results (In-plane)
EMA Prediction
Expt results - Powder

350
300
250
200
150
100
50
0
1

1.5

2.5

3.5

Measured Density [g/cm ]

Figure 9.3: Variation of effective Youngs modulus with measured density. Experimental results
for interpenetrating-network and powder composite samples, and EMA predictions ( = 6).

9-5

Figure 9.4(a) displays Youngs modulus values for several alumina preforms before and
after infiltration with epoxy, which are concordant with EMA predictions. Comparing
experimental results before and after infiltration indicates that the presence of the epoxy
increased the stiffness, compared with the porous alumina, as expected.

Composite Young's Modulus, E* [GPa]

400
350
300

Expt - infiltrated

250
200

EMA - infiltrated

150
100

EMA - porous
50

Expt - porous
0
0

0.1

0.2

0.3

0.4

0.5

Composition [Phase 2 Volume fraction]

Figure 9.4 Youngs Modulus for porous alumina specimens, before () and after () infiltration
with epoxy, and EMA with = 8.

The success of the EMA in simulating elastic properties for this type of composite structure
may be attributed to its lack of inherent assumption regarding internal geometry. Rather,
inclusions are assumed as being surrounded by an effective medium. In the composites
investigated, any particular microstructural element of alumina or epoxy would be in
contact with both alumina and epoxy though not in any regular configuration, and thus is
most accurately modelled via the effective medium concept. This is in agreement with the
conclusions of Wu [1966], Christensen [1990] and Munro [2001]. Accordingly, the EMA
was used to calculate elastic property distributions for the simulations of experimental
specimens, used for prediction and data analysis, in Chapter 10.

Due to the interpenetrating-network structure, percolation effects, identified in other FGM

systems [Reiter et al., 1997, Van Siclen, 2003] as discussed in Section 2.2 (p. 2-10), were
9-6

not present in the alumina-epoxy composites. It is possible that observed deviations from
the predictions of the EMA could be explained by the observed variation in microstructural
morphology across the composition range. Incorporating such effects into a model would
be difficult however, and of questionable benefit, given the reasonably good fit obtained
using the EMA in the formulation presented.

9.2

EMA for other IPN composites

To demonstrate the general applicability of the Effective Medium Approximation method

to composites with interpenetrating structures, EMA predictions have been obtained for
several composite systems investigated experimentally by other researchers. Results for
alumina-aluminium obtained by the authors and others [Prielipp et al., 1995, Neubrand et
al., 2002] are displayed in Figure 9.5 along with EMA predictions. Good agreement is
observed for Youngs modulus and also for thermal expansion coefficient, which has been
discussed previously by Hoffman et al. [1999]. This area has been pursued further [Moon et
al., 2004] with similar conclusions reached.

Effective Young's Modulus, E* [GPa]

EMA - CTE
350

11

EMA - E
10

300

Expt - CTE
9

250
8

200

0.65

0.7

0.75

0.8

0.85

0.9

0.95

6
1

-6

150
0.6

Expt - E

Effective Thermal Expansion Coefficient, * [x10 ]

12

400

Volume Fraction of Alumina

Figure 9.5 Effective Youngs modulus and thermal expansion coefficient for alumina-aluminium
composites, experiment () [Moon, 2004] and refs [Prielipp et al., 1996] () and [Neubrand et al.,
2002] (,), and EMA with = 10.

9-7

Normalised Effective Young's Modulus E*/E

(a)

Titanium-Magnesium
0.8
EMA Predictions
0.6
Experimental
Results
0.4

Tuchinskii
Model Bounds

Iron-Lead
0.2

0
0

0.2

0.4

0.6

0.8

(b)

Normalised Composite Young's Modulus E*/E

Volume Fraction (Second Phase)

1
0.9

EMA Predictions

0.8
0.7

Jedamzik
Results

Tuchinskii
Results

0.6
0.5

Tuchinskii
Model Bounds

0.4
0.3
0.2

0.2

0.4

0.6

0.8

Volume Fraction of Copper

Figure 9.6 Comparison of EMA predictions ( = 8) with experimental values and Tuchinskii model
predictions for Youngs modulus of (a) iron-lead and titanium-magnesium composites [Tuchinskii,
1983] and (b) tungsten-copper composites [Tuchinskii, 1983, Jedamzik et al., 1999] with
interpenetrating structures. E1 and Ew are Youngs modulus values for the more rigid phase in (a)
(Fe or Ti) and for tungsten in (b).

Tuchinskii [1983] compared theoretical predictions, from his model for co-continuous
structured composites, to experimental results for Fe-Pb, W-Cu and Ti-Mg composites. As
9-8

shown in Figure 9.6(a), the EMA predictions are very close to the measured values. In this
figure, E1 refers to the Youngs modulus of the first, more rigid, phase: Fe in the case of FePb composites and Ti in the case of Ti-Mg composites. The Tuchinskii model also provides
good predictions; however, these are given as upper and lower bounds, and hence are less
precise. Both Tuchinskii [1983] and Jedamzik et al. [1999] showed that Tuchinskiis model
could be used for tungsten-copper composites. Figure 9.6(b) shows that the EMA and
Tuchinskiis model provide very similar good predictions across a wide range of volume
fractions for W-Cu composites.

Wegner and Gibson have conducted a significant investigation into the mechanical
properties of interpenetrating-phase composites [2000a, 2000b, 2001a]. They produced and
characterised steel-bronze and steel-resin composites, and simulated their mechanical
response using a three-dimensional finite-element model. Whilst their computational model
provided good predictions, it would be very time-consuming to set up, and cannot be
applied across the full range of compositions due to the geometric assumptions involved.

As shown in Figure 9.7, the EMA provides sufficiently accurate predictions, which can be
improved by considering porosity in some specimens, as discussed by Wegner and Gibson,
or by varying the shape factor, . The steel-resin composites are of particular interest, due
to the extreme disparity in constituent properties, providing a useful comparison with
results obtained for alumina-epoxy composites. The sharp drop-off in stiffness with
increase in compliant phase volume fraction is even more significant in the steel-resin
system, which could be due to differences in stress transfer at phase boundary interfaces. A
good fit was obtained with a shape factor = 10. This suggests that the optimum choice of
shape factor may depend on the disparity in constituent material properties as well as the
geometry of the composite structure. However, as the appropriate choice of shape factor for
a particular composite system is not clear, experimental measurement of properties is
recommended.

9-9

Composite Young's Modulus, E* [GPa]

250
EMA including porosity
200

Steel-Bronze

150

Experimental
Results
100
=1
EMA Predictions
= 10

50

0.2

0.4

Steel-Resin

0.6

0.8

Volume Fraction of Steel

Figure 9.7 EMA predictions ( given) and experimental measurements [Wegner & Gibson, 2000b,
2001a] of Youngs modulus of steel-resin () and steel-bronze () composites with
interpenetrating-network structures.

9-10

9.3

Fatigue crack propagation

Crack initiation from a sharpened notch and subsequent propagation in specimens in the
composition range 5% to 50% epoxy was investigated under cyclic loading, as described in
Section 4.4. In alumina-rich specimens, as shown in Figure 9.8(a), cracks tended to steer
through the alumina phase, as this phase carried essentially the entire load. Some branching
was observed. The epoxy phase is known to be continuous so the crack must pass through
epoxy ligaments at some stage. In more epoxy-rich compositions, crack propagation
occurred in a brittle, discontinuous manner, illustrated in Figure 9.8(b). This was attributed
to the brittle nature of both phases and the relatively coarse, irregular composite structure.

The epoxy demonstrated highly brittle behaviour in bulk form, though it is possible that
fine ligaments could deform in a ductile manner. The possible micromechanisms are as
follows: crack-tip advance through alumina, through epoxy or along interfaces;
microcracking or interface delamination ahead of the crack tip; bridging of crack by
alumina grains, or epoxy ligaments. The choice of crack path depends on the comparative
toughnesses and the local stress distribution. Interfaces were shown to be much less tough
than either alumina or epoxy, though this could be related to the intrinsic stress
concentration associated with them. There was no apparent difference between crack
surfaces in fracture and fatigue. The occurrence of microcracking or delamination ahead
cannot be discounted, however it did not visibly manifest itself in any way, so will not be
discussed further. The occurrence of crack-extension effects implies that some kind of
bridging is present in these composites. The crack-extension distances over which
toughening occurred, and the extent of the toughening, indicate that it could not be entirely
due to grain-bridging. Hence, it was deduced that intact epoxy ligaments bridged the cracks
providing closure tractions that shielded the crack tip.

9-11

(a)

Crack tip

Steering through
alumina phase

(b)
Porosity

Tortuous
crack path
Discontinuous
crack growth

Branching

Notch-tip

500 m

Figure 9.8: Optical micrograph showing crack propagation path. Evidence of discontinuous crack
growth and surface porosity may be observed.

Figure 9.9 shows that an increase in epoxy content caused a decrease in the maximum SIF
required to initiate crack growth (ie. crack initiation toughness) under cyclic loading. This
is attributed to the significantly lower stiffness and higher susceptibility to fatigue
degradation of the epoxy phase. The curve fit in this figure was used to predict crack
initiation toughness for graded specimens in Section 10.1 in the following chapter.

9-12

Cyclic loading
Curve fit
Monotonic loading

1/2

Initiation Toughness, K [MPa m ]

00

10

20

30

40

50

Volume Fraction of Epoxy [%]

Figure 9.9: Effect of composition on initiation toughness, KC, under monotonic and cyclic (KC =
Kmax, R = 0.1) loading. Monotonic loading results obtained by L. Rutgers [2004].

Several specimens produced within the current project were tested under monotonic
loading by Lyndal Rutgers [2004], and several of these results are included in Figure 9.9,
showing the drop-off in initiation toughness is less significant under monotonic loading.
The monotonic loading results also indicated an increase in fracture toughness with crack
extension, as shown in Figure 9.10. This was attributed to crack-bridging by intact epoxy
ligaments, which have a higher failure strain than alumina, as was discussed in Sections 8.1
and 8.3. R-curves and crack bridging under monotonic loading were reviewed in Sections
2.5.2 and 2.5.3. The extent of crack-extension toughening in Figure 9.10, and the length
over which toughening occurs, are comparable with the R-curves for aluminium/alumina
composites shown in Figures 8.2 and 8.3.

Figure 9.11 shows that, under cyclic loading, crack propagation appeared to follow Paris
law, (da/dN) = A(K)m, with some scatter. As discussed in Section 4.4, the applied SIF
amplitude was varied, within the range resulting in crack growth, during fatigue testing so
that the correlation between crack growth rates and crack extension would be negligible.
While influences of crack-wake zone damage history may occur, they would contribute to
noise in the data rather than resulting in a systematic trend with crack extension. The results
9-13

in Figure 9.11 appear similar to fatigue crack growth rate results reported for other brittle

7
6

5% Epoxy
- Coarse
Microstructure

30% Epoxy

Effective Fracture Toughness, K

eff

[MPa m1/2]

materials [Ritchie, 1999, Gilbert et al., 1997].

5% Epoxy
Fine Microstructure

60% Epoxy

2
1
0

0.5

1.5

Crack Extension, a [mm]

2.5

Figure 9.10: R-curves for homogeneous specimens under monotonic loading. Results obtained by
L. Rutgers [2004], using specimens produced within the present study.

It was observed that the SIF amplitude required for crack growth generally increased with
crack extension. This is illustrated in Figure 9.12 which shows the applied SIF and fatigue
crack growth rate results for a 6% epoxy composite. Whilst applied SIF amplitudes
increased fairly uniformly with crack extension throughout the experiment, crack growth
rates fluctuated significantly after an initial decrease. It is believed the trends in SIF
amplitude, and some of the scatter in Figure 9.11, are due to the effect of crack-extension
toughening on fatigue crack growth rate. Assuming this is the case, and that these effects
may be extracted from the fatigue data, a novel data analysis approach was developed.

9-14

Fatigue Crack Growth Rate, (da/dN) [m/cyc]

-6

10

-7

30% Epoxy
m = 22

45% Epoxy
m = 25

10

6% Epoxy
m = 27

-8

10

-9

10

10

11% Epoxy
m = 24
-10
6

10

10

1/2

Applied Stress Intensity Factor Range, K [Pa m ]

Figure 9.11: Fatigue crack growth rates for several different compositions, showing Paris law
behaviour.

1/2

Applied SIF Amplitude, K [MPa m ]

10

-7

10

-8

10

2
-9

da/dN

10

-10

0.2

0.4

0.6

0.8

Crack Extension, a [mm]

10
1.2

Fatigue Crack Growth Rate, da/dN [m/cyc]

-6

Figure 9.12: Systematic decrease in crack growth rate with crack extension, with increased applied
SIF range, indicative of crack-extension toughening effects.

9-15

9.4

Fatigue analysis

As discussed in Section 2.4.5, the interaction between crack-bridging and cyclic loading
can be complicated and the traditional approach for quantifying fatigue behaviour by
presenting crack propagation rates as a function of SIF amplitude becomes insufficient
when R-curve effects are present. In this section, a novel approach is used to analyse
fatigue data in order to deconvolute the effects of crack growth rate and crack extension,
and examine the variation of fatigue crack growth resistance with crack extension. As shall
demonstrated in the following chapter, this is very important for graded specimens, where
crack-extension and crack-growth rate effects are accompanied by spatial variation in
intrinsic (crack-tip) fatigue resistance and bridging, as discussed in Sections 2.6 and 2.9.

The underlying assumption of the proposed analysis is that, for a given composition, the
mechanisms of crack advance at the crack tip, and accordingly the value of the Paris-law
exponent, do not vary with crack-extension. The Paris law (Equation 2.54) may be
rearranged to form an expression equal to the Paris law coefficient, A, which is assumed to
be constant for a given specimen:
m da
A = (K )
(9.1)

dN
This suggests that the measured values, Kmeas and (da/dN)(meas), for SIF amplitude and crack

growth rate could be related to adjusted values, Kadj and (da/dN)adj:

(meas)

(adj)

m da
da
K
= K (adj)
(9.2)

dN
dN
Here, the relationship between applied SIF range and resulting crack growth rate is being
meas m

utilised to infer the effect, of changing one parameter, on the other.

Applying this adjustment to the experimental results, measured K values that resulted in
differing crack-growth rates may be compared, by estimating the adjusted values that
correspond to a standard adjusted crack growth rate, (da/dN)adj = (da/dN)std:
K

adj

= K

meas

(da dN )(std )
(da dN )( meas)

(9.3)

9-16

The value for m was taken from the initial Paris law fit, as in Figure 9.11. The value for
(da/dN)std was arbitrarily chosen as 10-7 m/cycle. The choice of this value scales the results
by a constant factor but does not alter any trends in the adjusted data. Essentially, this
adjustment provides an estimate, assuming that the Paris law is applicable, of the SIF
amplitude value that would have resulted in crack extension at the standard rate. The
adjusted SIF was interpreted as a measure of fatigue crack propagation resistance, ie. Kadj
= KCF. Under conditions of constant load ratio, R, this is proportional to the adjusted
maximum loading, (Kmax)adj = Kadj/(1-R), so, even if the crack propagation is driven by
Kmax rather than K, the adjustment may still be applied. It is stressed that this adjustment
method does not provide absolute results; rather it is an approximate way of quantifying the
relative variation in fatigue crack propagation resistance, in a given specimen.

The variation in adjusted SIF with crack extension is shown in Figure 9.13 for several of
the alumina-epoxy composite specimens. A systematic increase in fatigue resistance with
crack extension is observed, comparable with the R-curves in Figure 9.10. It appears that
the crack-extension effects are less significant under cyclic than monotonic loading, which
is consistent with the susceptibility to fatigue degradation of the epoxy phase.

6% Epoxy

11% Epoxy
5

Fatigue Resistance, K

adj

1/2

[MPa m ]

30% Epoxy
2

45% Epoxy
0

0.5

1.5

Crack Extension, a [mm]

Figure 9.13: Variation of effective fatigue crack propagation resistance, ie. adjusted SIF as in
Equation (9.3), with crack length for a range of compositions.

9-17

Applying a fit to the fatigue resistance curve, the effective fatigue resistance, KCF(a), may
be predicted at a given crack extension, a. Rearranging Equation (9.2) and making the
substitution Kadj = KCF(a) yields:

da

dN

(std)

K meas
da

dN
K CF (a )

(meas)

(9.4)

A form of the Paris law similar to this, in which the applied SIF amplitude is normalised by
fracture toughness, has been used to quantify fatigue behaviour in grain-bridging ceramics
[Gilbert et al., 1997]. Expanding the logarithms of each side, and writing (da/dN)std as a
constant, A:

da
log

dN

(meas)

= log(A') + m log K meas m log(K CF (a ))

(9.5)

This implies that each measurement of fatigue crack growth rate may be corrected,
removing the dependence on crack-extension:

da
log

dN

(corr)

da
= log

dN

(meas)

+ m log(K CF ) = log(A') + m log K meas

(9.6)

In Figure 9.14, the fatigue crack growth rate results shown in Figure 9.11 are corrected to
remove the crack extension effect.

Results before and after correction are compared directly in Figure 9.15. This clearly
demonstrates that applying this correction to the data improves the linearity of the data
significantly, which confirms that much of scatter was attributable to R-curve effects. The
improvement is more significant for composites with lower epoxy volume fractions as these
exhibited a greater increase in crack-growth resistance with crack extension. Had a
different value of (da/dN)(std) been used, the values of (da/dN)(corr) would be multiplied by a
factor, without altering the general trend. The corrected data in Figures 9.14 and 9.15
represent estimates of how the fatigue crack growth rate data would have appeared were
there not an influence of bridging on fatigue crack growth resistance. The corrected da/dN
is higher than the measured da/dN, concordant with the expectation that crack-bridging
would be expected to slow down crack growth. This is also in agreement with Figure 9.10.

9-18

Fits to the effective fatigue crack propagation resistance data (fatigue R-curves) in Figure
9.13 were used to predict effective fatigue resistance for graded specimens in Chapter 10
(Section 10.4).
-2

Corrected Fatigue Crack Growth Rate, (da/dN)

corr

[m/cyc]

10

6% Epoxy
-3

10

-4

10

30% Epoxy
-5

10

-6

10

-7

10

45% Epoxy

11% Epoxy

-8

10

10

10

Applied Stress Intensity Factor Range, K [Pa m ]

1/2

Figure 9.14: (a) Fatigue crack growth rate results from Figure 9.11 corrected to remove crack
extension effect, as in Equation (9.6).

The increase in toughness, or more generally crack-growth resistance, with crack extension
introduces another variation into the crack propagation problem. Under monotonic loading,
this variation may be isolated and quantified by expressing the applied SIF required for
crack growth as a function of crack extension, the well-known R-curve. In the case of
cyclic loading, where crack growth rate varies with applied loading, the situation is more
complicated. The proposed analysis is one way to quantify this extra variation by adopting
an assumed relationship between loading and crack growth rate, in this case the Paris law,
9-19

so that loadings which result in different crack growth rates may be compared directly. This
approximately isolates the variation in resistance to fatigue crack growth with crack

Fatigue Crack Growth Rate, (da/dN) [m/cyc]

extension.

10

-4

Corrected
10-6

15% Epoxy
Measured

10

Corrected

-8

6% Epoxy
- 10

10

10- 12
106

Measured

7
SIF Amplitude, K [Pa m0.5] 10

Figure 9.15: Fatigue crack growth rate results before and after being corrected to remove crack
extension effect, as in Equation (9.6).

The analysis proposed here is applicable to any set of fatigue results containing scatter
where fatigue behaviour is understood to obey the Paris law. The adjustment, based on the
assumed Paris-law relation, provides a way of comparing results that, because they pertain
to different crack growth rates, would not otherwise be comparable. If the loading has been
applied in such a way that crack growth rates do not correlate with crack extension, then
any correlation between the adjusted SIF amplitudes and crack extension will be real, and
will reflect a systematic increase in crack-growth resistance. This will not be dependent on
the mechanism causing this increase. It is stressed that this data adjustment procedure does
not refer to the underlying physics of the situation, rather it provides a simple way of
capturing the overall effect of increasing crack growth resistance on fatigue crack growth.

9-20

This type of analysis could potentially be applied to grain-bridging ceramics such as

alumina and silicon carbide, and to fibre composites. Given this approach is not specific to
particular toughening mechanisms, transformation-toughening ceramics and piezoelectric
materials could also potentially be analysed. All that would be required is the data for
applied loading, crack growth rate and crack length. Generally crack growth rate and
applied loading results are presented in the literature, without the corresponding crack
length data being included.

The issue will also be relevant to the graded specimens, as crack-bridging is likely to
develop, as in the homogeneous specimens, which will influence the propagation of cracks.
The influence of crack-bridging is likely to vary across the graded region [Cai & Bao,
1998]. For crack propagation in a graded composite structure with property variation in the
x-direction, the applied SIF, intrinsic fracture toughness and the crack-tip shielding will all
depend on crack-tip position, xtip:
Ktip = Kappl - Kbr(a, Kappl, xbr) = Kc(xtip)

(9.7)

Under cyclic loading, there will be a rate-dependence also:

Ktip = Kappl - Kbr(a, Kappl, da/dN, xbr)

(9.8)

This is addressed in Section 10.4 in the following chapter.

9.5

R-curve Analysis

Increase in effective fracture toughness with crack extension under monotonic loading is
frequently analysed to quantify the relationship between bridging tractions and crackopening displacement, as discussed in Section 2.4.3 (p. 2-48). Indeed, due to the
dependence of R-curves on specimen geometry and loading configuration [Fett et al.,
2000c], this relationship is generally seen as being more fundamental than the R-curve
itself [Jacobsen & Srensen, 2001].

In this section, the crack-extension effects observed in the alumina-epoxy specimens under
monotonic and cyclic loading are analysed to obtain approximations for the underlying
bridging traction relations. The crack growth resistance data plotted as a function of crack
9-21

extension was fitted by calculating R-curves from tri-linear bridging traction relations. The
tri-linear bridging law was chosen to simplify calculations; it represents a very simplified
approximation to the real bridging traction relation expected for a large number of
statistically-distributed tractions from individual deforming ligaments of epoxy.

Weight functions provide a much faster method for back-calculation of traction relations
than finite element modelling approach described in Sections 8.1 and 8.2. The weightfunction expression for an edge-cracked beam was given by Fett and Munz [1992]:

h ( x, a ) =

2
1
3/ 2
+1

(
1
a
/
h
)
A
(
1
x
/
a
)
(
a
/
h
)

a 1 x / a (1 a / h ) 3 / 2
,

(9.9)

where x is position behind crack tip, a is crack length and h is specimen height, and the
values of A are given in Table 9.1.
Table 9.1: Values of coefficient, A, in Equation (9.9) [Fett & Munz, 1992].
=0
0.498
0.54165
-0.19277

A
=0
=1
=2

=1
2.4463
-5.0806
2.55863

=2
0.07
24.3447
-12.6415

=3
1.3187
-32.7208
19.763

=4
-3.067
18.1214
-10.986

This weight function was used to calculate the applied stress intensity factor:
a

K app = K 0 + h ( x ' , a )p(u ( x ' ))dx '

(9.10)

The initial bridging traction distribution, p(0)(x), was calculated from the assumed p(u)
expression and the Irwin profile for the crack-tip SIF:

u 0 (x) =

K0
E'

8(a x )

(9.11)

After the first calculation of applied SIF for a given crack extension, an iterative algorithm
was used. The crack-opening displacement profile was calculated using the applied SIF
value from the previous iteration:

u app ( x ) =

K app
E'

8(a x )

(9.12)

9-22

u ( x ) = u app ( x )

a
a a '

4
p
(
u
(
x
'
))
h
(
x
'
,
a
'
)
dx
'

h ( x , a ' )da '

E ' a x a

(9.13)

This was used, along with the assumed tri-linear p(u) function and the weight function, to
calculate the applied SIF value for the current iteration. Generally four or five iterations
were required to obtain stable solutions, to within 0.1%. This was conducted for a number
of crack extensions to produce an R-curve. Having set up all the calculations in an Excel
spreadsheet, the parameters describing the p(u) function could be adjusted and the effect on
predicted R-curve observed immediately. These parameters were adjusted manually until a
good agreement with experimental results was achieved.

Examples of calculated tractions and the resultant fit to fatigue crack propagation resistance
data are shown in Figures 9.16 and 9.17. These may be compared with the simple tractions
Figure 2.20. The significant difference between the tractions acting under cyclic and
monotonic loading is similar to the change in traction with cyclic loading observed by
Gilbert and Ritchie in SiC, shown in Figure 2.23.

Bridging Traction, p [MPa]

50
6% Epoxy Cyclic Loading
11% Epoxy Cyclic Loading
30% Epoxy Monotonic Loading

40

30

20

10

0
0

20

40

60

80

100

Crack Opening Displacement, u [ m]

Figure 9.16: Calculated traction distribution and the resultant fit to fatigue crack propagation
resistance data (cf. Figures 2.20 and 2.23)

9-23

1/2

Fatigue Resistance, K [MPa m ]

cf

0.2

0.4

0.6

0.8

1.2

Crack Extension, a [mm]

1/2

Fatigue Resistance, K [MPa m ]

cf

0.5

1.5

Crack Extension, a [mm]

Figure 9.17: Fit to fatigue crack propagation resistance profile for (a) 6% and (b) 11% epoxy.

Good fits to each set of data were obtained, with apparently reasonable approximations for
bridging traction relations. This indicates that the R-curve effects observed under both
9-24

monotonic and cyclic loading were most likely due to the development of a bridging zone
behind the crack-tip. A more rigorous approach, such as that used by Hu and Mai [1992] to
calculate bridging tractions from compliance measurements, could be employed and could
also be feasibly applied to FGMs; this was beyond the scope of the present study but could
be an interesting area for future work.

1/2

Fracture Toughness, K [MPa m ]

0.5

1.5

Crack Extension, a [mm]

Figure 9.18: Fit to fracture toughness profile for 30% Epoxy composite under monotonic loading.

9.6

Summary

The mechanical properties of composites with interpenetrating structures were investigated

and compared to theoretical predictions. The effective medium approximation (EMA) was
generally found to provide better predictions than other models. The EMA was applied
successfully to a range of interpenetrating composite systems produced by differing
methods, indicating a general suitability for use with interpenetrating structured
composites. Investigation of composites with significantly disparate constituent properties
allows improved differentiation between predictions from different models. This
highlighted the general applicability of the EMA in preference to other models. For
composites with significantly different constituent properties, EMA predictions appear to
9-25

be reasonably accurate, and are preferable to the model of Tuchinskii. The advantages of
the EMA over the unit-cell model of Tuchinskii, or more computationally involved models
such as those of Wegner and Gibson, are that internal phase geometry is not specified and
that exact values, rather than a range, are predicted. Nevertheless, experimental
determination of properties for validation of models is recommended.

The fatigue behaviour of the alumina-epoxy composites was similar to that expected for a
brittle material, as reviewed in Section 2.4.4: high sensitivity to applied SIF amplitude;
sensitivity to maximum applied SIF. Crack-extension effects were observed to influence
crack growth under both cyclic and monotonic loading. The crack growth rate under cyclic
loading was approximately related by Paris law to the applied loading magnitude. Within a
fairly narrow SIF range (about 10% of the critical SIF for fast-fracture), crack propagation
rate went from zero to critical, though this SIF range increased with crack extension due to
the development of a crack-wake zone. The presence of crack-extension toughening (Rcurve) effects influenced resistance to fatigue crack propagation. The increases in
toughness and fatigue crack propagation resistance were fitted well with apparently
reasonable approximations to the bridging traction relations.

The results in this chapter were interesting in their own right, as well as providing
important data for the FE simulations of graded specimens. In extending the results of the
present study to other FGM systems, the specific composite property relations and fatigue
micromechanisms become less relevant; only the resultant mechanical and fatigue
behaviour is important. This was characterised for the homogeneous composites and is
compared with the results for graded specimens in the following chapter.

9-26

10 Experimental Results: Graded Specimens

The propagation of fatigue cracks in graded composites is the focus of the present study.
Accordingly, experimental observations of this process, presented, compared with
simulation predictions and discussed in this chapter, are an important component of the
results of the study. Findings from FE simulations in Chapters 6 to 8, regarding the
influences of elastic gradient, fatigue crack propagation resistance gradient and crackextension toughening, are considered in terms of their contribution to the overall process.
Production and testing procedures for graded specimens were described in Sections 4.1, 4.4
and 4.5. Several examples of stable crack propagation paths are shown in Figure 10.1. In
each of these, cracks deflected toward the more compliant, epoxy-rich side, and
experienced a variation in properties whilst propagating across the gradient region.

(a)

1 mm

(b)
(c)

1 mm

1 mm

Figure 10.1: Crack propagation paths observed in graded specimens, showing deflection towards
the more compliant epoxy region. The effect of gradient width on crack propagation angle is
apparent in the comparison of (b) (w = 20) and (c) (w = 10).

10-1

10.1 Crack initiation

Predicted and measured values for crack initiation loads (peak) and initial deflection angles
are shown in Table 10.1. As discussed in Section 5.5 (p. 5-31), predictions were obtained
with the FE model, using data from homogeneous specimens and methods described in
Chapter 5. Displacement correlation was used to calculate SIFs and the local symmetry
criterion, with the algorithm in Section 5.4, was used to predict propagation direction.
Predictions of crack path and initiation load relied on assumptions for the elastic property
and initiation toughness profiles and, thus, inherited any inaccuracy in these assumptions.

Specimen
Code-Name

Gradient Width,
w [mm]

Notch Position,
= x/w

Number of steps,
Ns

Predicted Failure
Load, Pinitest [N]

Measured Failure
Load, Pinitmeas [N]

Predicted
Deflection Angle,
kest [degrees]

Measured
Deflection Angle,
kmeas [degrees]

Table 10.1: Comparison of predicted and measured values for failure load and deflection angle for
graded specimens. Specimens that fast-fractured are denoted with *.

G1-1
G1-2*
G2-1
G2-2
G3
G4
G5*
G6*
G7
G9
G10-1*
G10-2*
G11
G12-1*
G12-2
G14-1
G14-2
G15
G16
G17-1
G17-2
GM1*

19
19
22
19
19
19.6
20
19
16
8.3
9.8
9.8
9.55
9.55
9.7
19.1
19.4
18.7
21.6
4.86
5
8.7

0.5
0.83
0.5
0.3
0.75
0.42
0.55
0.57
0.5
0.24
0.82
0.82
0.69
0.93
0.54
0.76
0.73
0.37
0.76
0.73
0.76
.43

5
5
6.5
6
6
6
6.5
6
1
6
6
6
6
6
6
6
6
6
7
6
6
6

388
460
475
480
391
380
570
619
267
95
393
484
168
390
275
563
877
122
422
208
285
428

370
600
450
530
350
400
600
750
265
70
580
770
140
590
450
540
N/a
100
330
180
250
505

53.1
28.2
31.8
41.3
24.9
28.4
44.1
19.6
10.2
47.5
45.9
44.4
46.8
40.9
49.9
31.1
27.7
52.5
27.2
43.5
59.4
15.1

52.6
32
30
26.6
23.5
24.4
48.8
4.5
6.2
51.3
45.6
42.5
50.1
42
54.1
30.7
n/a
52.1
20.5
41.6
56.2
14.2
10-2

Crack initiation under cyclic loading led to fast fracture in some cases, whilst in others
crack propagation remained stable. The specimens were very brittle, particularly at
initiation, before a crack-wake zone had developed, so the difference between loads
resulting in stable crack propagation, and those resulting in fast fracture, was slight. This
presented a challenge for successfully initiating stable crack propagation.

The results in Table 10.1 are also depicted graphically in Appendix F. There was generally
good agreement between predicted and measured values of crack initiation load and initial
deflection angle. Most predictions of deflection angle were accurate to within 10%, with
the exceptions of G2-2, G4, G6 and G7. This provides strong evidence that crack deflection
was predominantly influenced by elastic gradient, and that this was successfully modelled
with the FE simulations. It is not surprising that there were some discrepancies between
predicted and observed deflection angles, given the stochastic nature of crack initiation
from a notch and the influences of microstructure. Deflection angles, varying from 5 to
60, spanned a similar range to those observed by Chapa-Cabrera [2002] and Hoffman et
al. [2001], and were significantly greater than those observed by Rousseau and Tippur
[2000], as discussed in Section 2.8.3 (p. 2-94).

Most predictions of initiation load were accurate to within 30%, with around half being
accurate to within 15%. Given the uncertainties involved in calculating stress concentration
and estimating initiation toughness, this is acceptable. There were several cases in which
initiation load was significantly underestimated (G10-1, G10-2, G12-1 and G12-2). This
may have been due to notch-effects, where a higher load is required to initiate fracture from
a slightly blunt notch. The fact that three of these specimens fast-fractured suggests this
may indeed be the case.

The good agreement, between predicted and measured values for loads and deflection
angles, indicates that the assumed property distributions, from results in Chapter 9, and the
FE modelling procedures, described in Chapter 5, were satisfactory.

10-3

10.2 Crack propagation paths

After initiation, the cracks were observed to propagate across the gradient. This occurred
irrespective of the stability of crack advance, so that both fast fracture and fatigue cracks
could be analysed in terms of the effects of elastic gradient on crack path. Examples of
crack propagation trajectories were shown in Figure 10.1. The mechanisms of crack
propagation appeared to be the same as in the homogeneous composites, described in
Section 9.3 (p. 9-11). Figure 10.2 shows that two specimens with similar gradient width
and profile, notched at similar positions, yield fairly similar crack paths. There is some
deviation, apparently due to microstructure-induced tortuosity and the influence of step
interfaces, but the average propagation directions are very similar.

(b)

(a)
Decreasing stiffness

Figure 10.2: Comparison of two cracks at similar positions ( 0.8) in specimens with gradient
width, w = 10mm.

Figure 10.3 compares an experimentally observed fast-fracture crack trajectory, in a

specimen with gradient width, w = 10mm, with the predicted path from FE simulation. As
discussed in Section 5.5 (p.5-31), the simulations of crack paths, presented throughout this
chapter, were conducted individually for each of the experimental specimens. Specimen
dimensions were measured and input to the model, and property distributions were
10-4

estimated from the measured compositional distributions and homogeneous specimen

results in Chapter 9. This contrasts with the simulations in Chapters 6 to 8, in which
specimen dimensions were standardised and property distributions were given by simple
functions. Figure 10.3 shows the good agreement between experiment and predictions
across the gradient and into the epoxy region for that particular specimen.

Distance from base of specimen, y [mm]

22

Interface
between
gradient
& epoxy

Experiment
FE prediction

20

Decreasing Stiffness
18

16

14

Notch-tip
12

Graded composite
4

Epoxy
10

Horizontal distance, x [mm]

Figure 10.3: Fast fracture crack path with initial crack position, 0.9, in a medium-width
(10mm) gradient specimen, compared with prediction from FE simulation.

In Figure 10.2, there is evidence of microstructural and step-interface influences on the

crack path. These influences are even more apparent in Figure 10.4. Figure 10.4(a) shows a
crack path in a wide-gradient specimen with poor connectivity at a step-interface, resulting
in inferior fatigue resistance for cracks propagating up the interface. After initial deflection
due to the elastic gradient the crack reached the interface then deflected up it. Eventually,
the crack deflected off the interface and continued into the next step. Figure 10.5 shows
another crack trajectory influenced by an imperfect interface. The crack path showed
10-5

excellent agreement with finite element predictions for the first few millimetres of crack
propagation. Upon reaching the interface, the crack deflected up it away from the direction
predicted from elastic stress fields.

(a)
(b)
Branching

Local
deflection
Imperfect
interface
Tortuosity

Local
branching

Initial macroscopic
deflection
Figure 10.4: Factors influencing crack path. (a) Crack in wide-gradient specimen showing
macroscopic deflection, cracking up imperfect interface, branching and deflection off the interface.
(b) Crack in narrow-gradient specimen showing effect of local heterogeneity, resulting from the
composite structure, on crack path.

10-6

3.5

Experiment
FE Prediction

Vertical Crack Extension, y [mm]

2.5

1.5
Crack tip for
calculations in
Figure 10.6

0.5

0
0

0.5

1.5

2.5

Horizontal crack extension, x [mm]

Figure 10.5: Fatigue crack path in a wide gradient (w 20) specimen, with inferior fatigue crack
propagation resistance at the interface between steps.

This issue is illustrated further in Figure 10.6, which shows the variations of Mode II SIF
and mechanical energy release rate with kink-angle, for the experimentally-observed crack
in Figure 10.5, at the point when its tip has just reached the interface. The local symmetry
criterion cannot be used due to the asymmetry introduced by the weak interface so we must
turn to the maximum energy release rate criterion, introduced in Section 2.3.5 (p.2-36). The
crack is understood to propagate in the direction in which the applied energy release rate,
G(), is equal to the critical value, Gc(). For the crack at the interface, the critical energy
release rate will be low for = 0 and higher for other angles, as discussed in Section 2.5.2.
Figure 10.6 shows that the applied mechanical energy release rate varies with kink angle
but much less significantly than does mode II SIF, kII. This implies that a reduction in
toughness, by a factor less than two, will result in crack propagation up the interface. The
local symmetry criterion is clearly inappropriate for predicting crack path in the presence of
10-7

anisotropy, as discussed in Section 2.3.5. A further observation may be made, that the
toughness anisotropy required to cause propagation along the interface must be greater for a
crack at a greater deflection angle, due to the lower value of G(0)/G(k). This implies that
cracks in narrower gradients are less likely to be influenced by toughness anisotropy at

G - FE
G - Parabolic fit 350

II

1/2

400

2.5

300

k - FE

1.5

II

G(k)

250

G(0)
G(k) 0.55

200

k - Linear fit
II

0.5

150

Mechanical Energy Release Rate, G [J/m ]

Mode II Stress Intensity Factor, k [MPa m ]

interfaces.

10

20

30

40

50

60

70

100

Propagation Direction, [deg]

Figure 10.6: Variation of mode II SIF at the kink-tip, and mechanical energy release rate, with
kink-angle for a crack, as in Figure 10.5, with crack tip situated at the interface.

Figure 10.7 shows crack paths in FGM specimens with similar gradient profiles but
differing gradient widths. The effect of gradient steepness on deflection is clearly observed
as cracks in the narrower, steeper, gradients deflected more (~50) than those in wider
gradients (~25). This comparison is illustrated in Figures 10.1(b) and 10.1(c). The effects
of varying initial notch position and length do not appear significant over the ranges
investigated, in concord with conclusions from FE simulations in Chapter 6. The agreement
between observed and predicted crack paths is better for cracks in the 10mm gradients than
in the 20mm gradients. This is understood to be due to the influence of interfaces, as in
Figure 10.4(a), as this reduces the overall extent of deflection. As discussed above, this is
more likely to occur for cracks with lower deflection angles. With the exception of the
10-8

influence of toughness anisotropy at weak interfaces, the effect of steps on path appears to

(a)

Displacement from base of specimen, y [mm]

be negligible, as was predicted in Chapters 6 and 7.

22

20

Experiment

Simulation

18

16

14

12

10

10

12

14

16

20

18

(b)

Displacement from base of specimen, y [mm]

Displacment from ceramic-gradient interface, x [mm]

20

18

16

14

12

10

10

12

Displacement from ceramic-gradient interface, x [mm]

Figure 10.7: Crack propagation paths in graded samples under four-point bending experimental
results (full lines) and modelling predictions (dashed lines) for cracks in gradients of (a) 20mm and
(b) 10 mm approximate width.

10-9

There were three influences on crack path, acting at different length scales:

Elastic gradient which caused macroscopic deflection at the scale of the gradient
region, ie. millimetres to tens of millimetres. This was taken into account by the FE
simulations, as described in Chapters 5 and 6.

Heterogeneous composite structure resulted in crack path a tortuosity, at the tens to

hundreds of microns scale. This was essentially a random fluctuation in crack path,
similar to that observed in homogeneous specimens, as in Figure 9.8 on page 9-12. This
is understood not to influence overall crack path so was not included in FE simulations.

Interfaces between steps, in some of the specimens, displayed inferior crack

propagation resistance, so cracks propagated up the interface rather than continuing to
traverse the graded region. This resulted in crack path perturbation in the millimetre
scale, causing a reduction in overall propagation angle. This effect was not taken into
account in FE simulations, so FE predictions for crack path provided a maximum bound
for the deviation from a straight line.

Several further comments may be made in relation to findings in previous chapters. In

Section 6.3, it was demonstrated that toughness gradient is unlikely to have a significant
influence on crack propagation path in a graded specimen. The results here do not provide
any evidence to the contrary, with the exception of the influence of step-interfaces. There
was some variability observed between predicted and experimental propagation paths. This
is comparable with the variability associated with different values of n as in Ch 6 or,
arguably, the variability due to bridging in Chapter 8. From the findings in Chapter 7, had
the full crack path shapes not been used in the FE simulations, the agreement between
predictions and observations would not have been as good. From Section 8.4, the effect of
bridging on path would not have been very great even if the bridges affected CT modemixity.

The generally good agreement between experiment and simulations provides verification of
the model, thereby validating the predictions in Chapters 6 and 7.

10-10

10.3 Fatigue crack propagation

As a result of their deflection, cracks traversed the gradient region as they propagated. In
doing so, they experienced a variation in composition, and hence properties, at the crack
tip. This section presents fatigue crack propagation resistance profiles, calculated from SIF
amplitude, K, values and measured crack growth rates,

da

/dN, using the adjustment

procedure introduced in Section 9.4 (p. 9-16). SIF amplitude, K, values were calculated
from measured loads and K/P (equivalent to K/P) values obtained from FE simulations.
As discussed in Section 4.4 (p. 4-19), there are no standard expressions for non-straight
cracks in graded specimens, so finite element modelling was required to obtain the factor
relating crack-tip stress intensity factor to applied load, K/P. These stress intensity
calculations relied on assumptions of spatial property variation and inherited any
inaccuracy in these assumptions. The K/P profiles determined from FE simulations are
included in most figures showing calculated fatigue crack propagation resistance profiles to
illustrate the process of calculation more clearly.

The microstructural processes associated with fatigue crack propagation were understood to
be similar to those in monotonic fracture: decohesion of interfaces; cracks in alumina;
degradation and cracks in epoxy. The processes at the crack tip were essentially brittle, and
the same as under monotonic loading. Crack propagation mechanisms appeared to be the
same in graded specimens as in homogeneous specimens (see Figure 9.8). Crack-tip
advance appeared to occur in a discontinuous (jumping) manner. Degradation of the epoxy
bridging ligaments in the crack wake is understood to have occurred more significantly
under cyclic loading. Figure 10.8 shows fatigue crack growth rate results for several graded
specimens. These may be compared with the equivalent results for non-graded composites
shown in Figure 9.11 on page 9-14. While the da/dN vs. K curves are steep, a hallmark of
brittle behaviour, the graded specimens exhibited a relatively high spread of SIF
amplitudes. This was understood to result from the variation in fatigue crack propagation
resistance experienced by cracks in graded regions. Results for specimens in which cracks
traversed more compositional steps were spread over a wider range of applied SIF, as
expected from the wide variation of fracture resistance with composition, as in Figure 9.13.
10-11

Fatigue Crack Growth Rate, (da/dN) [m/cyc]

10

-5

10

-6

10

-7

10

-8

= 0.35
Propagation
in one step

= 0.65
Propagation
across five steps

= 0.75
Propagation
across two steps
10

-9

1/2

Applied Stress Intensity Factor Range, K [MPa m ]

Figure 10.8: Fatigue crack propagation rate results for several FGM specimens. Propagation over
more of the gradient resulted in a wider range of applied SIF values.

In Figure 10.9(a), experimental and simulated crack paths are compared, for a widegradient specimen with the crack initially located toward the alumina side. The observed
crack path is very similar to the predicted path, even after the onset of fast fracture and after
the crack passed from the graded composite region to the epoxy. The adjusted SIF
amplitude, Kadj, shown in the figure, is interpreted as a measure of the fatigue crack
propagation resistance, as discussed in Section 9.4 (p.9-17). This increased with crack
extension as the crack propagated within the composition step in which it was initially
located, as shown in Figure 10.9(b). At the interface with the next step, which had higher
epoxy content and lower toughness, the crack suddenly fast fractured, due to the sharp
decrease in effective toughness. The increase in fatigue crack propagation resistance with
crack extension is similar to that observed for homogeneous specimens, in Figure 9.13.

10-12

(a)
10

Onset of
fast fracture

0
-7

-5

-3

adj

1/2

22

K/P - Graded

20

Experimental Results

18

Fast
Fracture

16
14

Interface
between
steps

12
10
8

K/P - Homogeneous
0

Crack Extension, a [mm]

/kN]

0.5

Fatigue Resistance, K

-1

Relative Stress Intensity, K/P [MPa m

(b)

[MPa m ]

-9

Figure 10.9: Fatigue crack propagation in a graded specimen (w = ~20mm, = 0.35): (a) predicted
and observed crack paths, and (b) variation in effective fatigue resistance, Kadj, and relative stress
intensity, K/P, calculated for the graded specimen, compared with that calculated for a
homogeneous specimen.

Figure 10.10 shows the effective fatigue resistance profile, Kadj, for a crack in a relatively
wide gradient, in which stable propagation occurred only within a single compositional step
(15% epoxy). A steady increase in effective crack propagation resistance was observed and
attributed to the development of a bridging zone behind the crack tip. At the interface with
the next step, which had higher epoxy content and lower toughness and fatigue resistance,
10-13

the crack suddenly fast fractured, due to the sharp decrease in effective toughness and the
increase in stress concentration. The wide variation between values for the factor, K/P,
calculated with gradient taken into account, and those calculated as for a homogeneous
material, highlights the importance of incorporating material gradient in such calculations.
14

1/2

Experimental Results
adj
K
12

10

K/P - Graded
8

1
6

K/P - Homogeneous
0

0.5

1.5

2.5

Crack Extension, a [mm]

3.5

Step
Interface

-1.5

Fatigue Resistance, K

adj

Normalised SIF, K/P [x10 m

[MPa m ]

Figure 10.10: Measured fatigue crack propagation resistance profile for a graded specimen in
which stable crack propagation occurred within one compositional step. Included is the assumed
variation in K/P used to calculate crack propagation resistance.

The variation in fatigue crack propagation resistance, Kadj, experienced by a crack,

moving across the interface from one compositional step to the next, is illustrated in Figure
10.11. The point at which the crack crosses the step interface is shown in Figure 10.11(a),
which also shows that the fatigue crack path was fairly close to the predicted path.
Approximate compositions for each step, which are included in the plot, were used, along
with results from homogeneous specimens, to estimate intrinsic (crack-tip) fatigue
resistance as a function of crack-tip position. As the crack moves from one step to the next,
the effective fatigue resistance, Kadj, and relative stress intensity, K/P, both decreased
suddenly. The decrease in fatigue resistance, Kadj, occurs due to the change in
composition at the crack tip. This is accompanied by a drop in intrinsic crack-tip fatigue
10-14

resistance, as shown in the figure. The decrease in relative SIF, |K|/P, occurs due to the
change in effective stiffness at the crack tip. The decrease in crack-growth resistance was
dominant, so the crack advanced significantly and a reduction in applied loading was
required in order to avoid unstable crack growth. The crack-extension effects are
understood to also change as the crack crosses the step-interface, for two reasons. Firstly,
the crack-opening displacement profile behind the crack tip depends on the crack-tip stress
intensity factor and the local stiffness, which both change suddenly across the interface.
Secondly, the bridging traction relation depends on volume fraction of bridging phase and
on microstructure which both change across the interface. As the crack tip crosses the
interface, the crack-wake zone will still be predominantly situated in the previous
compositional step, which could result in an effective dampening of the sharp change in
fatigue resistance at the interface.

A similar effect is observed in Figure 10.12. In this case, crack propagation remained stable
across a large portion of the gradient so the influence of several interfaces could be
examined throughout the crack propagation. A series of sudden reductions in toughness
was observed, corresponding to the crack moving from one composition step to the next.
Within the steps, the fatigue resistance increased, due to R-curve effects similar to those
observed in the homogeneous specimens. This led to fluctuations in the effective crack
propagation resistance. This is essentially a competition between two different effects. On
one hand, the fatigue resistance will decrease suddenly at an interface between
compositional steps, due to change in intrinsic (crack-tip) resistance. On the other hand, the
fatigue resistance increases steadily as the crack propagates within a compositional step,
due to R-curve effects. The R-curve effects diminished with increasing epoxy content, as in
Figures 9.10 and 9.13, so that the reductions in resistance at interfaces led to an overall
reduction in the total fatigue-crack propagation resistance. Furthermore, the development of
the bridging zone behind the crack is understood to have a dampening effect on the sharp
drops in fatigue crack growth resistance at interfaces. Comparison, in Figure 10.12(b), of
the crack-propagation resistance values calculated incorporating material gradient with
those from the expression for homogeneous materials, again highlights the importance of
including material gradient in calculations.
10-15

5
y [mm]

Interface
between
steps

(a)

Experiment
Simulation

0
0

x [mm] 4

10

1/2

K/P

-3/2

Fatigue Resistance, K

Estimated Crack-tip
Fatigue Resistance

Normalised SIF, K/P [x10 m

Adjusted
Experimental adj
Results K

adj

(b)

[MPa m ]

Step Interface
1

23% Epoxy

16% Epoxy
0

Crack Extension, a [mm]

Figure 10.11: Crack propagation across interface between steps: (a) predicted and observed crack
paths for crack in gradients of width 20 mm, and (b) variation in effective toughness and relative
stress intensity across interface between steps.

10-16

(a)
3

y [mm]

Experiment
Simulation

Fatigue Resistance, K

adj

1/2

[MPa m ]

(b)

x [mm] 4

Step Interfaces

3.5

adj

- calculated as for
graded specimen

3
2.5
2
1.5

Estimated Crack-tip
Fatigue Resistance

1
0.5
0
0

adj

- calculated as for nongraded specimen

1

Crack Extension, a [mm]

Figure 10.12: Fatigue crack propagation across gradient: (a) predicted and observed crack paths for
gradient of width 5 mm, and (b) variation in effective crack-propagation resistance, as calculated for
a graded and a non-graded specimen, with change in position across the gradient. Crack-extension
effects are in competition with effects of changing composition at crack tip.

It is apparent that the variation in effective crack-propagation resistance in graded materials

results from a competition between (1) crack-extension effects, which tend to increase
resistance, and (2) spatial variation in intrinsic and extrinsic contributions to crack10-17

propagation resistance. The adjustment method developed in this work for homogeneous
composites provides a useful approach for analysing this phenomenon in graded
composites. The combination of crack-extension and compositional variation effects led to
cyclic variations in effective crack propagation resistance across steps, though an overall
decrease was observed. Bridging had a notable effect on crack propagation resistance,
although, as discussed in the previous section, an effect on crack path was not discernible.

The observations of variation in fatigue crack propagation resistance with crack

propagation through the gradient region are similar to the variations in effective toughness
observed by Chung et al. [2001] and Moon et al. [2001] for crack propagation parallel to
the gradient under monotonic loading, as discussed in Section 2.7.5. The results here are
also influenced by propagation path, however, along with spatial property variation, which
adds an additional aspect to the process. The large elastic gradients and wide variation in
fatigue crack propagation resistance across the compositional range for the alumina-epoxy
system, as discussed in Section 3.3, enabled the effects of propagation path and spatial
compositional variation to be examined clearly.

The influence of phase continuity at step interfaces on the effective properties in that
region, and the possibility of smoothing of the compositional steps was introduced in
Section 2.2.6 (p. 2-19) and discussed further in Sections 4.2 and 5.1. Simulations were run
with and without smoothing, as discussed in Section 5.1. It was observed that without
smoothing, results could become erratic near the step-interfaces. This is shown in Figure
10.13 for the specimen featured in Figure 10.12. The influence on calculated fatigue crack
propagation resistance profiles suggests that some smoothing in the assumed effective
property distribution would be appropriate. This is a possible area for further work in the
future.

10-18

Fatigue Resistance, Kadj [MPa m0.5]

3.6

Without smoothing
With smoothing

3.2

2.8

2.4

Step-Interfaces
2

0.5

1.5

Crack Extension, a [mm]

Figure 10.13: Comparison of calculated fatigue crack propagation resistance profiles with and
without smoothing of the elastic property distribution at interfaces between steps. Results without
smoothing become erratic near the step-interfaces.

10-19

10.4 Comparisons of specimens

The effect of gradient width is illustrated in Figure 10.14. Figure 10.14(a) shows that a
narrower graded region, which has a steeper gradient, causes the most deflection whilst a
wide region causes the least deflection. This is in agreement with the findings from FE
simulations in Section 6.1, and those of other researchers discussed in Section 2.8.1. The
effect on toughness profile is twofold: cracks in steeper gradients propagate at a wider
angle, and the graded region is narrower, so these cracks will traverse the gradient region
much faster than those in less steep gradients. The implications of this are shown in Figure
10.14(b). The crack in the wide (w = 20) gradient remained within one composition step
and experienced a smooth increase in toughness, whilst the crack in the narrow gradient (w
= 5) moved across several steps, and experienced significant fluctuations in toughness.

Figures 10.15 and 10.16 demonstrate the effect of differing crack position for gradients of
width 10mm and 20mm respectively. In each case, there are only minor variations between
crack paths, with predicted and observed paths showing reasonably good agreement. This is
in concord with the conclusion from FE simulations in Section 6.2; initial crack position
does not have a major effect on average crack propagation angle. The influence on fatigue
crack propagation resistance is more significant. Cracks starting in different positions are
located within different compositions, and are therefore expected to experience different
intrinsic fatigue resistance at the crack tip and differing degrees of crack-extension
toughening effects. This is noted in both the 10mm wide (Figure 10.15(b)) and 20mm wide
(Figure 10.16(b)) gradients.

10-20

(a)

w = 20 - expt
w = 20 - fem
w = 10 - expt
w = 10 - fem
w = 5 - expt
w = 5 - fem

Vertical Crack Extension, y [mm]

1/2

Crack Propagation Resistance, K [MPa m ]

Horizontal Crack Extension, x [mm]

(b)

w = 20
w = 10
w=5

Crack Extension, a [mm]

Figure 10.14: Effect of gradient width on fracture in graded specimens: (a) predicted and observed
crack paths for gradients of width 5, 10, and 20 mm, and (b) variation of effective toughness for
these specimens.

10-21

Vertical Crack Extension, y [mm]

(a)

= 0.7 Expt
= 0.7 FE
= 0.54 Expt
= 0.54 FE
= 0.24 Expt
= 0.24 FE

1/2

cf

(b)

Crack Propagation Resistance, K [MPa m ]

Horizontal Crack Extension, x [mm]

1
= 0.7
= 0.54
= 0.24
0

Crack Extension, a [mm]

Figure 10.15: Comparison of (a) crack paths and (b) fatigue crack propagation resistance profiles
for cracks at different positions in gradients of medium width (w 10mm).

10-22

14

= 0.75 Expt

12

= 0.75 FE
= 0.5 Expt
= 0.5 FE

Vertical Crack Extension, y [mm]

(a)

= 0.35 Expt
10

= 0.35 FE

10

1/2

cf

(b)

Crack Propagation Resistance, K [MPa m ]

Horizontal Crack Extension, x [mm]

= 0.75
= 0.5
= 0.35

Crack Extension, a [mm]

Figure 10.16: Comparison of (a) crack paths and (b) fatigue crack propagation resistance profiles
for cracks at different positions in wide gradients (w 20mm).

10-23

10.5 Effective resistance to fatigue crack growth

Predicted variations of cyclic load amplitude, P, for crack growth with crack extension,
with and without consideration of crack extension toughening, are shown in Figures 10.17
and 10.18 for several specimens. Effective fatigue crack propagation resistance values were
estimated from results of homogeneous specimens with corresponding compositions, as in
Section 9.4. These were then used in the FE model to predict P values, assuming R=0.1,
with and without the inclusion of crack extension toughening effects. In Figure 10.17, the
cyclic load amplitude, P, profile from experiment is compared with the predicted profile.
The load profile, predicted with crack-extension toughening, shows much better agreement
with the experiment results than the profile predicted without these effects. This
demonstrates that R-curve effects have a major influence on effective propagation
resistance and must be included for accurate predictions.
6

P - experiment

400

Estimated K (a)
cf
- with R-curve

300

Fast
Fracture

200
2
100

Estimated K
o
Predicted P
- without R-curve
- without R-curve

Step
Interface

cf

Applied Cyclic Load Variation, P [N]

Predicted P - with R-curve

Estimated Fatigue Resistance, K (a) [MPa m ]

500

1/2

0
0

0.5

1.5

2.5

3.5

Crack Extension, a [mm]

Figure 10.17: Load profile for a graded specimen in which stable crack propagation occurred
within one compositional step. The assumed variation in Kceff, used to predict load profiles with and
without R-curve effects, is also included.

Figure 10.18 shows the predicted cyclic load amplitude, P, profile for a specimen in
which the crack remained stable after crossing a step interface. In this case, a change in
10-24

crack-tip fatigue crack propagation resistance was observed. The values in each
compositional step were estimated from homogeneous specimen results. The contribution
of crack-bridging to total crack propagation resistance was approximately estimated using
the following equation:
K cf ( a, x, x) = K cf (0, x tip ) (1 + f br ( a, x))

(10.1)

where Kcf(a=0,x=xtip) is the intrinsic crack-propagation resistance value at the crack tip,
estimated from results of homogeneous specimens in Section 9.3, and fbr is the relative
increase in fatigue resistance due to bridging. This was estimated for each composition,
from crack-extension dependence results for homogeneous specimens in Section 9.4. When
the crack moved across the interface, there would be a variation in composition along the
length of the bridging zone. In this case, the bridging factor was approximated based on an
average composition. As seen in Figure 10.18, there is a reasonably good agreement
between predicted and observed load profiles, and the inclusion of R-curve effects into
predictions is important.
5

Estimated K (a)
cf
- with R-curve

600

4
500

P - experiment

400
2

Estimated K
o
-without R-curve

300

Predicted P
- without R-curve

200

0.8

1.6

Step
Interface
2.4

3.2

1/2

100

cf

Applied Cyclic Load Variation, P [N]

Predicted P - with R-curve

Estimated Fatigue Resistance, K (a) [MPa m ]

700

4.8

5.6

0
6.4

Crack Extension, a [mm]

Figure 10.18: Load profile for a graded specimen in which stable crack propagation occurred
within one compositional step. Included in (a) is the assumed variation in K/P used to calculate
crack propagation resistance in (a), and in (b) the assumed variations in Kceff, used to predict load
profiles with and without R-curve effects.

10-25

Bridging effects were also examined in more detail, by deriving fatigue resistance profiles
from bridging traction relations, as for homogeneous composites in Section 9.5. The
bridging traction relations used are shown in Figure 10.19. The corresponding predicted
fatigue resistance profiles are compared in Figure 10.20 with experimentally observed
profiles for several specimens. Figure 10.21 compares predicted fracture toughness profile
with the experimentally measured profile obtained by Rutgers [2004] for a graded aluminaepoxy specimen under monotonic loading. The observed profiles are fitted well by the
predicted profiles, with an appropriate choice of bridging traction relation. It must be
stressed that this analysis is approximate, as the weight function used to relate bridging
closure tractions to crack-tip stress intensity factor (Equation 9.10) was that for a
homogeneous specimen. The equivalent weight function for a graded specimen was not
known. The bridging traction relations obtained here are similar to those for homogeneous
composites in Section 9.5.
= 0.75 Cyclic Loading
= 0.35 Cyclic Loading
= 0.65 Monotonic Loading

Bridging Traction, p [MPa]

50

40

30

20

10

0
0

10

20

30

40

50

Crack Opening Displacement, u [ m]

60

Figure 10.19: Bridging traction relation profiles obtained by fitting of predicted fatigue crack
propagation resistance profiles (or fracture toughness profile, in monotonic loading case) to
observed profiles, as shown in Figures 10.19 and 10.20.

10-26

1/2

Fatigue Resistance, K [MPa m ]

(a)

cf

0.4

0.8

1.2

1.6

2.4

2.8

3.2

Crack Extension, a [mm]

cf

1/2

Fatigue Resistance, K [MPa m ]

(b)

Crack Extension, a [mm]

Figure 10.20: Fatigue crack propagation resistance profiles for cracks, at initial positions (a) =
0.75 and (b) = 0.35, in graded alumina-epoxy specimens. Experimental measurements () are
compared with predictions from self-consistent solution of bridging equations (Equation 9.10 to
9.14) with bridging traction relations given in Figure 10.19.

10-27

1/2

Fatigue Resistance, K [MPa m ]

cf

Crossed
step interface
2

0.5

1.5

Crack Extension, a [mm]

Figure 10.21: Variation of fracture toughness with crack extension for a crack, at initial position
= 0.65, in a graded alumina-epoxy specimen under monotonic loading. Experimental results () of
Lyndal Rutgers [2004] are compared with predictions from self-consistent solution of bridging
equations (Equations 9.10 to 9.14) with bridging traction relation given in Figure 10.19.

The findings in this section indicate that the influence of crack-extension effects in graded
specimens was similar in nature and magnitude to that in homogeneous specimens, for each
of the cyclic and monotonic loading cases. The prediction of crack-extension effects for
cracks propagating across several layers can become complicated, although the simple
approximation in Equation 10.1 seems to provide a reasonable estimate. Further work in
this area, beyond the scope of the present study, may be useful.

10-28

10.6 Moir interferometry

To provide a verification of the finite element predictions, a specimen was examined with
Phase-shifted Moir interferometry (PSMI), as described in Section 4.5. An asymmetric
graded specimen (30x8x6mm) with gradient width 8.7mm was tested in a 20-10mm span
four-point-bend configuration. Figure 10.22 shows the initial Moir pattern at low applied
load, which is indicative of the displacement field around the crack and interface.

(a)

Interface

Surface
flaw
Crack tip

(b)

Interface

Surface
flaw
Crack tip

Figure 10.22: PSMI fields in the (a) y and (b) x directions at low applied load (100N, K 0.25
MPam). The influences of the crack and the polymer-gradient interface are visible.

10-29

As applied load was increased, the density of displacement contours increased as shown in
Figure 10.23. The interface may be observed, with the composite region on the left
exhibiting low displacements and the polymer region on the right exhibiting much greater
displacements. The deformation of the polymer region was so great that useful data could
not be obtained there as the applied load approached that for failure. The asymmetry of the
displacement field around the crack tip may be observed in these images also. These are
similar to the Moir patterns observed by Chapa-Cabrera [2002] for layered graded Cu/W
composites. Asymmetric displacement fields, obtained from FE simulation of the specimen,
are shown in Figure 10.24.
Gradient Region

Epoxy

(a)

Crack tip

(b)

Crack tip

Figure 10.23: PSMI fields in the (a) x and (b) y directions at higher applied load (400N, K 1
MPam). The influences of the crack may be observed, along with the massive disparity between
displacements in the composite region and those in the more compliant polymer region on the right.

10-30

(b)

(a)

Figure 10.24: Predicted displacement fields in the (a) y and (b) x directions from FE simulation.
These may be compared qualitatively with observed Moir fields in Figures 10.22 and 10.23 and
displacement fields in Figure 10.25.

Figure 10.25 shows the displacement fields extracted from Moir results, just prior to
failure. The asymmetry is visible in both the x and y direction displacement fields, and the
initial propagation can be approximately predicted, as indicated in the figure.
Estimated Propagation Direction

(a)

(b)

Crack tip
Figure 10.25: Experimental displacement fields (with contours), in the (a) y and (b) x directions at
high applied load, extracted from Moir interference fields. The asymmetric deformation around the
crack tip may be observed. Propagation direction predicted from this asymmetric field is shown.

10-31

The differences between the predicted and observed displacement fields in Figures 10.24
and 10.25 result from the Moir fields having been centred on the crack tip at each load
increment. This introduces a relative rotation between the two fields, though the asymmetry
is reflected in each case nonetheless. The asymmetry is also observed in the principal stress
field in Figure 10.26, which also shows good agreement with the estimate propagation
direction as in Figure 10.25.

Estimated
Propagation Direction

Figure 10.26: Principal stress field around crack tip for Moir specimen, as predicted with FEM.
The asymmetry is apparent, and shows good qualitative agreement with the propagation direction
predicted from Moir displacement fields in Figure 10.24.

As demonstrated in Figure 10.27, the observed propagation direction shows good

agreement with the direction predicted from Moir displacement fields in Figure 10.25, and
also with the initial propagation direction predicted with finite element analysis.
Subsequent deviation was observed, possibly due to toughness anisotropy at the stepinterface.

10-32

(a)
Surface-flaw

Epoxy

y [mm]

(b)

Step Interface

Notch-tip
Expt
FEM

Interface
0
0

x [mm]

Figure 10.27: Experimentally observed crack path compared with (a) propagation direction
predicted from Moir displacement fields and (b) finite element predictions.

Although only one specimen was tested with Moir interferometry, the results obtained
provide a valuable qualitative benchmark. Combined with the other experimental results
and finite element simulations, this strengthens the arguments that the crack deflection and
propagation direction is dictated by the elastic gradient in the specimens studied. The use
an alumina-epoxy composite specimen enabled large elastic gradient effects to be observed.
The results here also indicate that further Moir work on alumina-epoxy composites and
other types of step-graded composite specimen would be most fruitful. To summarise, the
Moir results, in agreement with FE predictions and experimental crack propagation
results, provide further validation of the FE predictions in Chapter 6.

10-33

10.7 Comparison with monotonic loading

The comparison of results from specimens under cyclic loading and those obtained for
monotonic loading by Lyndal Rutgers [2004] are illustrated in Figures 10.28 to 10.31.
Monotonic loading specimens were also simulated with the FE model, as discussed in
Section 5.5 (p.5-31), and predictions for crack paths are included in the figures. In Figure
10.28, results from two fatigue experiments and one fracture experiment on gradients of
width, w = 20mm, with notch position around =0.7. The crack paths in 10.28(a) are very
similar and crack propagation resistance profiles (ie. fracture toughness for monotonic
loading, fatigue crack propagation resistance for cyclic loading) in 10.28(b) all show a
significant rise over the first few millimetres of crack extension.

Crack Propagation Resistance, K [MPa m ]

1/2

(b)

Vertical Crack Extension, y [mm]

(a)

Cyclic A - Expt
Cyclic A - FEM
Cyclic B - Expt
Cyclic B - FEM
Monotonic - Expt
Monotonic - FEM

Horizontal Crack Extension, x [mm]

Cyclic Loading A
Cyclic Loading B
Monotonic Loading

Crack Extension, a [mm]

Figure 10.28: Comparison of (a) crack paths and (b) crack propagation resistance for specimens
under cyclic and monotonic loading, with gradient width w 20mm and notch position 0.7.

In Figures 10.29 and 10.30, the paths and propagation resistance profiles are presented for
cracks in a similar position ( = 0.7) in gradients of 5mm and 10mm width respectively.
For the 5mm wide gradients, both fracture and fatigue cracks are observed to wander
slightly away from the FE predictions, and they both exhibit fluctuations in crack
propagation resistance with crack extension across the gradient. For the 10mm wide
10-34

gradients, the paths are very similar and close to FE predictions, but there is a more
significant difference between the crack propagation resistance profiles under cyclic and
monotonic loading. This is likely to result from slight differences between the specimens
and from more significant degradation of the bridging zone in the cyclic loading case, as
observed for homogeneous specimens in Chapter 9.

Vertical Crack Extension, y [mm]

(a)
3

1/2

Crack Propagation Resistance, K [MPa m ]

(b)

Cyclic - Expt
Cyclic - FEM
Monotonic - Expt
Monotonic - FEM

Horizontal Crack Extension, x [mm]

Cyclic Loading
Monotonic Loading
0

Crack Extension, a [mm]

Figure 10.29: Comparison of (a) crack paths and (b) crack propagation resistance for specimens
under cyclic and monotonic loading, with gradient width w 5mm and notch position 0.7.

10-35

Similar results are observed for cracks initially located closer to the epoxy region, in
gradients of width, w = 10mm. The paths are very similar but the crack propagation
resistance is much lower under cyclic loading, due to more degradation of the epoxy phase,
as in homogeneous composites (ie. Figures 9.9 and 9.13 in Chapter 9).
5

Vertical Crack Extension, y [mm]

(a)
4

Cyclic - Expt
Cyclic - FEM
Monotonic - Expt
Monotonic - FEM

1/2

Crack Propagation Resistance, K [MPa m ]

Horizontal Crack Extension, x [mm]

(b)

Cyclic Loading
Monotonic Loading
4

1
0

Crack Extension, a [mm]

Figure 10.30: Comparison of (a) crack paths and (b) crack propagation resistance between
specimens under cyclic and monotonic loading, with gradient width around 10mm and notch
position around = 0.7.

10-36

Vertical Crack Extension, y [mm]

(a)
3

Horizontal Crack Extension, x [mm]

1/2

Crack Propagation Resistance, K [MPa m ]

(b)

Cyclic - Expt
Cyclic - FEM
Monotonic - Expt
Monotonic - FEM

Cyclic Loading
Monotonic Loading
3

Crack Extension, a [mm]

Figure 10.31: Comparison of (a) crack paths and (b) crack propagation resistance between
specimens under cyclic and monotonic loading, with gradient width around 10mm and notch
position around = 0.3.

Results from the parallel investigation by Lyndal Rutgers into crack propagation in
alumina-epoxy FGMs under monotonic loading were presented in Figures 10.28 to 10.31.
These were compared with FE crack path predictions and generally showed good
10-37

agreement, indicating that paths did not differ significantly between monotonic and cyclic
loading for alumina-epoxy graded composites. Crack paths showed no discernable
systematic variations. Effective crack propagation resistance was lower under cyclic
loading, as would be expected. It is understood that the microstructural failure mechanisms
associated with crack extension were the same in each case. The fracture toughness and Rcurve results observed for graded specimens under monotonic loading are similar to those
for homogeneous specimens, in Figures 9.9 and 9.10.

The similarity of crack paths under monotonic and cyclic loading, despite different R-curve
behaviours, in agreement with the results in Section 9.3, indicates the effect of cyclic
loading on bridges had negligible effect on crack path, and therefore that crack bridging has
a negligible effect on crack-path. From considerations of this issue in Section 8.4, it appears
that the effects of bridging on crack-tip stress intensity factors must be very similar in each
mode.
10.8 Summary
The experimental methods compared well with those used by others to investigate fracture
and fatigue in FGMs. An advantage of cyclic loading was that stable crack growth could be
observed as the crack moved through the graded region. While this had been achieved
previously by Chung et al. [1999] and Moon et al. [2002] for cracks parallel to the gradient
under monotonic loading, the current investigation enabled effects of crack deflection and
changing crack-tip environment to be observed simultaneously. Rousseau and Tippur
[2000] and Chapa-Cabrera [2002] previously observed crack deflection for cracks initially
oriented perpendicular to the gradient. Their specimens fast fractured, which prevented the
examination of effects of changing crack propagation resistance as the crack traversed the
gradient. Previous work on fatigue cracks in FGMs [Blumm et al., 1995, Forth et al., 2003,
Xu et al., 2003] has not addressed the issues of crack path and spatial variation in fatigue
crack propagation resistance to the same degree as in the present study.

Experimental investigation of crack initiation and propagation in graded alumina-epoxy

composite specimens under cyclic loading, along with parallel finite element simulations,
yielded the following findings:
10-38

Initial deflection angle and load amplitude values, and subsequent crack propagation
paths, were predicted accurately with FE simulations, using effective property values
obtained from homogeneous specimens in Chapter 9. This indicated that the FE
simulation methods and assumed material property distributions used were reasonably
accurate.

The combination of crack-extension and compositional variation effects led to cyclic

variations in effective crack propagation resistance across steps, though an overall
decrease was observed.

Material gradient influenced crack growth rate via (i) the concentration of stresses, due
to the elastic gradient, and (ii) the spatial variation in fatigue resistance. Together these
dictated the range of loads in which stable crack propagation would occur, and the
resultant crack growth rate.

The presence of interfaces within the graded region had a significant effect on crack
propagation behaviour. In some specimens, interfaces were observed to have a
significant influence on crack path, as it was more energetically favourable for the crack
to grow straight up the weaker interface. Subsequent deflection off the interface was
observed, indicating a close competition between the mixed-mode stresses acting to
draw the crack off the interface and the reduced fatigue crack propagation resistance of
the interface.

Bridging had a notable effect on crack propagation resistance, but not on crack path.

Comparison of crack propagation results under cyclic and monotonic loading indicated
little systematic difference in crack paths, though failure loads (and crack propagation
resistance) tended to be higher under monotonic loading, in concord with measurements
of homogeneous composites in Chapter 9.

10-39

11 Discussion
This chapter aims to integrate the various aspects of the investigation, and discuss the key
findings. The hypotheses in Chapter 3 are revisited and the key issues influencing crack
propagation under cyclic loading are addressed. Recommendations are made based on
these findings regarding methodologies for analysing and designing graded specimens to
optimise their fatigue resistance.

11.1 Hypotheses
The aim of the present study was to gain a fundamental understanding of crack
propagation in a graded region under cyclic loading, with the objective of developing
means of accurately prediction of crack paths and fatigue lifetimes for FGM components.
To this end, a number of hypotheses were proposed. This section discusses each of these
in relation to the findings of the present study.

1. Alumina-epoxy graded specimens may be used as a model system to examine the

influences of (i) elastic gradient and (ii) fatigue crack propagation resistance
gradient, on fatigue crack growth in graded materials.
The choice of material system, as discussed in Section 3.3 (p. 3-5), represented an
extreme case of an FGM system with very high elastic mismatch, and negligible thermal
mismatch and yielding. It is useful to have characterised this for comparison with other
types of FGM system, which may be more practical for engineering applications. The
alumina-epoxy system appears to have been an appropriate model system, despite some
processing issues and the brittle failure behaviour, which led to some experimental
challenges. The sample production process provided a reliable method for obtaining
connectivity of phases across the sample in both homogeneous and graded composites.

Crack deflection was observed in the graded alumina-epoxy specimens, in agreement

with FE predictions. The high elastic mismatch resulted in a high stiffness gradient over
relatively wide region, which provided the opportunity to observe the influence of the
gradient on crack propagation without significant influence of microstructural
heterogeneity, which would have been more dominant at smaller length scales.
11-1

Characterisation of fatigue behaviour of homogeneous composites revealed that fatigue

crack propagation resistance varied significantly across the pertinent composition range.
Crack-bridging behaviour under cyclic loading was quantified with the data adjustment
technique introduced in Section 9.4, which enabled the increase in fatigue crack
propagation resistance with crack extension to be visualised in a similar manner to Rcurves under monotonic loading. Fatigue crack propagation resistance in graded
specimens was observed to vary with crack extension, due to variations in both the
intrinsic (crack-tip) and extrinsic (crack-wake) contributions to fatigue resistance. The
fatigue data adjustment technique also proved most useful for deconvoluting the effects
of loading, material gradient and crack extension for cracks in graded specimens.

In summary then, the investigation of fatigue crack growth in graded alumina-epoxy

composites enabled insight into the effect of elastic gradient on stress concentration and
crack propagation path, and into the effect of fatigue resistance gradient on critical loads
for crack propagation. The results provide a benchmark for comparison with theoretical
models and other FGM systems.

2. Fatigue cracks will deflect due to elastic gradient and this deflection may be
predicted from stress field calculations and an appropriate deflection criterion.
Significant deflection was observed for fatigue cracks in alumina-epoxy FGMs in
Chapter 10, in concord with results from FE simulations in Chapter 6. In Section 10.6,
Moire interferometry revealed the asymmetrical displacement field around the crack-tip,
verifying FE predictions.

The initial deflection behaviour and crack path after deflection were dominated by the
elastic gradient in the graded alumina-epoxy composites studied. Accordingly, deflection
angles could be predicted very well using a finite element model with linear elastic
fracture mechanics analysis, as described in Chapter 5. That elastic gradient is the
dominant influence on crack path and stress concentration in alumina-epoxy FGMs is to
be expected considering the extreme elastic mismatch and the absence of residual stresses
and plastic yielding. In other composites where these are present, they may dominate

11-2

crack propagation behaviour and elastic gradient may have only a minor influence. This
was predicted for layered Cu/W specimens in Chapter 8.

While the possibility of toughness gradient or crack-bridging influencing crack-path

cannot be ruled out, they did not appear to influence fatigue crack path significantly, as
discussed below. FE predictions for crack path in Chapter 10 seem to provide a
maximum bound for the overall deflection. In the some of step-graded specimens
studied, toughness anisotropy at interfaces led to less overall deflection. If the anisotropy
were different, fault lines along the gradient direction for instance, then there may be
more deviation from straight-ahead propagation.

The effect of elastic gradient was twofold: crack deflection due to rotation of crack-tip
stress fields; and concentration of stresses relative to a similar crack in a homogeneous
material. FE simulations in Chapter 6 revealed that SIF concentration and mode-mixity
generally have similar trends with respect to gradient and crack parameters, though the
two parameters can exhibit different behaviours, especially for narrow gradients, and
approaching interfaces. The values of stress intensity factor and mode-mixity of a straight
crack in a graded region are influenced by crack position and length, and on the relative
steepness of material stiffness gradient both at the crack-tip and in close proximity to it.
Crack paths are influenced significantly by material gradient profile. Crack propagation
direction does not depend significantly on initial position within the gradient. These
findings were verified by experimental observations in Sections 10.1 to 10.4.

The good agreement between FE predictions and experimental results, for a range of
crack positions, lengths, specimen geometries and gradient widths, indicates that the FE
model provided a reasonably accurate simulation of the dominant influences of elastic
gradient on the crack propagation process. This implies that methods used for stress field
calculations and predictions of deflection angle were appropriate.

To summarise, deflection occurred due to elastic property gradient in the alumina-epoxy

FGMs studied and this was predicted well with the finite element model.

11-3

3. As a result of deflection, cracks will experience a variation in fatigue crack

propagation resistance as they propagate across the gradient.
Cracks were observed to propagate in a stable manner across the gradient in some
experimental specimens. Deflection was towards the more compliant epoxy-rich side,
which was understood, from results in Chapter 9, to have a lower fatigue crack
propagation resistance. An overall decrease in fatigue crack propagation resistance across
the gradient region was noted, with sharp drops at the interfaces between steps. This
agreed with the measurements of fatigue crack propagation resistance for homogeneous
composites of corresponding composition.

The combination of crack-extension and compositional variation effects led to cyclic

variations in effective crack propagation resistance across steps, though an overall
decrease was observed. Bridging zones could develop across several compositional steps,
as illustrated in Figure 11.1.

Step K

Crack-tip resistance
for composition K

Traction relation
for composition K

Step J

Traction relation
for composition J

End of
bridging zone

Interface
Figure 11.1: Schematic of a crack crossing an interface between compositional steps. The
bridging zone extends across both steps so different bridging traction relations will apply in
different sections. This could increase the magnitude of crack-extension effects.

This could result in crack-extension effects of greater magnitude than in homogeneous

composites of similar composition, due to the contributions of greater bridging tractions
in the more alumina-rich steps, in agreement with results in Sections 9.3 and 9.4. This

11-4

acted to reduce the magnitude of the abrupt drop in crack-tip fatigue crack propagation
resistance that occurred at the interface, and thereby stabilised crack growth.

The concentration of stresses due to elastic gradient and the spatial variation in fatigue
resistance were major influences on crack propagation rate and failure loads. At a given
point in the propagation of a crack, the range of loads for which further crack growth may
occur is dependent on:
(i)

intrinsic (crack-tip) fatigue crack propagation resistance, which depends on cracktip position;

(ii)

extrinsic (crack-wake) fatigue crack propagation resistance, which depends on the

location of the crack-wake zone; and

(iii)

stress concentration due to elastic gradient, which depends on crack position and
shape and elastic gradient profile.

Predictions of initiation load in Section 10.5, and of fatigue crack propagation resistance
profile in Section 10.4, agreed reasonably well with experimental observations, indicating
that these three factors were predicted reasonably well.

In conclusion, cracks experienced a significant variation in fatigue crack propagation

resistance as they propagated across the graded region. This resulted from spatial
variation in composition and crack-extension effects, due to crack-wake interactions.

4. Effective property relationships obtained for non-graded composites can be used

for graded composites.
The relation between composition and microstructure and effective thermomechanical
properties maps the compositional distribution in an FGM to the elastic property
distribution, which then dictates the distribution of stresses. The equivalent relation for
effective fatigue behaviour dictates the resistance to further propagation of the crack at a
given point in the gradient. These were shown in the framework in Section 3.4. The
effective elastic properties and fatigue behaviour were sufficient for the alumina-epoxy
IPN-structured composites studied, as they did not exhibit thermal stresses or yielding.
For other systems, effective yielding and thermal property relations may be required.

11-5

The effective Youngs modulus and other elastic properties showed good agreement with
EMA predictions, in Section 9.1. The sharp drop-off in stiffness with increasing epoxy
content agreed with predictions of Wegner and Gibson for co-continuous composites
with internal geometry similar to the alumina-epoxy composites. The success of the EMA
was attributed to the non-specific geometry assumed by this model, and it was shown in
Section 9.2 to be applicable to other IPN structured composites. Several of the effective
property models reviewed in Section 2.2 were shown to be inappropriate for the aluminaepoxy composites studied, though they may be more applicable for other types of
composites. This highlighted the importance of experimental characterisation of elastic
properties for several compositions across the relevant range. The influence of phase
continuity at step interfaces on the effective properties in that region, and the possibility
of smoothing of the compositional steps was introduced in Section 2.2.6 and discussed
further in Sections 4.2 and 5.1. It was concluded that a degree of smoothing must be
appropriate otherwise calculated results become erratic at step-interfaces.

Fatigue crack growth behaviour in the alumina-epoxy composites was fairly

characteristic of a brittle material, though crack-extension effects were observed.
Initiation and propagation were strongly influenced by composition. Higher epoxy
content resulted in increased probability of flaw initiation, and reductions in intrinsic and
extrinsic contributions to fatigue crack propagation resistance. The crack-extension
effect, probably due to bridging, resulted in increased resistance to fatigue crack
propagation. As discussed in Section 2.5.5, this is generally difficult to quantify for
fatigue, though a novel approach for adjusting fatigue results introduced in Section 9.4
proved very useful in this regard. In graded specimens, crack-extension effects could
result from bridging within a region of varying composition, ie. over several
compositional steps as in Figure 11.1, which would complicate matters further.

The good agreement, between predicted and measured values for initiation loads and
deflection angles in Section 10.1, indicates that the assumed property distributions,
obtained from homogeneous composite results in Chapter 9, were satisfactory.

11-6

Accordingly, it was concluded that the same effective property relationships apply in
homogeneous and graded composites.

5. Bridging influences crack path and effective resistance to crack growth

Bridging of the crack by intact ligaments of a reinforcing phase could have several effects
on crack propagation. It was speculated that altering the geometry and/or properties of
bridging-phase ligaments could influence crack deflection path and effective fatigue
crack propagation resistance.

Firstly, the bridging will shield the crack-tip, reducing the crack growth rate at a given
load and thereby increasing the effective crack propagation resistance. This would have
an influence under both monotonic and cyclic loading. In Chapter 9, crack-extension
toughening effects in homogeneous specimens, observed under monotonic loading, were
also detected under cyclic loading by adjusting SIF amplitude values. This approach was
applied to graded specimens in Chapter 10. For both homogeneous and graded aluminaepoxy composites, bridging was understood to influence the effective toughness and
crack propagation resistance. Furthermore, in graded specimens the presence of a
bridging zone was understood to smooth the sharp change in fatigue-crack propagation
resistance experienced by the crack as it crossed interfaces between compositional steps,
as in Figure 11.1. The effects of bridging on effective toughness in homogeneous and
graded composites were addressed in Chapters 9 and 10 respectively. These were shown
to be significant, and beneficial in improving resistance to failure.

Secondly, it is possible that, as the crack undergoes deflection and curvature due to
mixed-mode loading, the bridging tractions could modify the mode-mixity at the cracktip and thereby influence crack path. This issue was considered in depth in Chapter 8. FE
simulations and a simple analytical model were presented to show that bridging could
influence crack-tip mode-mixity significantly, depending on the shielding tractions acting
in each mode. This influence was shown by a simple model to vary with characteristics of
the bridges such as width and modulus.

11-7

An argument was presented in Section 8.4, that bridging should not have a major effect
on overall crack path. Even if bridging modifies crack-tip mode-mixity, the crack should
eventually propagate normal to the far-field loading direction. It was noted that bridging
could have a transient effect, impeding the deflection of the crack toward the far-field
propagation direction. Due to the sensitivity of crack paths in the gradient region to
perturbation, as observed in earlier simulations in Section 6.2, this slight effect of
bridging could lead to systematic deviation of crack trajectories. This was demonstrated
via simple FE simulations in which SIF in one or both modes was artificially suppressed
to examine effect on crack-path, as shown in Figure 8.17. It was also shown that if
bridging acts to increase mode-mixity, this would amplify any instability in the crack
path that may arise due to T-stresses.

Whilst bridging had a notable effect on crack propagation resistance in the experiments,
an effect on crack path was not discernible. This was further verified through the
comparison of experimental results for crack propagation under cyclic loading with those
for propagation under monotonic loading in Section 10.7. While crack-extension effects
were observed to differ significantly between these cases, indicating that bridging
tractions were different, crack paths were very similar.

It must be stressed, however, that the crack-tip mode-mixity is essentially determined by

applied mode-mixity and overall crack path is dictated by the elastic gradient. Bridging
can cause transient effects that lead to slight perturbations in crack path that, due to the
interactions between crack shape and material gradient, can lead to some deviation of
crack paths. Bridging could modify the path slightly, but it certainly could not cause the
crack to change direction and start deflecting towards the other side. For this reason,
attempts to engineer the reinforcement phase to influence propagation path are likely to
meet with limited success. The main consideration with regards to bridging should be its
potentially beneficial influence on effective crack propagation resistance.

11-8

6. Spatial variation in toughness and fatigue crack propagation resistance can have
an influence on crack propagation path, as propagation through material with
lower resistance to cracking is more energetically favourable.
The possibility that spatial variation in crack growth resistance could influence crack path
was considered. In this case, the path that maximises applied mechanical energy release
rate, G, and that which maximises (G-Gc), where Gc is the critical energy release rate,
may differ.

In Section 5.3, it was shown that the path that maximises G is very similar to that
predicted by the MTS and local symmetry criteria. In Section 6.4, the effects of
toughness variation on propagation direction were shown to be very minor, as the
variation of G with deflection angle is much more significant than the corresponding
variation of Gc. In the limit of an infinitely small test-kink, the effect of variation in Gc
will go to zero. Accordingly, the path that maximises G, and that which maximises (GGc), will be very similar, and will be in agreement with the predictions of the local
symmetry criterion. Indeed, experimental results in Section 10.2 indicated that the crack
does follow such a path. It is conceivable that at step-interfaces, spatial toughness
variation could cause slight perturbations in path that could lead to further deviation, due
to the tendency of crack paths to bifurcate as discussed above with regards to bridging.
This would only be a minor effect.

Whilst toughness gradient effects could conceivably influence crack path slightly, this
would occur at a small scale, at which microstructural heterogeneity is a dominant
influence on crack path. The effects of local heterogeneity on crack-path tortuosity were
observed in the alumina-epoxy composites, in Sections 9.3 and 10.2. These are likely to
override any effects of toughness gradient at the microstructural scale.

An exception to this, as discussed in Section 6.4, may occur if the crack propagation
direction is determined by the maximum stress some distance away from the crack-tip. In
this case, the use of a deflection criterion based on stress distribution rather than on
mechanical energy release rate would be expedient. This accords with the findings of

11-9

Becker et al. [2002]. In this case, higher-order terms in the stress field may need to be
taken into account.

Only in the presence of toughness anisotropy will there by a discernible influence of

toughness on crack path, as discussed in Chapter 10.2. This was observed for cracks at
weak interfaces. Upon reaching the interface, the crack would deflect away from the
propagation direction which maximised applied mechanical energy release rate, G, and
would travel up the interface, which had a lower critical energy release rate, Gc. In these
cases, it was seen that the path that maximised (G-Gc) was followed.
In summary, spatial toughness (or fatigue resistance) variation does not have a
significant effect on crack path, although at interfaces, toughness anisotropy can have a
major effect. In both cases, cracks follow the path for which (G-Gc) is maximised.

7. After crack deflection, the crack-shape has an effect on crack propagation

direction and on stress concentration at the crack-tip.
In Chapter 7, curvature was shown to have an effect on subsequent crack path and on
stress concentration. This was more significant in bimaterial and graded specimens than
in homogeneous ones, due to the interactions between crack-shape and material property
variation. Comparing the results in Sections 6.3 and 7.5, the crack path itself appears to
be less important, in terms of total structural integrity, than the location of the crack-tip
within the gradient. The latter has a more significant effect in terms of stress
concentration, and resistance to this stress concentration, than the effect of crack shape,
particularly in the vicinity of interfaces.

The issue of crack shape approximation was addressed in Chapter 7 via a simple
analytical model for cracks in homogeneous materials, and FE simulations of crack
propagation in homogeneous, layered and graded materials. Curved cracks in
homogeneous materials may be approximated by equivalent straight cracks for the
calculation of mechanical energy release rates and deflection angles. This was understood
in terms of the volume of material around the crack from which strain energy will be

11-10

released during crack propagation. Cracks in graded materials may also be reasonably
approximated by equivalent straight cracks also, although there is systematic cumulative
divergence between simulations using the entire crack shape and those using
approximated crack shapes. Piecewise-linear crack shapes provided a significantly better
approximation than diagonal or straight crack shapes. Accordingly, analytical solutions
for piecewise-linear cracks in graded materials would be a useful area for further work.

8. Crack propagation behaviour in stepped gradients will differ from that in

continuous gradients.
Finite element simulations in Chapter 6 showed that the profiles of stress intensity factor
and mode-mixity values differ significantly with crack propagation between continuous
and stepped gradients. The latter exhibit fluctuations in both parameters as the crack
position varies from the middle of steps to near the interfaces between steps. The
fluctuations in mode-mixity tend to average out over propagation, however, so crack
paths in stepped and continuous gradients are very similar. Steps were not observed to
significantly influence crack paths in experimental specimens, except when the interface
between compositional steps was weaker, as discussed below.

The SIF magnitude also exhibits fluctuations across a stepped gradient. For a crack on the
right-hand side (stiffer) of a step-interface, elastic anti-shielding will result in additional
stress concentration compared with a continuous gradient, whereas elastic shielding
reduces stress concentration for cracks on the more compliant side of a step-interface. As
the crack crosses the interface, the SIF magnitude will usually drop, though this is
preceded by an increase as the crack approaches the interface. This is in concord with the
findings of He and Hutchinson [1989] and Fleck et al. [1991] for cracks near interfaces.

Fluctuations in stress concentration represent a compromise in reliability of the overall

structure due to the additional variability introduced. For bigger (i.e. less) steps, there will
be more variability in stress concentration and more chance of a sudden catastrophic
failure. On the other hand, the fluctuations could mean that a crack that extends near a
step interface will then arrest in the next step, which could be beneficial.

11-11

For the stepped-gradient specimens studied, there was an abrupt drop in intrinsic (cracktip) toughness and fatigue crack propagation resistance as cracks crossed the interfaces.
This compares with the smooth decrease expected in a continuous gradient. As discussed
above (Figure 11.1), crack-extension effects can reduce this somewhat as these do not
change as abruptly. In some cases, however, the sudden change led to fast fracture. This
was discussed above in response to Hypothesis 3. Taking both SIF magnitude and crack
growth resistance into account, as in Section 6.3, the presence of steps tends to decrease
the critical load when the crack-tip is near step-interfaces. However, the subsequent
increase in critical load as the crack moves past the interface is likely to cause crack
arrest, so the structure could undergo graceful rather than catastrophic failure.

Toughness anisotropy, e.g. weakness at interfaces between compositional steps, can have
a significant effect on crack path. For lower deflection angles (<20), only a slight
(<10%) reduction in toughness along the interface is sufficient to cause the crack to
propagate along it rather than continue to traverse the gradient. This was observed in
experiments, as was initiation from interfaces.

An issue, which was identified but could not be fully investigated within the scope of the
current study, was that of representative volume element size. In a graded material with
compositional steps, the effect of averaging over finite region could lead to a smoothing
of effective property gradient. This was shown to have a significant effect on stress field
calculations for cracks near to the interfaces. The influence on apparent SIF profiles
suggested that some smoothing in the assumed effective property distribution would be
appropriate. This is a possible area for further work in the future.

9. Crack propagation under cyclic loading will differ from that under monotonic
loading.
Differences between cracks in FGMs under monotonic and cyclic loading could arise for
several reasons: different crack-tip processes, different extrinsic mechanisms or nonlinear behaviour away from the crack-tip.

11-12

Crack-tip advance mechanisms in the alumina-epoxy composites studied did not appear
to differ between monotonic and cyclic loading, although crack-wake interactions were
understood to degrade under cyclic loading. Comparison of crack propagation results in
Section 10.7 indicated little systematic difference in crack paths, though failure loads and
crack propagation resistance tended to be higher under monotonic loading, particularly
for more epoxy-rich compositions. This was understood to result from degradation of the
bridging zone under cyclic loading, in agreement with measurements of homogeneous
composites presented in Chapter 9. Despite this influence on the bridging zone, there was
no discernible effect on crack paths. As discussed above, this provided strong evidence
that crack-bridging does not influence crack path appreciably in these composites. For
materials with significant non-linear behaviour, ie. crack-tip SIF prior to loading due to
residual stresses, or plastic mismatch across the gradient, the differences between cyclic
and monotonic loading are likely to be more important.

10. The fatigue crack propagation process can be modelled, taking the various
influences into account, and providing accurate predictions for crack
propagation path and failure loads.
The use of finite element modeling, and in particular the use of ANSYS, was very
appropriate due to its flexibility in modeling arbitrary specimen, gradients and cracks,
and the fact that models could be developed, modified and adapted throughout the study.

Throughout the FE simulations, it was demonstrated that simplified approaches generally

gave very similar predictions to more rigorous models, on issues of crack shape,
deflection criteria and fracture parameter calculations. Section 5.3 showed that the
calculation of stress intensity factors with the displacement correlation technique is
convenient and appropriately accurate, as compared with results from other approaches
and analytical solutions. The local symmetry, or kII = 0, criterion for predicting crack
deflection, with a test-kink, is consistent with other criteria. Any variations between
different criteria appear to average out over several increments of crack propagation. The
variation of mode II SIF at the kink-tip, kII, with kink angle, , was shown in Section 5.4
to be predominantly linear, which enabled the use of a more efficient algorithm for

11-13

predicting deflection angle. Although test-kink size influences predictions of deflection

angle, this influence is averaged out after several increments of crack propagation so that
predicted crack paths do not depend on test-kink size. Hence, it is not crucial to use
extremely small kinks. Furthermore, crack extension increment size does not influence
crack path predictions, within the range of increment sizes examined (2% to 10% of
original crack length).

The connection between experimental work and finite element modeling in the present
study enabled direct comparison and verification of the FE simulation methodology and
analysis of experimental data. This has been achieved previously [Rousseau & Tippur,
2000, Chapa-Cabrera, 2002], though these researchers only modelled cracks before
deflection. In contrast, cracks were successfully simulated and analysed throughout
propagation in the present study. The use of finite element modeling, as opposed to
theoretical models such as those of Erdogan and co-workers [1995], Gu and Asaro [1997]
and Jin and Batra [1998], was necessary for simulating cracks of arbitrary shape in finite
specimen geometries with real, rather than idealised, material property gradients. In
general, the findings from the present study are in good qualitative agreement with those
from theoretical models; however such models could not be used for comparison with
real experimental specimens. Kim and Paulino [2004] have also simulated crack
propagation in FGMs, utilising more rigorous computational methodologies. It may be
beneficial in the future to compare predictions from their model and that developed in the
present study.

11.2 Other FGM systems

Some discretion is required in extrapolating the findings obtained for the model system
examined in the present study to other types of FGM system. By extending the FE
simulations beyond the issues pertinent to the alumina-epoxy system, various aspects
were examined relating to other types of FGM systems. This enabled the findings of the
present study to be more confidently applied to FGMs in general. In particular, the effects
of plastic compliance mismatch, thermal residual stresses and process zone effects were
examined in Chapter 8. This section outlines areas in which the behaviour of other FGM

11-14

systems may differ from the alumina-epoxy system, and suggests methods for
incorporating these differences into analyses.

11.2.1 Thermomechanical property gradient

Crack deflection and stress concentration can occur due to mismatch in any one or more
of elastic, plastic and thermal properties. Residual stresses and plasticity were understood
to not be present in the alumina-epoxy specimens, as the infiltration of the epoxy was
conducted at room temperature and both phases are brittle. Nonetheless, these issues were
investigated with FE simulations in Sections 8.5 to 8.8. It was shown that reasonable
predictions may be obtained for crack-tip stresses and critical loads for cracks under the
influence of plastic and thermal mismatch.

Other FGM systems may not have as significant elastic mismatch between constituent
phases as the alumina-epoxy system, as illustrated in Section 6.1. However, the gradient
region may be much narrower so the stiffness gradient would have a similar magnitude to
the specimens in the present study. Mismatch in plastic yielding, as it results in
compliance mismatch, can have essentially the same effect as elastic mismatch on stress
concentration and mode-mixity. As shown in Section 8.7, the occurrence of plastic flow
in the more ductile region tends to increase stress intensity factors and cause deflection
towards that region, due to enhanced compliance. The presence of plastic yielding can
introduce three-dimensional constraint effects which must be included in analyses
[Chapa-Cabrera, 2002].

Section 8.6 showed that residual stresses, due to thermal expansion coefficient mismatch,
can have an important influence on crack-tip stresses and propagation paths, particularly
in metal-ceramic systems. Accordingly, it is important to characterise the residual-stress
distribution in the graded region to ascertain whether it is likely to have a deleterious
effect on structural reliability. The influence of residual stress tends to be more
complicated for materials with plasticity than for purely elastic materials. In metalceramic FGMs, the more compliant metal region is likely to have a higher thermal
expansion coefficient, and therefore experience tensile thermal stresses. In this case,

11-15

plastic mismatch and residual stresses may have opposing effects on crack path.
Furthermore, elastic and thermal mismatch can influence plastic deformation, and thereby
amplify plastic mismatch effects. Interactions between these may be complicated and
should be quantified with experiments and FE simulations wherever possible.

11.2.2 Crack propagation resistance gradient

Other materials could exhibit significantly different fatigue behaviours from those
studied. These could involve, for example, more ductility, a diffuse damage zone or
transformation toughening. More ductility would lead to a wider range of loads for which
crack propagation would occur, which could lead to more complicated overall behaviour.
Taking such behaviour into account in a model would require experimental
characterisation and corresponding modifications to the simulation code, in terms of
material stress-strain response and crack growth and deflection criteria. As shown in
Chapter 8, plastic deformation during earlier crack propagation can lead to R-curve
effects, whereby effective critical stress intensity factor increases due to accumulation of
plastic strains adjacent to the crack-tip. This corresponds to an increase in effective
toughness due to the work absorbed during plastic deformation. Accordingly, considering
the effects of prior strain history on crack-tip stresses in simulations is important.

As discussed in Chapter 6, the gradient in crack-propagation resistance is unlikely to

influence crack path, except in materials where crack extension depends on stresses a
certain distance away from the crack-tip, as in the statistical fracture model of Becker et
al. [2002], or materials with toughness anisotropy.

11.2.3 Effective property relations

The effective property relations required for analysing a particular FGM system will
depend on the factors influencing failure behaviour: FGMs with plastic yielding and
thermal stresses will require relationships for composition-dependence of yield stress,
non-linear stress-strain behaviour and thermal expansion coefficient. Relations will
depend on composite internal geometry, and this may vary across the gradient region.
Effective time-dependent properties, such as viscoelasticity, viscoplasticity and creep
behaviour or constitutive damage relations (as in Section 2.2.7) may be required also.
11-16

11.2.4 Crack-extension effects

These will depend on the composite structure and the failure behaviour of each phase. As
most FGMs are composites, it is likely that some crack-bridging will occur, especially in
those exhibiting interpenetrating-network structures. Of particular importance is the
effect of ductile phase reinforcement on fatigue crack propagation, as in aluminaaluminium FGMs [Moon, 2004]. This may require a more detailed model for crackbridging, than that in the present study, to take the progressive degradation of bridges into
account. Correspondingly, more detailed experimental characterisation of both
homogeneous and graded specimens under a range of loadings would be required to
understand the cyclic degradation process.

Phase-transformation, microcracking and fatigue crack closure can also lead to crackextension effects, and could be included into models in a similar manner to bridging. The
R-curve behaviour and, in the case of bridging, the underlying traction relation should be
quantified as a function of composition under both monotonic and cyclic loading.
Bridging or other process-zone effects are unlikely to have a major influence on crack
path, as discussed in Section 8.4, rather they may modify the effects of thermomechanical
property gradient on propagation direction. In some cases, the scale of the process or
crack-wake zone may be greater compared with the scale of the gradient, so that crackextension effects would become more complicated as the crack-wake zone develops
across a large portion of the gradient. In other cases, the scale of the crack-wake zone
may be very small compared to the gradient, so the effective toughness is dependent on
crack-tip position only.

11.2.5 Crack path & shape

Crack deflection and the resultant non-straight crack path are likely to be important issues
for cracks in many different types of FGMs. The choice of deflection criteria for
predicting crack path may be dependent on crack-tip plasticity and process zone or
anisotropy effects may also influence crack propagation direction. The findings in the
present study indicate that crack shape should be taken into account in simulations. When
deflection angles are low (<10), the use of a straight-crack approximation rather than
11-17

exact crack path would probably be acceptable. In some systems, the crack-front may not
remain straight through the thickness of the specimen and three-dimensional effects may
be relevant, as shown by Jin and Dodds [2004].

11.2.6 Interfaces
It is likely that many experimentally or industrially produced FGM components will not
be continuously graded, rather they will contain discrete interfaces between sections of
different composition, as considered in present study. Depending on composite structure
and connectivity across interfaces, smoothing of effective property profiles at the
interface may be applicable. Toughness anisotropy at the interfaces could affect crack
path. Interfaces are unlikely to have other effects on crack path, as discussed in Section
11.1.8. On the other hand, the effects of interfaces on fluctuations in stress concentration
cannot be ignored. The intersections of interfaces with a free edge represent probable
locations for crack initiation due to stress peaks there, along with the possibility of
microstructural weakness and manufacturing flaws.

11.2.7 Monotonic & cyclic loading

For materials with significant non-linear behaviour, the differences between cyclic and
monotonic loading are likely to be more important. One example is the Cu/W layered
specimens in which crack-tip mode-mixity varies with applied load due to the presence of
residual stresses, as discussed in Chapter 8. Another example is graded aluminaaluminium IPN composites in which cyclic loading of intact aluminium ligaments is
understood to lead to progressive degradation of bridging tractions [Moon, 2004].

The effects of cyclic loading will depend on:

(i)

Crack-tip fatigue behaviour, especially plastic yielding.

(ii)

Cyclic deformation behaviour of bridging zone or process zone.

(iii)

Irreversible or nonlinear deformation away from the crack-tip (as in Section

8.7.3), which could lead to progressive redistribution of stresses in the structure.

(iv)

Time dependent deformation ie. viscoelastic or viscoplastic behaviour, which can

lead to frequency dependence

11-18

It should be remembered that, if one material is tougher, it may be less resistant to

fatigue, as the deformation/energy absorption mechanisms that improve toughness often
enable damage to accumulate at subcritical loadings [Ritchie, 1999].

11.2.8 Modeling of crack propagation

The present study has demonstrated that simulations of crack propagation in FGMs are
important for investigating fatigue performance. Simulations for a given FGM system
must take into account all the issues discussed above. The general framework presented
in Section 3.4 may be applied to any type of graded material system. The present study
demonstrated the applications of this framework to a particular system. The specific
details, such as choice of mixing laws, crack growth criteria and relations, calculation
methods, will vary between different systems.

The prediction of fatigue behaviour and lifetime for FGMs in general requires an
understanding of the effective mechanical properties and fatigue crack growth rate
relations for the composite system in question. Once these are established, the mechanical
and fatigue aspects of the problem may be separated. A mechanics analysis (using FEA,
for example) should be used to determine the crack-tip stresses under loading, accounting
for elastic, thermal and plastic mismatches, and predict the crack path and corresponding
K/P (or G/P2) profile. Fatigue crack growth relations should then be used, along with K/P
values, to calculate crack growth rate.

FE simulation was demonstrated in the present study to be an appropriately flexible and

accurate method for calculating stress and displacement fields in FGMs. The choice of
methods for calculation of fracture parameters (ie. SIFs or J-integral) and prediction of
propagation direction will depend on the extent of crack-tip plasticity. For more
significant plastic yielding, the J-integral and a modified deflection criterion should be
used.

Cohesive zone modeling (CZM), discussed briefly in Section 2.4, has been receiving
significant attention for modeling fracture and fatigue crack propagation [Yang et al.,

11-19

1999, De Borst et al., 2004]. To date, work on CZM of crack propagation under mixedmode loading has been very limited. Whilst it has been used to simulate crack
propagation parallel to the gradient in FGMs [Jin & Dodds, 2004], it has not been applied
to propagation in other directions. The FE simulation approach demonstrated here could
be extended to incorporate cohesive zone elements; this could be an interesting area for
future work.

11.2.9 Analysis of FGMs

The present study has provided useful insight into the analysis of graded material
systems, in terms of their mechanical and fatigue performance. The following general
methodology for analysing an FGM structure is suggested:
1. Establish requirements for application, or context/objectives of investigation. Identify
processing issues, loading configuration, probable load spectra, operating
environment, constraints on constituent materials, composite internal geometry,
specimen and gradient geometry.
2. Measure composite elastic, thermal and plastic properties for several compositions
across the relevant range.
3. Measure composite fracture and fatigue behaviour for several compositions. Obtain
crack initiation toughness, fatigue crack propagation rate relations, R-curve behaviour
(if present) under monotonic and cyclic loading.
4. Conduct thermoelastic-plastic stress/deformation analyses under expected loadings
without cracks, and obtain stress fields.
5. Conduct thermoelastic-plastic stress/deformation analyses under expected loadings
with cracks at likely initiation locations, to obtain stress intensity factors and
determine likely propagation direction.

11-20

11.3 Optimisation of graded interfaces

As functionally graded interfaces may potentially be employed in a range of engineering
components, the question of optimisation for particular applications arises. The benefit of
a gradient region, over a bimaterial interface, in terms of structural integrity was
discussed at length in Chapter 2. However, there has not been a great deal of work on
how precisely to optimise the performance of FGMs. Work in this area has been directed
towards optimising a particular parameter such as peak stress [Raob et al., 1997,
Shimojima et al., 1999, Afsar & Sekine, 2001] or optimising performance in a particular
application [Bahr et al., 2003]. The deflection of cracks from the weak interface region
into the tougher material has been suggested as a mechanism for strengthening joints
between different materials [Hoffman et al., 2001, Chapa-Cabrera & Reimanis, 2002a,b].

A comprehensive model for design against failure would require a detailed knowledge of
the crack-tip toughness, and extrinsic toughening mechanisms within each of the phases
and the composite region. Crack initiation will depend on stress concentration, toughness
variation and defect distributions, hence an analysis of stress and predicted damage
accumulation within an initially uncracked component would be expedient. Variation of
effective toughness with crack extension, including R-curve effects, should be included
in considerations. The optimal gradient configuration will depend on the composite
system in question, on the mechanical and fatigue properties of each constituent phase
and effective properties across the composition range.

The process of crack propagation in graded materials is complex, with numerous

contributing influences that will differ between material systems. It is difficult, therefore,
to put forward generic recommendations regarding the design of optimal FGM structures
for resistance to fatigue damage. It has been shown that the general behaviour of cracks
in a graded structure is dictated by the materials on either side. This limits the extent to
which such a structure may be optimised. Rather, the general behaviour should be
identified based on the elastic, thermal, yielding and failure properties of each material,
then a local optimisation strategy be adopted.

11-21

The optimum design will obviously depend on the intended application, and the expected
operating conditions. Considering a joint or interface between two given materials with
given properties, as in Figure 11.2, the present discussion is restricted to structural
applications under cyclic loading in a relatively non-aggressive environment. This would
also be applicable to monotonic loading situations, to an extent. The optimisation
methodology could possibly be extended or adapted for other types of application and
conditions.

E2, Kc2
2, y2

E1, Kc1
1, y1
n

Stepped or
Continuous?
Where will initiation occur?

Which way will

the crack grow?

Figure 11.2: Joining two different materials: how can the interface region be optimised?

If a graded region is used at the joint between two different materials, the properties of
these materials are likely to be fixed, as shown in Figure 11.2. This leaves several
controllable parameters that may be varied in the optimisation process:

Gradient width

Gradient shape: profile exponent; continuous or stepped; number of steps

Composite internal geometry - shape and scale

The major issues to be considered are:

1. Where are cracks likely to start?
This will occur where the stresses are sufficient for the initiation of microscopic flaws, in
ductile materials, or the extension of existing flaws, in brittle and composite materials.
The latter is more likely in FGMs as they often have at least one brittle constituent phase
and are usually composites. The location of initiation will depend on (i) spatial variation
11-22

of fatigue resistance, which depends on composition and microstructure, (ii) distribution

of flaws or damage that may result from manufacturing and (iii) distribution of stresses
resulting from thermomechanical property mismatch. The presence of interfaces between
compositional steps could also provide likely initiation points due to stress concentration
and possible weaknesses there.

2. Which way is the crack likely to go? How will this affect crack propagation
resistance?
The combination of elastic, plastic and thermal mismatch effects, and possibly toughness
anisotropy, will determine the overall crack deflection behaviour for the FGM. The
resistance to fatigue crack propagation may be higher in one material or the other or,
possibly, in the graded region, if crack-bridging or diffuse damage leads to significant
energy absorption. As discussed above, if one material is tougher, it may be less resistant
to fatigue, as the deformation and energy absorption mechanisms that improve toughness
often enable damage to accumulate at subcritical loadings.

It was shown in Chapter 6 that variations in gradient profile and continuity (ie. stepped or
continuous) did not influence crack propagation paths greatly, which limits their use as
optimisation variables. Similarly, Section 8.4 demonstrated that crack-bridging would
have, at most, a secondary influence on path. The gradient width controls the steepness of
the gradient, and therefore the angle of deflection. A narrower gradient may be used to
promote deflection though this will result in higher stress concentrations for cracks in that
region.

There are essentially three avenues that may be taken for optimisation, depending on the
whether the side towards which the crack will deflect is the more or less tough side:
1. One side of the gradient is much tougher, and it is possible to divert the crack to
this side. The crack should be diverted to that side as quickly as possible, whilst
keeping stress concentration below an acceptable level. This is probably the most
likely configuration for FGMs in applications, e.g. a graded joint between metal and
ceramic. Optimisation of this mechanism would require a balance between an

11-23

increase in mode-mixity, , and a decrease in the normalised stress intensity factor,

K/P.

2. The composite region offers improvements, in terms of fatigue resistance, over

either constituent material. This situation is conceivable although due to it being a
composite, it is likely to be susceptible to fatigue degradation under cyclic loading. In
this case, deflection should be reduced, and the crack-tip loading minimised.
Increasing the width of the graded composite region will reduce stress concentration
and less deflection, and will provide more tough material for the crack to propagate
through. Also, the microstructure in this region should be tailored to maximise
improvements in crack propagation resistance.

3. Both sides are brittle, and the composite region does not offer significant
toughening. A risk reduction approach is required, based on careful manipulation of
gradient variables, predominantly focussed on minimising K/P. The width should be
increased to reduce stress concentration, and a higher value of gradient profile
exponent.

If a stepped gradient is necessary, then weakness at the step-interfaces should avoided as

this can alter propagation behaviour. More steps would be favourable as the variability in
the stress concentration profile, and corresponding likelihood of catastrophic failure,
would be reduced. Steps of different widths could be useful; more steps would be used in
the regions where interfaces are likely to cause most stress fluctuations, whilst less steps
would be used in the regions where they are not.

In the case of alumina-epoxy composites, for example, the concentration of stresses at the
crack-tip (|K|/P) can be very large due to elastic mismatch. Deflection is not helpful due
to the poorer fatigue resistance on the more compliant side. Hence the optimisation
strategy should focus on reducing |K|/P. To achieve this, the width should be increased,
and a higher gradient profile exponent, ie. n = 2, should be used. From the FE simulations
in Chapter 6, the optimum gradient would appear to be one with width of 10 mm, profile

11-24

around n = 2 and continuous or stepped with more than 5 steps, though these are
approximate values only. While a stepped gradient will not affect the average crack-tip
stress concentration significantly, it does reduce overall reliability by increasing the
variability in |K|/P and the likelihood of catastrophic failure from a step-interface.

The following methodology for overall optimisation of the fatigue resistance performance
of an FGM structure is suggested:
1. Establish requirements and limitations pertinent to application. Processing
practicalities, loading configuration, load spectra, environment, constituent phases,
composite internal geometry, etc.
2. Measure/obtain composite elastic, thermal, plastic, fracture and fatigue properties for
several compositions across the relevant range.
3. Adopt general strategy from those listed above.
4. For a number of gradients based on the adopted strategy, conduct thermoelasticplastic stress/deformation analysis under expected loadings without cracks to obtain
stress fields, and establish likely locations of failure initiation. Discard the gradient
profiles which result in unacceptably high likelihood of failure initiation.
5. For the remaining gradients, with cracks at a number of possible initiation locations,
conduct thermoelastic-plastic stress/deformation analysis under expected loadings to
obtain stress intensity factors and expected propagation directions.

There may be a compromise between average performance and reliability, i.e. the
average failure load as compared with the number failing below a given minimum
threshold load. The compromise reached will depend on the application. In practice, other
issues such as manufacturing ease and cost will also be taken into consideration but these
were outside the scope of the present study.

11-25

12 Conclusions
Crack propagation under monotonic and cyclic loading was investigated in graded
alumina/epoxy composites. FE simulations of propagation in graded specimens were
conducted, employing measured properties of homogeneous specimens, along with
simulations to investigate various aspects of the crack propagation process. The
conclusions reached may be divided into several groups as follows.

Firstly, several generic conclusions, which pertain not only to graded materials:
1. Effective elastic properties for interpenetrating-network structured composites may be
predicted accurately with the effective medium approximation (EMA) method.
2. Crack-extension toughening effects in homogeneous specimens, observed under
monotonic loading, may also be detected under cyclic loading by adjusting SIF
amplitude values with respect to resultant crack-growth rates.
3. The relationship between mode II SIF, kII, and kink angle for a small kink extending
from an existing crack is predominantly linear. This enables the use of a more
efficient algorithm for predicting deflection angle.
4. Non-straight cracks in homogeneous materials may be approximated by equivalent
straight cracks for calculations of mechanical energy release rate and predicting
deflection angle.

A number of conclusions related to simulation and analysis methods for graded materials:
5. The finite element model developed in the present study was adequate for predicting
critical loads and crack propagation directions in graded specimens, despite
employing more approximate methods than some more rigorous models in the
literature. Crack paths may be predicted reasonably well through FE simulations,
under linear-elastic conditions. Observed disparity was attributed to toughness
anisotropy.
6. Correlation of nodal displacements near the crack-tip may be appropriately used to
calculate stress intensity factors for elastic FGMs, and also for elastic-plastic FGMs
in the case of small scale yielding.

7. The local symmetry, or kII = 0, criterion for predicting crack deflection under elastic,
or small-scale yielding, conditions provides predictions consistent with other criteria.
Any variations between different criteria appear to average out over several
increments of crack propagation.
8. Although test-kink size does influence predictions of deflection angle, this influence
appears to be averaged out after several increments of crack propagation so that
predicted crack paths do not depend on test-kink size. Hence, it is not crucial to use
extremely small kinks. Furthermore, the size of crack-extension increments does not
influence crack path predictions, within the range of increment sizes examined (2% to
10% of original crack length).
9. Cracks in graded materials may also be reasonably approximated by equivalent
straight cracks for calculation of mechanical energy release rate and propagation
direction. However, systematic divergence occurs between simulations using the
entire crack shape and those using approximated crack shapes. Piecewise-linear crack
shapes provide a significantly better approximation than diagonal or straight cracks.
Accordingly, analytical solutions for stress intensity factors of piecewise-linear cracks
in graded materials would be a useful addition to the existing theory, and should be a
focus of further work in this area.
10. The fatigue data adjustment procedure, proposed for homogeneous composites with
crack-extension effects under cyclic loading, enables effects of material gradient,
applied load and crack extension to be deconvoluted in results for graded specimens
under cyclic loading.

Most importantly, there were conclusions regarding the actual crack propagation
behaviour of graded materials under cyclic loading:
11. The initial deflection behaviour of cracks in the graded alumina-epoxy composites
studied was dominated by the elastic gradient. Accordingly, deflection angles could
be predicted very well using a finite element model with linear elastic fracture
mechanics analysis. Crack path after deflection was also dominated by the elastic
gradient. While the possibility of crack-bridging influencing crack-path cannot be
ruled out, it does not appear to be a major influence.

12. As cracks propagated through the graded region, they experienced a variation in
composition at the crack-tip. This led to decreases in crack-tip toughness, which
occurred suddenly at interfaces between steps, and a decrease in crack-extension
effects, in agreement with results from homogeneous composites. The combination of
crack-extension and compositional variation effects led to cyclic variations in
effective crack propagation resistance across steps, though an overall decrease was
observed.
13. Crack shape has an effect on crack-tip stresses and subsequent propagation path, but
this is secondary to the effect of elastic gradient.
14. The effect of toughness variation on crack path was generally negligible, except at
interfaces where toughness anisotropy could result in deflection up the interface.
15. Bridging had a notable effect on crack propagation resistance, though an effect on
crack path was not discernible. This suggests that the shielding of the crack tip due to
bridging under mixed-mode loading is equivalent in each mode. Accordingly, crack
propagation paths were very similar under monotonic and cyclic loading, although
crack-extension toughening effects differed between these two cases.
16. The presence of interfaces within the stepped graded region had a significant effect
on crack propagation behaviour. In some specimens, interfaces were observed to have
a significant influence on crack path, as it was more energetically favourable for the
crack to grow straight up the weaker interface. Subsequent deflection off the interface
was observed, indicating a close competition between the mixed-mode stresses acting
to draw the crack off the interface and the reduced fatigue crack propagation
resistance of the interface. Compositional steps in the gradient lead to fluctuations in
stress concentration and in intrinsic and extrinsic fatigue crack propagation resistance
that can reduce reliability.
17. FE simulations revealed that compliance mismatch, due to mismatch in plastic
yielding behaviour, leads to similar effects on crack-tip stress fields as elastic
mismatch, namely mode-mixity and stress concentration. The development of a
plastic zone can result in an increase in effective toughness with crack extension for
high plastic mismatch cases.

18. Despite the diversity of possible behaviours and properties of different FGM systems,
the investigative framework and findings from the present study may be applied to
other FGM systems. It is very important to experimentally characterise FGMs rather
than relying on theoretical models.
19. Opportunities for optimising graded structures are limited by the properties of
constituent materials, and the mismatch between these properties. Several strategies
were recommended based on relative toughness of each side of the gradient and the
predicted propagation direction.

Appendices
Appendix A: Formulation for Effective medium approximation
The full set of equations used in modelling composite properties via the Effective Medium
Approximation technique is included here, for completeness. A more detailed derivation is
given in [4] which follows the work of Eshelby [34] and others [28,32]. The shape
functions, f1 and f2, are defined in a piecewise manner, in terms of shape parameter, :
2
1
cosh 1 (1 / )
2 3/ 2
2(1 )
2( 2 1)

f 1 ( ) =

1/3
2
21

f 2 ( ) =

2 3/ 2

3f1 ( ) - 1
1 2
0.2

cos 1 (1 / )

1
2( 2 1)

0<<1

(prolate ellipsoids)

=1

(spheres)

1<

(oblate ellipsoids)

1
=1

The shape polarisation tensor P is given in condensed Voigt notation as [34]:

3B* + *
P11 = P22 = *
((13 16 * )f1 () 3f 2 ())
*
*
8 (3B + 4 )
P33 =

3B* + *
(1 - 2 * - (1 4 * )f1 () f 2 ())
*
*
*
(3B + 4 )

P1 2 = P21 =

- 3B* *
(f1 () + f 2 ())
8 * (3B* + 4 * )

P1 3 = P31 = P23 = P32 =

P44 = P55 =

- 3B* *
(f1 () f 2 ())
2 * (3B* + 4 * )

3B* + *
(1 - * - (2 - * )f1 () + f 2 ())
*
*
*
2 (3B + 4 )

P66 = 0.5 (P11 P12 )

The summative terms are given as:
A-1

P1 = P2 =

3
f 1 ( )
3B + 4 *
*

P3 =

3
(1 - 2f 1 ( ))
3B + 4 *
*

Tensor, A(n), for each phase is determined from the terms of the components of P and the
summative terms:
*
*
*
2
A (n)
= (, ) + 2( n )P + [ B n B 3 ( n )]P

, = 1,2,3

*
A (n)
44 = 0.5 + 2( n )P44

where (,) is the Kronecker delta, equal to one for = , and zero otherwise.
The total magnitude are given by:
(n)
(n)
(n)
(n)
(n)
(A (n) ) 2 = A 33
( A 11
+ A 12
) 2A 13
A 31

Inverting A(n) gives the corresponding strain relation tensor, T(n):

( n)

= ( A ) 1
( n)

The strain relation coefficients are then calculated as:

(n)
B =

1
(n)
(n)
(n)
(n)
(n)
A 11
+ 2A 33
+ A 12
2( A 13
+ A 31
)
(n) 2
3(A )

(n) =

2
1
2
1
(n)
(n)
(n)
(n)
(n)
A11
) + (n)
+ 0.5A 33
+ A12
+ A13
+ A 31
+ (n)
(n) 2
(n)
5 A11 A12 A 44
15(A )

These calculations were done iteratively via MathCad for a number of compositions. Initial
guess values, B(0) and (0), were determined from Equation 8, then the following iteration
scheme was utilised to obtain stable solutions:
( B n B (i) ) TB(n) (B (i-1) , (i-1) ) = 0
( n (i) ) T(n) ( B (i -1) , (i -1) ) = 0

where the strain relation coefficients were calculated using the effective property values
from the previous iteration. An example of the MathCad code used to self-consistently
solve the EMA equations is shown in the following pages. The full code extended to
around 40 pages, so only a representative section is included, to enable the calculation
process to be better understood.
A-2

Analytical solution for Prediction of Composite Properties

via Effective Medium Approximation
Properties of constituents
E1

K1

397
E1

3.( 1

0.2

E2

2 . 1 )

2 (1

K1 = 220.556

K2

E1

G1

3.4

1 )

0.35
E2
3.( 1

2 )

2 (1

K2 = 3.778

G1 = 165.417

E2

G2

2 . 2 )

G2 = 1.259

Set range of volume fractions:

n

0 .. n
vf1j

0.5 .
( j)
n

0.5
vf2j

vf1j

Solution incorporating effect of inclusion shape

The expression for terms in matrices P and A involve the effective properties of the composite. As
self-consistent solution of equations involving matrices or excessive numbers of variables is not
possible with MC, an initial guess of the values without considering shape-factor (ie. = 1) is
used then iterations, including shape factor are used to improve and confirm result.

Shape parameter must be defined initially

Define shape functions

f1( )

if < 1 ,
2 1
f2( )

. acosh 1
3

2 2

1
2

1
, if 1 , ,
3
1

2
3

. acos 1

1
2

1 3 . f1( ) 1
if 1 , ,
2
5
1
0.4

f2( ) =

f1( ) =

f1( x )
f2( x )

0.2

0
0

A-3

Derivation for j=0

NB. Spherical inclusions are assumed initially, to

enable first order values to be obtained

Same type of derivation was used

for all other values of j

Guess values required by MathCad to guide the solution:

Kc

Gc

No-shape-factor equations (in fundamental form)

Given
( K1 Kc ) . vf10 ( K2 Kc ) . vf20
( G1 Gc) . vf10
0
4 .
4 .
1 . . 9 . Kc 8 . Gc
K1
Gc
K2
Gc
G1
Gc
3
3
6
Kc 2 . Gc
Kc > 0

( G2 Gc) . vf20
G2

1 . . 9 . Kc 8 . Gc
Gc
6
Kc 2 . Gc

Gc> 0
Find( Kc , Gc)

altpropvec

23.659

altpropvec =

15.427

First Iteration

Km

altpropvec 0

Gm

altpropvec 1

3 . Km 2 . Gm
0.5.
3 . Km G
Gm

All other iterations are essentially the same, with values of Km

and Gm taken from preceding iteration

Defining shape tensor P

P1 , 1

P3 , 3

P1 , 3

1
. ( ( 13
.
. Gm
3
Km
4
Gm
.
16.
3 . Km Gm
2
1
.( 1
.
. Gm
3
Km
4
Gm
.
2.
3 . Km Gm
2
1
Gm. 3 . Km 4 . Gm
4.
2
3 . Km Gm
P2 , 3

P4 , 4

Pm1

2 . m ( 1

. ( f1( )

Gm. 3 . Km 4 . Gm
4.
3 . Km Gm
2
3
. f1( )
.
3 Km 4 . Gm

.( 1

P1 , 2

m ( 2

Pm1

f2( ) )

P3 , 1

m) . f1( )

Pm3

P2 , 2

P1 , 1

1
. ( f1( )
.
. Gm
3
Km
4
Gm
.
16.
3 . Km Gm
2

P1 , 3

P5 , 5
Pm2

3 . f2( ) )

4 . m) . f1( )

f2( ) )

P3 , 2

P1 , 3

16. m) . f1( )

P4 , 4

f2( ) )

P1 , 3
P6 , 6

f2( ) )
P2 , 1

3
.( 1
.
3 Km 4 . Gm

1.
2

P1 , 1

P1 , 2

P1 , 2

2 . f1( ) )

A-4

Defining matrix A
A1 ,

( , )

A2 ,

( , )

A14 , 4

1
2

2 . ( G1 Gm) . P ,

K1

2 . ( G2 Gm) . P ,

K2

2.

Km

3
2.

Km

2 . ( G1 Gm) . P4 , 4

A13 , 3 . A11 , 1

A1sq

1 .. 3

2 . A11 , 3 . A13 , 1

A11 , 2

( G1 Gm) . Pm
( G2 Gm) . Pm

A24 , 4

1 .. 3

2 . ( G2 Gm) . P4 , 4

2
A2sq

A23 , 3 . A21 , 1

A21 , 2

2 . A21 , 3 . A23 , 1

Defining strain relation tensors

TK1

1 .
A11 , 1
.
3 A1sq

2 . A13 , 3

A11 , 2

2 . A13 , 1

A11 , 3

TK2

1 .
A21 , 1
.
3 A2sq

2 . A23 , 3

A21 , 2

2 . A23 , 1

A21 , 3

TG1

. A1
1, 1

0.5. A13 , 3

A11 , 2

A13 , 1

A11 , 3

0.2.

2
. A2
1, 1
.
15 A2sq

0.5. A23 , 3

A21 , 2

A23 , 1

A21 , 3

0.2.

15. A1sq

TG2

TK1 = 0.248

2
A11 , 1

A11 , 2

2
A21 , 1

A14 , 4

A21 , 2

A24 , 4

TG1 = 0.237
TG2 = 2.492

TK2 = 2.714

Solve for properties

Given
( K1

Kc ) . vf10 . TK1

Soln1

( K2

Kc ) . vf20 . TK2 0

( G1 Gc) . vf10 . TG1 ( G2 Gc) . vf20 . TG2 0

Find( Kc , Gc)
Soln10 = 21.955

Second Iteration

Gm

Soln11

Soln11 = 15.527
Km

Soln10

and so on.... for another four iterations, after which a stable solution is reached.

A-5

Appendix B: Sample ANSYS macro

The following macro is given as an example of the FE code, written in APDL (ANSYS
Parametric Design Language), used for simulations of crack growth. Specifically, it is the
macro for simulating the experimental specimen pictured in Figure 10.9. All the main
features of the FE simulations discussed in Chapter 5 are included: parameter-based
geometry definition, application of the temperature gradient, remeshing, SIF calculation by
displacement correlation, use of a test-kink and local symmetry criterion to predict
propagation direction. It should be noted that exclamation marks (!) are used for comments.
Also, lines of code which extend beyond the width of the page have been separated by ,
The addition of these was the only change to the code for presentation here.
!###########################################################
! MACRO for CRACK PROPAGATION in FGM in 4PB Configuration
!###########################################################
finish
/clear, start
/prep7

! INPUT specimen parameters

nsteps=6

! # compositional steps

nprof=1.5
nochpos=37
nincrs=42
lastoink=5

INPUT number of crack points

wstep=0.0187/nsteps
lcer=0.0313
lspec=0.135
hspec=0.0231
tspec=0.00486
nochlen=0.00978
wgrad=wstep*nsteps
lpol=lspec-2*wgrad-2*lcer
suppos=(lspec-0.12+0.005)
loadpos=(lspec-0.12+0.035)
buffer=2*wgrad/5
buffer1=2*wgrad/5
Erat=97
! Enter the PROPERTIES of POLYMER [GPa,n/a]:
Ep=3.4
vp=0.25

A-6

! Enter the PROPERTIES of CERAMIC [GPa,n/a]:

Ec=Ep*Erat
vc=0.25
!##### Enter the PROPERTIES of STEPS (from mixing law) [GPa,n/a]:
*DIM,gradprop,array,nsteps,2
gradprop(1,1)=270
! INPUT properties of gradient steps
gradprop(1,2)=0.25
! Youngs modulus & Poissons Ratio
gradprop(2,1)=190
gradprop(2,2)=0.25
gradprop(3,1)=115
gradprop(3,2)=0.25
gradprop(4,1)=80
gradprop(4,2)=0.25
gradprop(5,1)=40
gradprop(5,2)=0.25
gradprop(6,1)=25
gradprop(6,2)=0.25
!gradprop(7,1)=14
!gradprop(7,2)=0.25
!Define notch parameters
nochwide=0.0005
incrsize=0.00035
increm=incrsize/10

! width of notch for aesthetics only

! propagation increment size
! test-kink size

npositn=lcer+wgrad+lpol+wgrad*(nochpos/100)
j=0.000002

!define size of mesh at CT

kpn=130
*AFUN,deg

! keypoint number

TOFFST,273
! Define Ex as a function of temp
MPTEMP,,0,Ec-Ep
MP,EX, 1, 3.4e9,0,(1e9)/(Ec-Ep)
MP, NUXY, 1, 0.25
!@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
! Define keypoints
k,1,0,0
!define keypoints
k,2,lcer+wgrad+buffer,0
k,3,lcer+wgrad+lpol-buffer,0
k,4,lspec,0
k,5,lspec,hspec
k,6,lcer+wgrad+lpol-buffer,hspec
k,7,lcer+wgrad+buffer,hspec
k,8,0,hspec
k,9,lcer+wgrad+lpol-buffer,0.25*hspec
k,10,lspec-lcer+buffer1,0.45*hspec
k,11,suppos,0
k,12,lspec-0.005,0
k,13,lspec-0.035,hspec
k,14,loadpos,hspec

A-7

k,15,lcer-buffer1,0
k,16,lcer-buffer1,hspec
k,17,lspec-lcer+buffer1,0
k,18,lspec-lcer+buffer1,hspec
k,101,npositn-0.5*nochwide,0
!at front face of specimen
k,102,npositn+0.5*nochwide,0
k,kpn-2,npositn-0.25*nochwide,0.85*nochlen
k,kpn-3,npositn+0.25*nochwide,0.85*nochlen
k,kpn,npositn,nochlen-incrsize
k,kpn-1,npositn,nochlen-incrsize
k,20,lspec-lcer-0.5*wgrad,hspec
!define important structure lines lines
numstr,line,1
l,17,10
l,10,18
l,3,9
l,9,6
l,3,101
l,102,17
numstr,line,47
l,14,7
l,7,2
l,2,15
l,15,16
l,16,14
numstr,area,1
a,1,8,16,15,11
a,17,10,18,5,4,12
a,2,7,6,9,3
a,15,16,14,7,2

! Define areas
! ceramic left
! ceramic right
! polymer
! uncracked composite

! Mesh areas (except crack areas)

!define element type: planestress plane82
et,1,plane82,,,0
shpp,modify,1,12
shpp,modify,2,20
lesize,1,,,13
lesize,2,,,11
lesize,3,,,6
lesize,4,,,18
lesize,47,,,8
lesize,48,,,20
lesize,49,,,25
lesize,50,,,20
lesize,51,,,16
!mesh area 1-4 (ceramics,polymer,uncracked composite)
mat,1
smrtsize,2

A-8

amesh,3
mat,1
smrtsize,2
amesh,4
mat,1
smrtsize,2
amesh,1,2
! Assign temps to nodes
*DIM,T_,ARRAY,40000

! Vector for nodal temperatures

! First to Area 4
!select area
asel, s, area,, 4
!select nodes and elements attached to area
nsla, s, 1
nplot
esln, s,1
*GET,NMAX_,NODE,0,NUM,MAX ! Get the maximum node number
*GET,NNUM_,NODE,,COUNT,,,, ! Get # nodes selected
NDNEXT_ = 0
! Next Highest Node
*DO,intI,1,NNUM_,1
*GET,NDNEXT_,NODE,NDNEXT_,NXTH
*IF,NX(NDNEXT_),le,lcer,THEN
prop=Ec
*ELSEIF,NX(NDNEXT_),lt,lcer+wgrad,THEN
postep=NINT(((wgrad+lcer-NX(NDNEXT_))/wgrad)*nsteps+0.5)
prop=gradprop(nsteps+1-postep,1)
*ELSE
prop=Ep
*ENDIF
*SET,T_(NDNEXT_), prop-3.4
BF, NDNEXT_, TEMP, T_(NDNEXT_)
*ENDDO
! Then to Areas 1,2,3
ASEL, S, AREA,, 1
BFA, 1, TEMP, Ec-3.4
ASEL, S, AREA,, 2
BFA, 2, TEMP, Ec-3.4
ASEL, S, AREA,, 3
BFA, 3, TEMP, 0
BFTRAN
/PBF, TEMP,, 1
EPLOT

!Plot body force

! Input crack parameters

nincrmax=50
*DIM,crpx,array,nincrmax+1
*DIM,crpy,array,nincrmax+1
*DIM,Load,array,nincrmax+1,2
*DIM,crpang,array,nincrmax+1

A-9

*DIM,K1result,array,nincrmax,2
*DIM,K2result,array,nincrmax,2
*DIM,K2valz,array,20
*DIM,ThetaG,array,20
*DIM,Kmin,array,nincrmax,30
*DIM,Wvalz,array,50
*DIM,Gvalz,array,50
crpx(1)=npositn
crpy(1)=nochlen
Load(1,1)=20000
crpang(1)=0
! LOOP1 over CRACK PROPAGATION increments
*DO,Baa,1,nincrs,1
! Define crack keypoints
npts=Baa
!Define Intraloopous Parameters
ctposx=crpx(npts)
ctposy=crpy(npts)
ctang=crpang(npts)
ctpostep=NINT(((ctposx-lcer-lpol-wgrad)/wgrad)*nsteps+0.5)
*IF,ctposx,gt,lcer+wgrad+lpol,THEN
ctprop=gradprop(nsteps+1-ctpostep,1)
Kcguess=1+3*(ctpostep/(1+nsteps))**nprof
*ELSE
Kcguess=1
ctprop=3.4
*ENDIF
*IF,ctposx,gt,lcer+wgrad+lpol-0.002,THEN
*IF,ctposy,LT,0.9*hspec,THEN
/prep7
csys,0
MP, EX, 2, ctprop*1e9
MP, NUXY, 2, 0.2
allsel,all
*IF,npts,gt,1,THEN
*DO,I,1,npts-1
k,kpn-1+(2*I),crpx(I),crpy(I)
k,kpn+(2*I),crpx(I),crpy(I)
*ENDDO
*ENDIF
*AFUN,deg
! near crack-tip
k,98,ctposx+4*j*sin(ctang),ctposy-4*j*cos(ctang)
k,97,ctposx+8*j*sin(ctang),ctposy-8*j*cos(ctang)
k,96,ctposx+4*j*sin(ctang),ctposy-4*j*cos(ctang)
k,95,ctposx+8*j*sin(ctang),ctposy-8*j*cos(ctang)
k,94,ctposx+2*incrsize*cos(ctang)+1*incrsize*sin(ctang),
ctposy+2*incrsize*sin(ctang)-1*incrsize*cos(ctang)
k,93,ctposx-2*incrsize*cos(ctang)+1*incrsize*sin(ctang),
ctposy-2*incrsize*sin(ctang)-1*incrsize*cos(ctang)

A-10

k,92,ctposx-2*incrsize*sin(ctang),
ctposy+2*incrsize*cos(ctang)
k,99,ctposx,ctposy
!crack-tip
k,19,ctposx-2*incrsize*sin(ctang),ctposy+6*incrsize !above CT
*IF,npts,gt,2,THEN
k,91,0.7*(ctposx-incrsize*(cos(ctang)+sin(ctang)))+
npositn/3, 0.9*nochlen
k,90,incrsize+1*npositn,1*(ctposy+2*incrsize*sin(ctang)incrsize*cos(ctang))-incrsize/2
*ENDIF
k,103,ctposx-2*incrsize*cos(ctang)-1*incrsize*sin(ctang),
ctposy-2*incrsize*sin(ctang)+1*incrsize*cos(ctang)
k,104,ctposx+2*incrsize*cos(ctang)-1*incrsize*sin(ctang),
ctposy+2*incrsize*sin(ctang)+1*incrsize*cos(ctang)
k,105,ctposx-10*j*cos(ctang),ctposy-10*j*sin(ctang)
k,106,ctposx+10*j*cos(ctang),ctposy+10*j*sin(ctang)
k,107,ctposx-10*j*sin(ctang),ctposy+10*j*cos(ctang)
numstr,line,500
l,6,20
l,13,18
l,101,kpn-2
l,102,kpn-3
numstr,line,504
l,kpn-2+(2*npts),93
l,kpn-3+(2*npts),94
numstr,line,506
l,93,103
l,94,104
l,20,19
l,19,92
l,103,92
l,104,92
l,97,105
l,105,107
l,107,106
l,106,95
l,107,92
numstr,line,520
l,kpn-2+(2*npts),97
l,kpn-3+(2*npts),95
l,kpn-4+(2*npts),kpn-2+(2*npts)
l,kpn-5+(2*npts),kpn-3+(2*npts)
numstr,line,524
l,107,92
l,20,18
! Define crack areas
*IF,npts,eq,1,THEN
numstr,area,5
a,3,9,6,20,19,92,103,93,kpn,kpn-2,101
a,18,10,17,102,kpn-3,kpn-1,94,104,92,19,20,13
a,93,103,92,107,105,97,kpn
a,94,kpn-1,95,106,107,92,104
a,107,99,98,97,105
a,95,96,99,107,106

A-11

*ELSEIF,npts,eq,2,THEN
numstr,area,5
a,3,9,6,20,19,92,103,93,kpn+2,kpn,kpn-2,101
a,18,10,17,102,kpn-3,kpn-1,
kpn+1,94,104,92,19,20,13
a,93,103,92,107,105,97,kpn+2
a,94,kpn+1,95,106,107,92,104
a,107,99,98,97,105
a,95,96,99,107,106
*ELSE
numstr,area,60
a,3,9,91,kpn+2,kpn,kpn-2,101
*DO,I,1,npts-2,1
numstr,area,60+(2*I)-1
a,kpn+(2*I),91,kpn+2+(2*I)
aadd,60+(2*I)-2,60+(2*I)-1
*ENDDO
numstr,area,60+(2*npts)-3
a,9,6,20,19,92,103,93,kpn+2*npts-2,91
numstr,area,5
aadd,60+(2*npts)-4,60+(2*npts)-3
numstr,area,7
a,kpn-2+(2*npts),93,103,92,107,105,97
numstr,area,260
a,102,kpn-3,kpn-1,kpn+1,90,10,17
*DO,I,1,npts-2,1
numstr,area,260+(2*I)-1
a,kpn-1+(2*I),kpn+1+(2*I),90
aadd,260+(2*I)-2,260+(2*I)-1
*ENDDO
numstr,area,260+(2*npts)-3
a,kpn-3+2*npts,94,104,92,19,20,13,18,10,90
numstr,area,6
aadd,260+(2*npts)-4,260+(2*npts)-3
numstr,area,8
a,kpn-3+(2*npts),95,106,107,92,104,94
a,107,99,98,97,105
a,95,96,99,107,106
*ENDIF
! Mesh crack areas
lesize,5,,,25
lesize,6,,,10
lesize,500,,,20
lesize,501,,,10
lesize,502,,,10
lesize,503,,,10
lesize,504,,,6
lesize,505,,,6
lesize,506,,,14
lesize,507,,,14
lesize,508,,,20
lesize,509,,,6
lesize,510,,,12

A-12

lesize,511,,,12
lesize,512,,,4
lesize,513,,,4
lesize,514,,,4
lesize,515,,,4
lesize,516,,,10
lesize,520,,,12
lesize,521,,,12
lesize,524,,,16
lesize,525,,,16
!set concentration point
kscon,99,j,1,,0.5
!mesh areas 5,6,7,8
mat,1
mshape,0,2D
!quadrangle meshing in 2D model
asel,s,area,,5,10
smrtsize,2
amesh, all
! Assign temps to nodes near crack
asel, s, area,, 5,10,1
nsla, s, 1
nplot
esln, s,1

!select area
! select nodes and elements
! attached to area

*GET,NMAX_,NODE,0,NUM,MAX ! Get the maximum node number

*GET,NNUM_,NODE,,COUNT, , , ,
! Get # nodes selected
NDNEXT_ = 0
! Next Highest Node
*DO,intI,1,NNUM_,1
prop=Ep
*GET,NDNEXT_,NODE,NDNEXT_,NXTH
*IF,NX(NDNEXT_),lt,lcer+wgrad+lpol,THEN
prop=Ep
*ELSEIF,NX(NDNEXT_),eq,lcer+wgrad+lpol,THEN
prop=Ep
*ELSEIF,NX(NDNEXT_),lt,lspec-lcer,THEN
postep=NINT(((NX(NDNEXT_)-(lspec-lcerwgrad))/wgrad)*nsteps+0.5)
prop=gradprop(nsteps+1-postep,1)
*ELSE
prop=Ec
*ENDIF
*SET,T_(NDNEXT_),prop-3.4
BF,NDNEXT_,TEMP,T_(NDNEXT_)
*ENDDO
allsel,all
BFTRAN
/PBF, TEMP,, 1
!Plot body force
EPLOT
*DO,Moo,1,2,1

!Loop to get critical Load

dk,11,uy
! Apply Structural Loading
dk,12,uy
dk,11,ux
ksel,s,kp,,13,14

A-13

fk,all,fy,-(Load(Baa,Moo)/2)
allsel,all
sbctran
/OUTPUT

! sends text output to default system output

file
OUTPR,,ALL ! store all basic stuff and strain energies
OUTPR,VENG,ALL
! Locate immediately prior to SOLU
/solu

! Run FEM
solve
finish
/POST1
!Calculate SIFs via DCT
allsel,all
! Determine Stress Intensity Factors
cskp,11,0,99,98,3
PATH,fredericka,5,30,20,
KSEL,S,KP,,99
NSLK,S
*GET,PNode1,NODE,0,NXTH
PPATH,1,PNode1, , ,
KSEL,S,KP,,98
NSLK,S
*GET,PNode2,NODE,0,NXTH
PPATH,2,PNode2, , ,
KSEL,S,KP,,97
NSLK,S
*GET,PNode3,NODE,0,NXTH
PPATH,3,PNode3, , ,
KSEL,S,KP,,96
NSLK,S
*GET,PNode4,NODE,0,NXTH
PPATH,4,PNode4, , ,
KSEL,S,KP,,95
NSLK,S
*GET,PNode5,NODE,0,NXTH
PPATH,5,PNode5, , ,
RSYS,11
KCALC,0,2,3,1
allsel,all

! AUTO PICKS PATH NODE 1

,11
! AUTO PICKS PATH NODE 2
,11
! AUTO PICKS PATH NODE 3
,11
! AUTO PICKS PATH NODE 4
,11
! AUTO PICKS PATH NODE 5
,11

*GET,K1result(Baa,Moo),KCALC,,K,1
*GET,K2result(Baa,Moo),KCALC,,K,2
KSEL,S,KP,,97,,,0
! Determine kink direction
NSLK,S
*GET,NONU260,NODE,,NUM,MAX
*GET,CODYA,NODE,NONU260,U,Y
*GET,CODXA,NODE,NONU260,U,X
*SET,UYLO,CODYA
*SET,UXLO,CODXA
KSEL,S,KP,,95,,,0
NSLK,S

A-14

*GET,NONU261,NODE,,NUM,MAX
*GET,CODYB,NODE,NONU261,U,Y
*GET,CODXB,NODE,NONU261,U,X
*SET,UYUP,CODYB
*SET,UXUP,CODXB
CODY=(UYUP-UYLO)/2
CODX=(UXUP-UXLO)/2
*AFUN,deg
COD2=CODY*cos(ctang)+CODX*sin(ctang)
*IF,COD2,GT,0,THEN
kinkflag=-1
*ELSEIF,COD2,LT,0,THEN
kinkflag=1
*ELSE
kinkflag=0
*ENDIF
allsel,all
!Calculate LLDs
rsys,0
KSEL,S,,,13,
NSLK,S
*GET,NONU11,NODE,,NUM,MAX
*GET,LLD1,NODE,NONU11,U,Y
KSEL,S,KP,,14,
NSLK,S
*GET,NONU12,NODE,,NUM,MAX
*GET,lld2,NODE,NONU12,U,Y
allsel,all
Winit=-0.25*Load(Baa,2)*(LLD1+LLD2)
*IF,Baa,eq,1,THEN
Winit1=Winit
*ENDIF
finish
! Delete Structural Loads
/prep7
nsel,all
fdele,all,fx
fdele,all,fy
ksel,all
fkdele,all,all
dkdele,all,all
! Calculate next load
*IF,Moo,EQ,1,THEN
Ktotal=((K1result(Baa,1))**2+(K2result(Baa,1))**2)**0.5
Load(Baa,2)=Load(Baa,1)*1000000*(Kcguess/Ktotal)
*ELSE
Load(Baa+1,1)=Load(Baa,2)
*ENDIF
*ENDDO

!end of Moo Loop

A-15

!Calculate initial guess angle and corresponding crack points

!(increment 10 um)
momix=K2result(Baa,2)/K1result(Baa,2)
kinkdet=(1-SQRT(1+8*momix**2))/(4*momix)
*AFUN,deg
*IF,Baa,lt,5,THEN
kinkMTS=-1*kinkflag*2*ATAN(kinkdet)
*ELSE
kinkMTS=-1*kinkflag*2*ATAN(kinkdet)!*0.5
*ENDIF
thetaMTS=crpang(Baa)+kinkMTS
triang=crpang(Baa)+kinkMTS/10
*IF,thetaMTS,gt,90,THEN
thetaMTS=crpang(Baa)-kinkMTS
*ENDIF
*IF,npts,LT,nincrs+1,THEN
/prep7
asel,s,,,5,10,1
nsla,s,1
BFDELE, ALL, TEMP

!Delete temps/mesh/areas around crack-tip

asel,all
nsel,all
ACLEAR, 5,10,1
KSCON, 99,,,

!delete previous stress concentration pt

ADELE,5,10,1
ksel,s,kp,,90,99,1
lslk,s,0
ldele,all,,, 0
ksel,s,kp,,19
lslk,s,0
ldele,all,,, 0
ksel,s,kp,,103,107,1
lslk,s,0
ldele,all,,, 0
*IF,npts,gt,1,THEN
ksel, s, kp,, kpn+1,kpn-2+2*npts,1
lslk, s, 0
ldele, all,,, 0
*ENDIF
ksel,all
lsel,all
kdele,90,99,1
kdele,103,107,1
kdele,19
*IF,npts,gt,1,THEN
kdele,kpn+1,kpn-2+2*npts,1
*ENDIF
allsel,all

A-16

! Loop over different angles for testkink

thetag(2)=thetaMTS+16+kinkMTS/4
thetag(3)=(thetaMTS/2)-6
Thetup=thetag(2)
Thetdo=thetag(3)
*DO,Oink,2,lastoink,1
! Define crack with extra test kink
npts=Baa+1 !Define Intraloopous Parameters
*AFUN,deg
ctposx=crpx(Baa)-increm*sin(thetag(Oink))
ctposy=crpy(Baa)+increm*cos(thetag(Oink))
ctang=thetag(Oink)
ctpostep=NINT(((ctposx-lcer-lpol-wgrad)/wgrad)*nsteps+0.5)
*IF,ctposx,gt,lcer+wgrad+lpol,THEN
ctprop=gradprop(nsteps+1-ctpostep,1)
Kcguess=1+3*(ctpostep/(1+nsteps))**nprof
*ELSE
Kcguess=1
ctprop=3.4
*ENDIF
!Crack keypoints
/prep7
csys,0
MP, EX, 2, ctprop*1e9
MP, NUXY, 2, 0.2
allsel,all
k,kpn-1+(2*npts)-2,crpx(npts-1),crpy(npts-1)
k,kpn+(2*npts)-2,crpx(npts-1),crpy(npts-1)
*AFUN,deg
k,98,ctposx+3*j*sin(ctang),ctposy-3*j*cos(ctang)
!near CT
k,97,ctposx+6*j*sin(ctang),ctposy-6*j*cos(ctang)
k,96,ctposx+3*j*sin(ctang),ctposy-3*j*cos(ctang)
k,95,ctposx+6*j*sin(ctang),ctposy-6*j*cos(ctang)
k,99,ctposx,ctposy
!crack-tip
k,105,ctposx-10*j*cos(ctang),ctposy-10*j*sin(ctang)
k,106,ctposx+10*j*cos(ctang),ctposy+10*j*sin(ctang)
k,107,ctposx-10*j*sin(ctang),ctposy+10*j*cos(ctang)
*IF,Oink,eq,2,THEN
*IF,npts,gt,2,THEN
*DO,I,1,npts-2
k,kpn-1+(2*I),crpx(I),crpy(I)
k,kpn+(2*I),crpx(I),crpy(I)
*ENDDO
*ENDIF
k,94,ctposx+2*incrsize*cos(triang)+1*incrsize*sin(triang),
ctposy+2*incrsize*sin(triang)-1*incrsize*cos(triang)
k,93,ctposx-2*incrsize*cos(triang)+1*incrsize*sin(triang),

A-17

ctposy-2*incrsize*sin(triang)-1*incrsize*cos(triang)
k,92,ctposx-2*incrsize*sin(triang),
ctposy+2*incrsize*cos(triang)
k,19,ctposx-6*incrsize*sin(2*ctang/3),
! above crack-tip
ctposy+6*incrsize*cos(2*ctang/3)
*if,npts,gt,2,then
k,91,0.7*(ctposx-incrsize*(cos(triang)sin(triang)))+0.3*npositn,0.9*nochlen
k,90,incrsize+npositn,ctposy+incrsize*(2*sin(triang)cos(triang)-0.5)
*endif
k,103,ctposx-2*incrsize*cos(triang)-2*incrsize*sin(triang),
ctposy-2*incrsize*sin(triang)+2*incrsize*cos(triang)
k,104,ctposx+2*incrsize*cos(triang)-2*incrsize*sin(triang),
ctposy+2*incrsize*sin(triang)+2*incrsize*cos(triang)
numstr,line,500
l,6,20
l,20,18
l,101,kpn-2
l,102,kpn-3
numstr,line,504
l,kpn-4+(2*npts),93
l,kpn-5+(2*npts),94
numstr,line,506
l,93,103
l,94,104
l,19,20
l,92,19
l,104,92
l,103,92
numstr,line,522+2*npts
l,kpn-4+(2*npts),kpn-2+(2*npts)
l,kpn-5+(2*npts),kpn-3+(2*npts)
*ENDIF
numstr,line,512
l,97,105
l,105,107
l,107,106
l,106,95
numstr,line,520
l,kpn-2+(2*npts),97
l,kpn-3+(2*npts),95
numstr,line,524+2*npts
l,107,92
!Crack areas
*IF,npts,eq,1,THEN
*IF,Oink,eq,2,THEN
numstr,area,5
a,3,9,6,20,19,92,103,93,kpn-2,101
a,18,10,17,102,kpn-3,94,104,92,19,20,13
*ENDIF
numstr,area,7
a,93,103,92,107,105,97,kpn,kpn-2
a,94,kpn-3,kpn-1,95,106,107,92,104
a,107,99,98,97,105

A-18

a,95,96,99,107,106
*ELSEIF,npts,eq,2,THEN
*IF,Oink,eq,2,THEN
numstr,area,5
a,3,9,6,20,19,92,103,93,kpn,kpn-2,101
a,18,10,17,102,kpn-3,kpn1,94,104,92,19,20,13
*ENDIF
numstr,area,7
a,93,103,92,107,105,97,kpn+2,kpn
a,94,kpn-1,kpn+1,95,106,107,92,104
a,107,99,98,97,105
a,95,96,99,107,106
*ELSE
*IF,Oink,eq,2,THEN
numstr,area,60
a,3,9,91,kpn,kpn-2,101
*DO,I,1,npts-2,1
numstr,area,60+(2*I)-1
a,kpn+(2*I)-2,91,kpn+(2*I)
aadd,60+(2*I)-2,60+(2*I)-1
*ENDDO
numstr,area,60+(2*npts)-3
a,9,6,20,19,92,103,93,kpn+2*npts-4,91
numstr,area,5
aadd,60+(2*npts)-4,60+(2*npts)-3
*ENDIF
numstr,area,7
a,kpn+2*npts-2,kpn-4+(2*npts),93,103,92,107,105,97
*IF,Oink,eq,2,THEN
numstr,area,260
a,102,kpn-3,kpn-1,90,10,17
*DO,I,1,npts-2,1
numstr,area,260+(2*I)-1
a,kpn-3+(2*I),kpn-1+(2*I),90
aadd,260+(2*I)-2,260+(2*I)-1
*ENDDO
numstr,area,260+(2*npts)-3
a,kpn-5+2*npts,94,104,92,19,20,13,18,10,90
numstr,area,6
aadd,260+(2*npts)-4,260+(2*npts)-3
*ENDIF
numstr,area,8
a,kpn-5+(2*npts),kpn+2*npts-3,95,106,107,92,104,94
a,107,99,98,97,105
a,95,96,99,107,106
*ENDIF
! Mesh crack areas
lesize,5,,,10
lesize,6,,,10
lesize,500,,,14
lesize,501,,,10
lesize,502,,,10
lesize,503,,,10
lesize,504,,,12
lesize,505,,,16
lesize,506,,,22

A-19

lesize,507,,,28
lesize,508,,,20
lesize,509,,,6
lesize,510,,,12
lesize,511,,,12
lesize,512,,,4
lesize,513,,,4
lesize,514,,,4
lesize,515,,,4
lesize,520,,,2
lesize,521,,,2
lesize,522+2*npts,,,12
lesize,523+2*npts,,,12
lesize,524+2*npts,,,20
kscon,99,j,1,,0.5
!set
mat,1
mshape,0,2D
!triangle
*IF,Oink,eq,2,THEN
asel, s, area,,5,6
smrtsize,2
amesh,all
*Endif
mshape,0,2D
!triangle
asel, s, area,,7,10
smrtsize,1
amesh, all

concentration point
meshing in 2D model
!mesh areas 5 to 10

meshing in 2D model

!Assign Temps
*IF,Oink,eq,2,THEN
!select area
asel, s, area,,5,10
*ELSE
asel, s, area,,7,10
*ENDIF
nsla, s, 1
!select nodes and elements attached to area
nplot
esln, s,1
*GET,NMAX_,NODE,0,NUM,MAX ! Get the maximum node number
*GET,NNUM_,NODE,,COUNT, , , ,
! Get the number of nodes
number selected
NDNEXT_ = 0
! Next Highest Node
*DO,intI,1,NNUM_,1
prop=Ep
*GET,NDNEXT_,NODE,NDNEXT_,NXTH
*IF,NX(NDNEXT_),lt,lcer+wgrad+lpol,THEN
prop=Ep
*ELSEIF,NX(NDNEXT_),eq,lcer+wgrad+lpol,THEN
prop=Ep
*ELSEIF,NX(NDNEXT_),lt,lspec-lcer,THEN
postep=NINT(((NX(NDNEXT_)-(lspec-lcerwgrad))/wgrad)*nsteps+0.5)
prop=gradprop(nsteps+1-postep,1)
*ELSE
prop=Ec
*ENDIF
*SET,T_(NDNEXT_),prop-3.4

A-20

BF,NDNEXT_,TEMP,T_(NDNEXT_)
*ENDDO
allsel,all
BFTRAN
/PBF, TEMP,, 1
!Plot body force
EPLOT
!Load with prior crit Load & solve
dk,11,uy
! Apply Structural Loading
dk,12,uy
dk,11,ux
ksel,s,kp,,13,14
fk,all,fy,-(Load(Baa,2)/2)
allsel,all
sbctran
/OUTPUT
! sends text output to default system output file
OUTPR,,ALL ! store all basic data
OUTPR,VENG,ALL
! and strain energies
/solu

! Run FEM
solve
finish
/POST1
!@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
!Calculate SIFs via DCT - get KII
allsel,all
! Determine Stress Intensity Factors
cskp,11,0,99,98,3
PATH,fredericka,5,30,20,
KSEL,S,KP,,99
NSLK,S
*GET,PNode1,NODE,0,NXTH
PPATH,1,PNode1, , ,
KSEL,S,KP,,98
NSLK,S
*GET,PNode2,NODE,0,NXTH
PPATH,2,PNode2, , ,
KSEL,S,KP,,97
NSLK,S
*GET,PNode3,NODE,0,NXTH
PPATH,3,PNode3, , ,
KSEL,S,KP,,96
NSLK,S
*GET,PNode4,NODE,0,NXTH
PPATH,4,PNode4, , ,
KSEL,S,KP,,95
NSLK,S
*GET,PNode5,NODE,0,NXTH
PPATH,5,PNode5, , ,
RSYS,11
KCALC,0,2,3,1
allsel,all

! AUTO PICKS PATH NODE 1

,11
! AUTO PICKS PATH NODE 2
,11
! AUTO PICKS PATH NODE 3
,11
! AUTO PICKS PATH NODE 4
,11
! AUTO PICKS PATH NODE 5
,11

*GET,K1resul,KCALC,,K,1

A-21

*GET,K2resul,KCALC,,K,2
K2valz(Oink)=K2resul
Gdct=(K1resul**2+K2resul**2)/(ctprop*1e9)
Gvalz(Oink)=Gdct
!Calculate LLDs
rsys,0
KSEL,S,KP,,13,
NSLK,S
*GET,NONU11,NODE,,NUM,MAX
*GET,LLD3,NODE,NONU11,U,Y
KSEL,S,KP,,14,
NSLK,S
*GET,NONU12,NODE,,NUM,MAX
*GET,LLD4,NODE,NONU12,U,Y
Wincr=-0.25*Load(Baa,2)*(LLD3+LLD4)
Wvalz(Oink)=Wincr-Winit
*IF,Oink,eq,2,THEN
Wincr1=Wincr
*ENDIF
finish
! Get new thetahi and thetalo from LLD-Work criterion
*IF,Oink,eq,3,THEN
K3up=K2valz(2)
K3do=K2valz(3)
thetag(Oink+1)=(Thetdo+(K3do/(K3up+K3do))*(Thetup-Thetdo))+4
thetag(Oink+2)=(Thetdo+(K3do/(K3up+K3do))*(Thetup-Thetdo))-4
Thetup=thetag(Oink+1)
Thetdo=thetag(Oink+2)
*ENDIF
*IF,Oink,EQ,5,THEN
K3up=K2valz(4)
K3do=K2valz(5)
thetag(Oink+1)=(Thetdo+(K3do/(K3up+K3do))*(Thetup-Thetdo))
thetag(Oink+2)=thetag(Oink+1)
*ENDIF
delflag=1
*IF,Baa,eq,nincrs,THEN
*IF,Oink,eq,lastoink,THEN
delflag=0
*ENDIF
*ENDIF
*IF,delflag,eq,1,THEN
/prep7
! Delete everything
nsel,all
! Delete Structural Loads
fdele,all,fx
fdele,all,fy
ksel,all
fkdele,all,all
dkdele,all,all

A-22

asel,s,,,7,10,1
nsla,s,1
BFDELE, ALL, TEMP
asel,all
nsel,all
ACLEAR, 7,10,1
KSCON, 99,,,

!delete previous stress concentration pt

ADELE,7,10,1
ksel,s,kp,,95,99,1
lslk,s,0
ldele,all,,, 0
ksel,s,kp,,105,107,1
lslk,s,0
ldele,all,,, 0
ksel,all
lsel,all
kdele,95,99,1
kdele,105,107,1
ksel,all
lsel,all
*ENDIF
!end of delflag condition
*ENDDO

! end of Oink-loop

*IF,delflag,eq,1,THEN
/prep7
asel,s,,,5,6,1
nsla,s,1
BFDELE, ALL, TEMP
asel,all
nsel,all
ACLEAR, 5,6,1
ADELE,5,6,1
ksel,s,kp,,90,94,1
lslk,s,0
ldele,all,,,0
ksel,s,kp,, 19
lslk,s,0
ldele,all,,,0
ksel,s,kp,,103,104,1
lslk,s,0
ldele,all,,,0
*IF,npts,gt,1,THEN
ksel,s,kp,,kpn+1,kpn-2+2*npts,1
lslk,s,0
ldele,all,,, 0
*ENDIF
ksel,all
lsel,all
kdele,90,94,1
kdele,19
kdele,103,104

A-23

*IF,npts,gt,1,THEN
kdele,kpn+1,kpn-2+2*npts,1
*ENDIF
allsel,all
*ENDIF
!end of delflag condition
*ENDIF
!End of not-last-increment condition
Kmin(Baa)=K2valz(lastoink)
crpang(Baa+1)=thetag(lastoink+1)
crpx(Baa+1)=crpx(Baa)-incrsize*sin(crpang(Baa+1))
crpy(Baa+1)=crpy(Baa)+incrsize*cos(crpang(Baa+1))
*ENDIF
*ENDIF

!End of crack-within-gradient condition

! End LOOP1 (Baa)

*ENDDO
*MSG,UI,
View G4_Resu and G4criteres files for stress intensity factors
*DIM,Allres,array,nincrmax+1,6
*DO,I,1,nincrs+1,1
Allres(I,1)=I
Allres(I,2)=K1result(I,2)
Allres(I,3)=K2result(I,2)
Allres(I,4)=Load(I,2)*tspec
Allres(I,5)=crpx(I)
Allres(I,6)=crpy(I)
*ENDDO

! Present results

*MWRITE,Allres,G15_Resu,,,JIK,
Incr = %G K1 = %G K2 = %G Load = %G cpx = %G cpy = %G
*DO,I,1,20,1
Allres(I,1)=ThetaG(I)
Allres(I,2)=Jvalz(I)
Allres(I,3)=Gvalz(I)
Allres(I,4)=K2valz(I)
Allres(I,5)=Wvalz(I)
*ENDDO
*MWRITE,Allres,G15criteres,,,JIK,
Theta = %G Jsim = %G Gdct = %G K2 = %G Wcom = %G
allsel,all
/PBF, TEMP,, 1
EPLOT

!Plot body force

!End of macro

A-24

Appendix C: Comparison with analytical model

Semi-infinite plane

E(x) = Eoex
Far field
stress o

Vertical symmetry BCs

1.2

Analytical
Computational

1/2

0.3

0.2

1.1

II

0.1

II

Analytical
Computational

Normalised Mode II SIF K /( a )

Normalised Mode I SIF K /( a )

1.3

0.9

1/2

0.8

0.2

0.4

0.6

0.8

Normalised crack length, a

Figure A.1: Comparison of computational results with those from the analytical model of Konda
and Erdogan [1994], for a vertical crack in an infinite plane with an exponential stiffness gradient in
the horizontal direction. Good agreement is observed.

A-25

Appendix D: Crack Propagation in Graded Specimens: Additional FE results

12

12

Continuous Gradient (b)

(a)

10

x [mm]

12

Stepped Gradient: 5 Steps

(c)

12

10

n=1
n = 0.5
n=2

y [mm]

y [mm]

10

n=1
n = 0.5
n=2

10

Stepped Gradient: 3 Steps

(d)
10

n=1
n = 0.5
n=2

10

x [mm]

12

n=1
n = 0.5
n=2

12

x [mm]

12

y [mm]

y [mm]

10

Stepped Gradient: 11 Steps

10

x [mm]

12

Figure A.2: Crack propagation paths for initially straight cracks in (a) a continuous gradient, and
stepped gradients with (b) 3 steps, (c) 5 steps and (d) 11 steps (E2/E1 = 100, a0/h = 0.35, w = 15
mm). The effects of differing gradient profile and initial crack position ( = 0.17, 0.5, 0.83) are
illustrated.

A-26

12

(a)

Gradient Profile, n = 0.5

y [mm]

10

3 steps
5 steps

11 steps
Continuous

0
12

y [mm]

(b) 10

10

14 x

12

[mm]

Gradient Profile, n = 1

3 steps
5 steps

11 steps
Continuous

0
12

(c)
y [mm]

10

10

12

14

x [mm]

Gradient Profile, n = 2

3 steps
5 steps

11 steps
Continuous

0
2
4
6
8
10
12
14 x [mm]
Figure A.3: Propagation paths for initially straight cracks in stepped and continuous gradients
(E2/E1 = 100, a0/h = 0.35, w = 15 mm), with property gradient profiles given by (a) n = 0.5, (b) n =
1 and (c) n = 2. The effects of gradient continuity and initial crack position ( = 0.17, 0.5, 0.83) are
illustrated.

A-27

Gradient Profile, n = 1
5 steps
11 steps
3 steps
Continuous

y [mm]

10

12

x [mm]

Figure A.4: SIF profiles for initially straight cracks in stepped and continuous gradients (E2/E1 =
100, a0/h = 0.35, w = 15 mm), with property gradient profiles given by (a) n = 0.5, (b) n = 1 and (c)
n = 2. The effects of gradient continuity and initial crack position ( = 0.17, 0.5, 0.83) are
illustrated.

A-28

Continuous Gradient
n=1
n = 0.5
n=2

K gr/Khom

10

x [mm]

12

Stepped Gradient: 11 Steps

n=1
n = 0.5
n=2

K gr/Khom

10

x [mm]

12

Figure A.5: SIF profiles for initially straight cracks in stepped and continuous gradients, (E2/E1 =
100, a0/h = 0.35, w = 15 mm), with property gradient profiles given by (a) n = 0.5, (b) n = 1 and (c)
n = 2. The effects of gradient continuity and initial crack position ( = 0.17, 0.5, 0.83) are
illustrated.

A-29

Stepped Gradient: 3 Steps

6

n=1
n = 0.5
n=2

K gr/Khom

5
4
3
2
1
0

10

12

x [mm]
6

Stepped Gradient: 5 Steps

n=1
n = 0.5
n=2

K gr/Khom

10

x [mm]

12

Figure A.6: SIF profiles for initially straight cracks in stepped gradients, (E2/E1 = 100, a0/h = 0.35,
w = 15 mm), with (a) 3 steps and (b) 4 steps. The effects of gradient profile and initial crack
position ( = 0.17, 0.5, 0.83) are illustrated.

A-30

Gradient Profile, n = 0.5

5 steps
11 steps
3 steps
Continuous

gr

K /K

hom

10

12

x [mm]

3.5

Gradient Profile, n = 2
3

gr

K /K

hom

2.5

1.5

5 steps
11 steps
3 steps
Continuous

1
0.5
0

10

12

x [mm]

Figure A.7: SIF profiles for initially straight cracks in stepped and continuous gradients, (E2/E1 =
100, a0/h = 0.35, w = 15 mm), with property gradient profiles given by (a) n = 0.5 and (b) n = 2. The
effects of gradient continuity and initial crack position ( = 0.17, 0.5, 0.83) are illustrated.

A-31

22000

20000

w = 2 mm

18000

w = 5 mm

16000

w = 10 mm
14000

12000

w = 20 mm

10000
Graded Region

Epoxy

Alumina

8000
0

0.5

Figure A.8: Stress profile along the lower face of a step-graded beam, with property profile given
by n = 1. Effect of gradient width is shown.

Stress concentration factor, /P [Pa/N]

22000

n=4

n=2

20000

18000

n=1
n = 0.5
n = 0.2

16000

Epoxy
14000

Graded Region

0.5

Alumina

Relative position across gradient,

Figure A.9: Stress profile along the lower face of a step-graded beam, with gradient width 3 mm.
Effect of gradient profile is shown.

A-32

(a)

Normalised Energy Release Rate, GE/ [m]

Appendix E: Curved cracks in graded specimens: Additional FE results

0.8
Noda, Oda & Ishi
Ioakimidis & Theocaris
Present analysis

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

10

20

30

40

50

60

70

Crack Arc Angle, [deg]

(b)

Figure A.10: Mechanical energy release rate predictions for surface crack with circular arc shape
and crack-tip normal to surface, compared with results of Noda et al. [1994] and Ioakimidis &
Theocaris [1979].

Results are compared with those of Noda et al. in Figures A.10 and A.11, and also with
results of Ioakimidis and Theocaris (1979) in Figure A.9. Results for circular arc-shaped
surface cracks with crack-tip normal to the curved crack are shown in Figure A.10. This
configuration is equivalent to the case of = 0, = 0 and, as in Figure 7(a), good
agreement is observed. Results for cracks comprised of a straight section and a circular arc
section, similar to crack-shape C in Figure 7.4, are shown in Figure A.11. Better
agreement is observed for predictions of mechanical energy release rate than for deflection
A-33

angle predictions. Both models predict that G and will be insensitive to arc-section
length. A very slight decrease in for small arc-section lengths is predicted by the model of
Noda et al., which may be compared with the slight decrease calculated with FE

0.8

60

( = 60)
k

G ( = 60)

0.6

( = 45)
k

G ( = 45)

Predicted Deflection Angle, k [deg]

(a)

Normalised Energy Release Rate, GE/2 [m]

simulations for small radius of curvature in Figures 7.7 and 7.10.

50

40

30

G ( = 30)

k ( = 30)

20
0.4

( = 15) 10
k

G ( = 15)
0

0.2

0.4

0.6

0.8

Relative Arc-Section Length, c 2/c1

(b)

C2 C1

Figure A.11: Mechanical energy release rate predictions (solid lines) for surface crack with linear
section normal to surface and circular arc shaped section, compared with results of Noda et al.
[1994] (dashed lines). Both crack-tip angle and relative length of arc section were varied
independently.

A-34

Appendix F: Additional experimental results

Crack initiation results from Table 10.1, shown graphically in Figure A.12, demonstrate the
good agreement between predictions and measurements.

(a)

800

Measured Initiation Load [N]

700
600
500
400
300
200
100
0
0

100

200

300

400

500

600

700

800

Predicted Initiation Load [N]

70

Measured Initial Deflection Angle [deg]

(b)

60

50

40

30

20

10

0
0

10

20

30

40

50

60

70

Predicted Initial Deflection Angle [deg]

Figure A.12: Predicted and measured values of (a) peak load and (b) deflection angle for crack
initiation under cyclic loading in graded alumina-epoxy specimens. (From Table 10.1, p.10-2)

A-35

References
1)

Aboudi J, Arnold SM, Pindera MJ (1994). Response of functionally graded

composites to thermal gradients. Compos Engrg 4: 1-8.

2)

Aboudi J, Pindera M-J, Arnold SM (1999). Higher-order theory for functionally

graded materials. Composites: B 30(8): 777-832.

3)

Afsar AM, Sekine H (2002). Inverse problems of material distributions for prescribed
apparent fracture toughness in FGM coatings around a circular hole in infinite elastic
media. Compos Sci Technol 60: 1063-1077.

4)

Aliabadi MH (1997). Boundary element formulations in fracture mechanics. Appl

Mech Rev 50: 83-96.

5)

Alpa G, Bozzo E, Gambarotta L (1980). Some observations on the path stability in

fracture propagation for biaxially stressed plates. Engrg Fract Mech 13: 791-799.

6)

Amundsen A (2001). Fracture Mechanics of Functionally Graded Materials, Masters

Thesis, University of New South Wales, School of Materials Science and
Engineering.

7)

Anderson TL (1995). Fracture Mechanics: Fundamentals & Applications, CRC Press,

Boca Raton.

8)

Anderson RM (1992). Testing Advanced Ceramics. Advanced Materials and

Processes 3:89-94.

9)

Anlas G, Santare MH, Lambros J (2000). Numerical calculation of stress intensity

factors in functionally graded materials. Int J Fracture 104: 131-143.

10) Anlas G, Lambros J, Santare MH (2002). Dominance of asymptotic crack tip fields in
elastic functionally graded materials. Int J Fracture 115:193204.
11) Ang WT, Clements DL (1987). On some crack problems for inhomogeneous elastic
materials. Int J Sol Struct 23(8): 1089-1104.
12) ANSYS (2002) Version 6.1 Documentation, ANSYS Inc. Canonsburg, PA.
13) Araki M, Sasaki M, Kim S, Suzuki S, Nakamura K, Akiba M (1994). Thermal
response experiments of SiC/C and TiC/C functionally gradient materials as plasma
facing materials for fusion application. J Nucl Mater 212-215: 1329-1334.
14) Ashby MF, Jones DRH (1996). Engineering Materials 1: An Introduction to their
Properties and Applications, ed 2, Butterworth Heinemann.
R-1

15) Ashby MF, Blunt FJ, Bannister M (1989). Flow characteristics of highly constrained
metal wires. Acta Met 37(7): 847-857.
16) Ashby MF, Brchet YJM (2003). Designing hybrid materials. Acta Mater 51: 5801
5821.
17) ASTM Standard C1259-98. Standard Test Method for Dynamic Youngs Modulus,
Shear Modulus and Poissons Ratio for Advanced Ceramics by Impulse Excitation of
Vibration.
18) ASTM Standard E1820-01. Standard Test Method for Measurement of Fracture
Toughness.
19) ASTM Standard E399-90. Standard Test Method for Plane Strain Fracture Toughness
of Metallic Materials.
20) Atkins AG, Mai Y-W (1985). Elastic and Plastic Fracture. 2nd ed., Ellis-Norwood,
Chichester.
21) Atkinson C, List RD (1978). Steady State Crack Propagation into Media with
Spatially Varying Elastic Properties. Int J Engrg Sci 16: 717-730.
22) Atri RR, Ravichandran KS, Jha SK (1999). Elastic properties of in-situ processed TiTiB composites measured by impulse excitation of vibration. Mater Sci Eng A 271:
150-159.
23) Ayatollahi MR, MJ Pavier, DJ Smith (1998). Determination of T -stress from finite
element analysis for mode I and mixed mode I/II loading. Int J Fracture 91: 283-298.
24) Ayatollahi MR, Pavier MJ, Smith DJ (2002). Mode I cracks subjected to large Tstresses. Int J Fracture 117: 159174.
25) Bahr H-A, Balke H, Fett T, Hofinger I, Kirchhoff G, Munz D, Neubrand A, Semenov
AS, Weiss H-J, Yang YY (2003). Cracks in functionally graded materials. Mater Sci
Eng A 362: 2-16.
26) Balke H, Bahr H-A, Semenov AS, Hofinger I, Husler C, Kirchhoff G, Weiss H-J
(2001). Graded Thermal Barrier Coatings: Cracking due to Laser Irradiation and
Determining of Fracture Toughness. Ceram Trans 114: 205-212.
27) Banichuk NV (1970). Determination of the form of a curvilinear crack by small
parameter technique. Izv. An SSR, Mekhanika Tverdogo Tela, 7(2): 130-137 (in
Russian).
R-2

28) Banks-Sills L (1991). Application of the finite element method to linear elastic
fracture mechanics. Appl Mech Rev 44(10): 447-461.
29) Banks-Sills L, Eliasi R, Berlin Y (2002). Modeling of functionally graded materials in
dynamic analyses. Composites: B 33: 7-15.
30) Bao G, Suo Z (1992). Remarks on crack-bridging concepts. Appl Mech Rev 24: 355366.
31) Bao G, Fan B Evans AG (1992). Mixed mode delamination cracking in brittle matrix
composites. Mech Mater 13:59-66.
32) Bao G, Wang L (1995). Multiple Cracking in Functionally Graded Ceramic/Metal
Coatings. Int J Sol Struct 32(19): 2853-71.
33) Barbero E, Fernndez-Sez J, Navarro C (2000). Statistical analysis of the
mechanical properties of composite materials. Composites: B 31: 375-381.
34) Barenblatt GI (1962a). The mathematical theory of equilibrium cracks in brittle
fracture. Adv Appl Mech, Academic Press, New York, vol 7: 55-129.
35) Barsoum RS (1976). On the use of isoparametric finite elements in linear fracture
mechanics. Int J Numer Meth Eng 10:25-37.
36) Bartolom J, Moya J, Requena J, Llorca J, Anglada M (1998). Fatigue Crack Growth
Behaviour in Mullite/Alumina Functionally Graded Ceramics. J Am Ceram Soc 81(6):
1502-08.
37) Batra RC, Love BM (2005). Crack Propagation due to Brittle and Ductile Failures in
Microporous Thermoelastoviscoplastic Functionally Graded Materials. Engrg Fract
Mech (accepted).
38) Becker TL Jr, Cannon RM, Ritchie RO (2000). An approximate method for residual
stress calculation in functionally graded materials. Mech Mater 32: 85-97.
39) Becker TL Jr, Cannon RM, Ritchie RO (2001). Finite crack-kinking and T-stresses in
functionally graded materials. Int J Sol Struct 38: 5545-5563.
40) Becker TL Jr, Cannon RM, Ritchie RO (2002). Statistical fracture modeling: crack
path and fracture criteria with application to homogeneous and functionally graded
materials. Engrg Fract Mech 69: 1521-1555.
41) Begley MR, McMeeking RM (1995). Numerical analysis of fibre bridging and fatigue
crack growth in metal matrix materials. Mater Sci Eng A 200: 12-20.
R-3

42) Belytschko T, Black T (1999). Elastic crack growth in finite elements with minimal
remeshing. Int J Numer Meth Eng 45, 601-620.
43) Belytschko T, Lu YY, Gu L (1994). Element-free Galerkin methods. Int J Numer
Meth Eng 37: 229-256.
44) Belytschko T, Krongauz Y, Organ D, Fleming M, Krysl P (1996). Meshless methods:
An overview and recent developments. Comput Meth Appl Mech Eng 139: 3-47.
45) Berg A, Wagner L (1999). Near-surface Gradient Microstructures in Metastable BetaTitanium Alloys for Improved Fatigue Performance. Mat Sci Forum 308-311: 301312.
46) Berger C, Eulitz K-G, Heuler P, Kotte K-L, Naundorf H, Schuetz W, Sonsino C.M.,
Wimmer A, Zenner H (2002). Betriebsfestigkeit in Germany - an overview. Int J
Fatigue 24: 603-625.
47) Bergkvist H, Guex L (1979). Curved crack propagation. Int J Fracture 15(5): 429-41.
48) Bian L-C, Kim KS (2004). The minimum plastic zone radius criterion for crack
initiation direction applied to surface cracks and through-cracks under mixed mode
loading. Int J Fatigue 26: 1169-78.
49) Bigerelle M, Iost A (1999). Bootstrap analysis of FCGR, application to the Paris
relationship and to lifetime prediction. Int J Fatigue 21: 299307.
50) Bilby BA, Cardew GE (1975). The crack with a kinked tip. Int J Fracture 11: 708-711.
51) Bilotsky YD, Gasik MM (1997). Gauge field theory for functional graded materials
and components. Composites: B 28: 121-125.
52) Bittencourt TN, Barry A, Ingraffea AR. Comparison of mixed-mode stress intensity
factors obtained through displacement correlation, J-integral formulation and modified
crack closure integral. Fracture Mechanics Vol II ASTM STP 1131, 1992, p. 69-82.
53) Bittencourt TN, Wawrzynek PA, Ingraffea AR, Sousa JLA (1996). Quasi-automatic
simulation of crack propagation for 2D LEFM problems. Engrg Fract Mech 55: 321324.
54) Blanks KS, Kristofferson A, Carlstrom E, Clegg WJ (1998). Crack deflection in
ceramic laminates using porous interlayers. J Europ Ceram Soc 18: 1945-51.

R-4

55) Bleek O, Munz D, Schaller W, Yang YY (1998). Effect of a Graded Interlayer on the
Stress Intensity Factor of Cracks in a Joint under Thermal Loading. Engrg Fract Mech
60(5-6): 615-623.
56) Blmm M, Dollmeier K, Ilschner B (1995). Experimental Investigation of FatigueCrack Propagation in Room Temperature Bending Tests with Cu-Ni Graded Alloys.
Ilschner B, Cherradi N (eds), Proceedings 3rd International Symposium on Structural
and Functionally Graded Materials, Presses Poly. et Uni. Romandes, Lausanne, 315320.
57) Boresi A, Chong K (1987). Elasticity in Engineering Mechanics. 2nd ed., Elsevier,
New York.
58) Bose K, Mataga PA, Ponte Castaneda P (2004). Stable crack growth along a brittleductile interface II: Small scale yielding solutions and interfacial toughness
predictions. Int J Sol Struct 25: 1-23.
59) Bouchard PO, Bay F, Chastel Y (2003). Numerical modelling of crack propagation:
automatic remeshing and comparison of different criteria. Comput Meth Appl Mech
Eng 192: 3887-3908.
60) Broek D (1991). Elementary Engineering Fracture Mechanics, 4th ed., Kluwer,
Dordrecht.
61) Bruck HA, Rabin BH (1999). Evaluating microstructural and damage effects in ruleof-mixtures predictions of the mechanical properties of Ni-Al2O3 composites for use
in modeling functionally graded materials. J Mater Sci 34: 2241-2251.
62) Budiansky B (1965). On the elastic moduli of some heterogeneous materials. J Mech
Phys Sol 13: 223-227.
63) Budiansky B, O'Connell RJ (1976). Elastic Moduli of a Cracked Solid. Int J Sol
Struct 12: 81-97.
64) Budiansky B, Hutchinson JW, Evans AG (1986). Matrix Fracture in Fibre-Reinforced
Ceramics. J Mech Phys Sol 34: 167-89.
65) Budiansky B, Amazigo JC (1989). Toughening by Aligned Frictionally Constrained
Fibres. J Mech Phys Solids 37(1): 93-109.
66) Bueckner H (1970). A novel principle for the computation of stress intensity factors.
ZAMM 50: 529-546.
R-5

67) Butcher RJ, Rousseau C-E, Tippur HV (1999). A Functionally Graded Particulate
Composite: Preparation, Measurements and Failure Analysis. Acta mat 47(1): 259268.
68) Cai H, Bao G (1998). Crack Bridging in Functionally Graded Coatings. Int J Sol
Struct 35(7-8): 701-17.
69) Cannillo V, Carter WC (2001). Functionally Graded Materials: Effect of Elastic
Heterogeneity on the Toughness. In: Ravi-Chandar K, Karihaloo BL, Kishi T, Ritchie
RO, Yokobori AT, Yokobori T (eds). Advances in Fracture Research, Proc. ICF10;
Pergamon CDROM.
70) Cannon RM, Dalgleish BJ, Dauskhardt RH, Oh TS, Ritchie RO (1991). Cyclic
fatigue-crack propagation along ceramic/metal interfaces. Acta met mat 39(9): 214556.
71) Cao HC, Thouless MD, Evans AG (1988). Residual stresses and cracking in brittle
solids bonded with a thin ductile layer. Acta met 36(8): 2037-46.
72) Cao HC, Evans AG (1991). On crack extension in ductile/brittle laminates. Acta
metall mater 39(12): 2997-3005.
73) Cao J-W, Sakai M (1996). Crack-face fibre bridging: Finite element analysis,
analytical model and experimental result. J Mater Res 11(6): 1537-44.
74) Carpenter RD, Liang WW, Paulino GH, Gibeling JC, Munir ZA (1999). Fracture
Testing and Analysis of a Layered Functionally Graded Ti/TiB Beam in 3-Point
Bending. Mater Sci Forum 308-311: 837-842.
75) Carrillo-Heian EM, Carpenter RD, Paulino GC, Gibeling JC, Munir ZA (2001).
Dense Layered Molybdenum Disilicide-Silicon Carbide Functionally Graded
Composites Formed by Field Activated Synthesis. J Am Ceram Soc 84(5): 962-68.
76) Chaboche JL, Maire JF (2001). New progress in micromechanics-based CDM models
and their application to CMCs. Compos Sci Tech 61: 2239-46.
77) Chaboche J-L, Maire J-F (2002). A new micromechanics based CDM model and its
application to CMCs. Aero Sci Tech 6: 131-145.
78) Chan HM (1997). Layered Ceramics: Processing and Mechanical Behavior. Ann Rev
Mater Sci 27: 249-82.

R-6

79) Chan Y-S, Paulino GC, Fannjiang AC (2001). Gradient Elasticity Theory for Mode I
Crack in Functionally Graded Materials. Ceram Trans 114: 731-38.
80) Chandra N, Li H, Shet C, Ghonem H (2002). Some issues in the application of
cohesive zone models for metal-ceramic interfaces. Int J Sol Struct 39: 2827-55.
81) Chapa JG, Reimanis IE (2002). Modeling of thermal stresses in a graded Cu/W joint.
Journal of Nuclear Materials 2002; 303(2-3): 131-136.
82) Chapa-Cabrera JG, Rozenburg K, Reimanis IE, Steffler ED (2001). Fracture in
Ductile/Brittle Graded Composites. Ceram Trans 114: 797-804.
83) Chapa-Cabrera JG, Reimanis IE (2001). Crack deflection in layered, graded
composites. In: Ravi-Chandar K, Karihaloo BL, Kishi T, Ritchie RO, Yokobori AT,
Yokobori T (eds). Advances in Fracture Research, Proc. ICF10 Pergamon CDROM.
84) Chapa-Cabrera J, Reimanis IE (2002a). Effects of Residual Stress and Geometry on
Predicted Crack Paths in Graded Composites. Engrg Fract Mech 69: 1667-78.
85) Chapa-Cabrera JG, Reimanis IE (2002b). Crack propagation in CuW FGM. Phil
Mag A 82(17/18): 3393-3403.
86) Chapa-Cabrera JG (2002). Fracture and Deformation in Cu/W Graded Joints. PhD
Thesis, Colorado School of Mines, Golden, CO, USA.
87) Charalambides PG, Mataga PA, McMeeking RM, Evans AG (1990). Steady-state
mechanics of a growing crack paralleling an elastically constrained thin ductile layer.
Appl Mech Rev 43(5) II: S267-S270.
88) Chen YZ, Gross D, Huang YJ (1991). Numerical Solution of the Curved Crack
Problem by means of Polynomial Approximation of the Dislocation Distribution.
Engrg Fract Mech 29(5): 791-97.
89) Chen YZ (1993). New Fredholm integral equation for multiple crack problem in
plane elasticity and antiplane elasticity. Int J Fracture 1993; 64(1):63-77.
90) Chen YZ (1994). New Singular Integral Equations for Circular Arc Crack and Rigid
Line Problem. Engng Fract Mech 1994; 41(1):139-145.
91) Chen YF, Erdogan F (1996). The interface crack problem for a nonhomogeneous
coating bonded to a homogeneous substrate. J Mech Phys Sol 44: 771-87.
92) Chen YZ (1999). Stress intensity factors for curved and kinked cracks in plane
extension. Theor Appl Frac Mech 31:223-232.
R-7

93) Chen YZ (2000). Closed form solutions of T-stress in plane elasticity crack problems.
Int J Sol Struct 37: 1629-37.
94) Chen J, Wu L, Du S (2000a). A Modified J Integral for Functionally Graded
Materials. Mech Res Comm 27(3): 301-306.
95) Chen J, Wu L, Du S (2000b). Element-Free Galerkin Method for Fracture of
Functionally Graded Materials. Key Engrg Mat 183-187: 487-492.
96) Chen C-S, Krause R, Pettit RG, Banks-Sills L, Ingraffea AR (2001). Numerical
assessment of T-stress computation using a p-version finite element method Int J
Fracture 107: 177199.
97) Chen K-Z, Feng X-A (2004). CAD modeling for the components made of multi
heterogeneous materials and smart materials Comput-Aid Design 36: 51-63.
98) Cherradi N, Kawasaki A, Gasik M (1994). Worldwide Trends in Functional Gradient
Materials Research and Development. Compos Engrg 4(8): 883-94.
99) Cho JR, Ha DY (2000). Dual-Phase Functionally Graded Composite Materials:
Overall and Discrete Analysis Models. Key Engng Mater 183-187: 373-378.
100) Cho JR, SW Shin (2004). Material composition optimization for heat-resisting FGMs
by artificial neural network. Composites: A 35: 585-594.
101) Choi SR, Salem JA (1995). Dynamic, static and cyclic fatigue of alumina with
indentation-induced flaws. J Mat Sci Lett 14(18): 1286-88.
102) Christensen RM (1990). A critical evaluation for a class of micro-mechanics models.
J Mech Phys Sol 6: 379-404.
103) Chu C, Zhu J, Yin Z, Lin P (2001). Structure optimization and properties of
hydroxyapatite-Ti symmetrical functionally graded biomaterial. Mater Sci Eng A 316:
205-10.
104) Chung TJ, Neubrand A, Rdel J, Fett T (2001). Fracture Toughness and R-Curve
Behaviour of Al2O3/Al FGMs. Ceram Trans 114: 789-96.
105) Chung TJ, Neubrand A, Rdel J (2002). Effect of Residual Stress on the Fracture
Toughness of Al2O3/Al Gradient Materials. Key Engrg Mater 206-213: 965-68.
106) Clark DR (1992). Interpenetrating Phase Composites. J Am Ceram Soc 75(4):739-59.
107) Clegg WJ (2000). Layered Ceramic Structures. Comprehensive Composite Materials
Volume 4; (ISBN: 0-080437222); Chapter 4.15 pp. 449-470.
R-8

108) Cichocki FR, Trumble KP, Rdel J (1998). Tailored Porosity Gradients via Colloidal
Infiltration of Compression Molded Sponges. J Am Ceram Soc 81(6): 1661-64.
109) Connell SJ, Zok FW (1997). Measurement of the cyclic bridging law in a titanium
matrix composite and its application to simulating crack growth. Acta mat 45(12):
5203-11.
110) Cotterell B, Rice JR (1980). Slightly curved or kinked cracks. Int J Fracture 16: 15569.
111) Cotterell B (1965). Notes on the Paths and Stability of Cracks. Int J Fracture 1966; 2:
526-533.
112) Cotterell B, Mai Y-W (1996). Fracture Mechanics of Cementitious Materials. Blackie
Academic and Professional, Glasgow, UK.
113) Cotterell B (2002). The past, present, and future of fracture mechanics. Engrg Fract
Mech 69: 533-553.
114) Cox BN (1991). Extrinsic factors in the mechanics of bridged cracks. Acta met mat
39: 1189-1201.
115) Cox BN (1993). Scaling relations for bridged cracks. Mech Mater 15: 87-98.
116) Cox BN, Marshall DB (1991a). Stable & Unstable Solutions for Bridged Cracks in
Various Specimens. Acta metall mater 39(4): 579-89.
117) Cox BN, Marshall DB (1991b). The determination of crack bridging forces. Int J
Fracture 49:159-76.
118) Cox BN, Marshall DB (1991c). Crack bridging in the fatigue of fibrous composites.
Fat Fract Engrg Mat Struct 14: 847-61.
119) Cox BN, Marshall DB (1994). Concepts for bridged cracks in fracture and fatigue.
Acta met mat 42: 341-63.
120) Cox BN, Lo CS (1992a). Load ratio, notch and scale effects for bridged cracks in
fibrous composites. Acta met mat 40(1): 69-80.
121) Cox BN, Lo CS (1992b). Simple approximations for bridged cracks in fibrous
composites. Acta met mat 40(7): 1487-96.
122) Cox BN, Rose LRF (1994). Time- or cycle-dependent crack bridging. Mech Mater
19(1): 39-57.

R-9

123) Cox BN, Rose LRF (1996). A self-consistent approximation for crack-bridging by
elastic/perfectly plastic ligaments. Mech Mater 22(4): 249-63.
124) Daehn GS, Starck B, Xu L, Elfishawy KF, Ringnalda J, Fraser HL (1996). Elastic and
Plastic Behavior of a Co-Continuous Alumina/Aluminum Composite. Acta mat 44(1),
249-61.
125) Dalgleish BJ, Trumble KP, Evans AG (1989). The strength and fracture of alumina
bonded with aluminium alloys. Acta met 36(7): 1923-31.
126) Damani R, Gstrein R, Danzer R (1996). Critical Notch-Root Radius Effect in SENBS Fracture Toughness Testing. J Eur Ceram Soc 16: 695-702.
127) Dao M, Gu P, Maewal A, Asaro RJ (1997). A micromechanical study of residual
stresses in functionally graded materials. Acta mater 45(8): 3265-76.
128) Dauskhardt RH (1993). A frictional wear mechanism for fatigue crack growth in
crack bridging. Acta met mat 41: 2165-81.
129) De Borst R, Remmers J, Needleman A (2004). Computational aspects of cohesive
zone models. Proc. 15th European Conference on Fracture, Stockholm.
130) Delale F, Erdogan F (1983). The Crack Problem for a Nonhomogeneous Plane. J Appl
Mech 50: 609-14.
131) Delale F, Erdogan F (1988a). Interface crack in a nonhomogeneous medium. Int J
Engng Sci 26: 559-68.
132) Delale F, Erdogan F (1988b). On the mechanical modelling of the interface region in
bonded materials. J Appl Mech 55: 317-24.
133) Deng X (1995). Plane strain near-tip fields for elastic-plastic interface cracks. Int J Sol
Struct 32(12): 1727-41.
134) Dhaliwal RS, Singh BM (1978). On the Theory of Elasticity of a Non-homogeneous
Medium. J Elasticity 8: 211-19.
135) Dhirendra VK, Narasimhan R (1998). Mixed-mode steady-state crack growth in
elastic-plastic solids. Engrg Fract Mech 59(4): 543-59.
136) DIN51109 Testing of advanced technical ceramics: Determination of fracture
toughness KIc , 1999.
137) Dolbow JE, Mos N, Belytschko T (2000). Discontinuous enrichment in finite
elements with a partition of unity method. Fin Elem Anal Design 36: 235-60.
R-10

138) Dolbow JE, Gosz M (2002). On the computation of mixed-mode stress intensity
factors in functionally graded materials. Int J Sol Struct 39: 2557-74.
139) Dolbow JE, Nadeau JC (2002). On the use of effective properties for the fracture
analysis of microstructured materials. Engrg Fract Mech 69: 1607-34.
140) Dong M, Lele P, Weber U, Schmauder S (1999). Mesomechanical Modelling of
Composites Containing FGM-related Interpenetrating Microstructures Based on
Micromechanical Matricity Models. Mat Sci Forum 308-311: 1000-05.
141) Dugdale DS (1960). Yielding of steel sheets containing slits. J Mech Phys Sol 8: 1008.
142) Dundurs J (1969). Edge-bonded dissimilar orthogonal elastic wedges. J Appl Mech
36: 650-52.
143) Dvorak G, Zuiker J (1994). The effective properties of functionally graded
composites I. Extension of the Mori-Tanaka method to linearly varying fields.
Compos Eng 4: 19-35.
144) Dvorak GJ (2000). Composite materials: Inelastic behaviour, damage, fatigue and
fracture. Int J Sol Struct 37: 155-170.
145) Einstein A (1906). Eine Neue Bestimmung der Molekuldimensionen. Annalen der
Physik 19: 289-306. (Translated in: Investigations on the Theory of Brownian Motion,
Dover, 1956: 36-62.
146) Eischen JW (1987a). Fracture of nonhomogeneous materials. Int J Fracture 34: 3-22.
147) Eischen J (1987b). An improved method for computing the J2 integral. Engrg Fract
Mech 26(5): 691-700.
148) Elber W (1974). Effects of shot peening on residual stresses on the fracture and
crack-growth properties of D6ac steel. Amer Soc Test Mater 559: 45-58.
149) El-Borgi S, Erdogan F, Hidri L (2001a). Stress Intensity Factors for a Crack
Arbitrarily Oriented in a Functionally Graded Layer. Ceram Trans 114: 723-30.
150) El-Borgi S, Erdogan F, Hila H, Smaoui H (2001b). Thermo-mechanical Stress
Intensity Factors for a Partially Insulated Crack in a Functionally Graded Medium.
Ceram Trans 114: 755-62.
151) El-Shaer Y, Derby B (2004). Modelling of R-curve behaviour in ceramicmetal
laminates. Mater Sci Eng A 365: 196-201.
R-11

152) England AH (1965). A Crack Between Dissimilar Media. J Appl Mech 32: 400-02.
153) Epstein M, de Len M (2000). Homogeneity without uniformity: towards a
mathematical theory of functionally graded materials. Int J Sol Struct 37: 7577-91.
154) Erdogan F, Sih GC (1963). On the crack extension in plates under plane loading and
transverse shear. J Basic Eng ASME Trans 85: 519-25.
155) Erdogan F, Gupta GD, Cook TS (1973). Numerical Solution of Singular Integral
Equations. in: Sih GC (ed). Mechanics of Fracture 1. Noordhoff, Leiden.
156) Erdogan F (1978). Mixed Boundary Value Problems in Mechanics. Mech Today 4:186.
157) Erdogan F (1983). Stress Intensity Factors. J Appl Mech 50: 992-1002.
158) Erdogan F (1985). The crack problem for bonded nonhomogeneous materials under
antiplane shear loading. J Appl Mech 52: 823-28.
159) Erdogan F (1995a). Fracture Mechanics of Functionally Graded Materials. Materials
Research Society Bulletin 20(1): 43-44.
160) Erdogan F (1995b). Fracture Mechanics of Functionally Graded Materials. Compos
Eng 5(7): 753-70.
161) Erdogan F, Ozturk M (1995). Periodic cracking of functionally graded coatings. Int J
Engrg Sci 33(15): 2179-2195.
162) Erdogan F, Wu BH (1997). The surface crack problem for a plate with functionally
graded properties. J Appl Mech 64: 449-56.
163) Erdogan F (2000). Fracture Mechanics. Int J Sol Str 2000; 37: 171-183.
164) Erdogan F, Dag S (2001). Crack and Contact Problems in Functionally Graded
Materials. Ceram Trans 114: 739-46.
165) Ergven ME, Gross D (1999). On the penny-shaped crack in homogeneous elastic
materials under normal extension. Int J Sol Struct 36: 1869-82.
166) Eriksson K. A domain independent integral expression for the crack extension force
of a curved crack in three dimensions. J Mech Phys Sol 2002;50:381-403.
167) Eshelby JD (1957). The determination of the elastic field of an ellipsoidal inclusion,
and related problems. Proc Roy Soc A 349: 376-396.
168) Evans AG, Rhle M, Dalgleish BJ, Charalambides PG (1990). The fracture energy of
bimaterial interfaces. Mater Sci Eng A 126:53-64.
R-12

169) Feng X-Q, Mai Y-W, Qin Q-H (2003). A micromechanical model for interpenetrating
multiphase composites. Comput Mater Sci 28: 486493.
170) Feng X-Q, Tian Z, Liu Y-H, Yu S-W (2004). Effective Elastic and Plastic Properties
of Interpenetrating Multiphase Composites. Appl Compos Mater 11: 33-55.
171) Ferrari M, Rooney F, Nadeau JC (1999). Optimal FGMs and Plain Awful
Composites. Mater Sci Forum 308-311: 989-94.
172) Fett T (1992). Procedure for the determination of the weight function for semielliptical surface cracks by direct adjustment to reference solutions. Engrg Fract Mech
43(4): 519-528.
173) Fett T (1998). T-stresses in rectangular plates and circular disks. Engrg Fract Mech
60(5-6): 631-52.
174) Fett T, Munz D (1992). Determination of fracture toughness at high temperatures
after subcritical crack extension. J Am Ceram Soc 75(11): 3133-36.
175) Fett T, Munz D (1995). Bridging stress relations for ceramic materials. J Eur Ceram
Soc 15: 377-83.
176) Fett T, Munz D (1997). A weight function for cracks in gradient materials. Int J
Fracture 84: L3-L7.
177) Fett T, Munz D (2002). Kinked cracks and the Richard fracture criterion. Int J
Fracture 115: L69-73.
178) Fett T, G Gerteisen, S Hahnenberger, G Martin, D Munz (1995). Fracture Tests for
Ceramics under Mode-I, Mode-II and Mixed-Mode Loading. J Euro Ceram Soc 15:
307-12.
179) Fett T, Munz D, Yang YY (2000a). Applicability of the Petroski-Achenbach weight
function procedure to graded materials. Engrg Fract Mech 65: 393-403.
180) Fett T, Munz D, Yang YY (2000b). Direct adjustment procedure for weight functions
of graded materials. Fat Fract Engrg Mat Struct 23: 191-98.
181) Fett T, Munz D, Geraghty RD, White KW (2000c). Influence of specimen geometry
and relative crack size on the R-curve. Engrg Fract Mech 66: 375-386.
182) Fett T, D Munz, X Dai, KW White (2000d). Bridging stress relation from a combined
evaluation of the R-curve and post-fracture tensile tests. Int J Fracture 104: 375-385.
183) FGM 2000, Proceedings, 6th Int Symp on FGMs. Ceramic Trans 2001; 114.
R-13

184) FGM 1998, Proceedings, 5th Int Symp on FGMs 1998. Mat Sci Forum 1999; 308-311.
185) Fish J, Yu Q (2002). Computational mechanics of fatigue and life predictions for
composite materials and structures. Comput Meth Appl Mech Eng 191: 4827-49.
186) Fitzpatrick ME, Hutchings MT, Withers PJ (1999). Separation of measured fatigue
crack stress fields in a metal matrix composite material. Acta mater 47(2) 585-593.
187) Fleck NA, Kang KJ, Ashby MF (1994). The cyclic properties of engineering
materials. Acta met mat 42(2): 365-381.
188) Fleck NA, Hutchinson JW, Suo Z (1991). Crack path selection in a brittle adhesive
layer. Int J Sol Struct 27(13): 1683-1703.
189) Foote RML, Mai YM, Cotterell B (1987). Crack growth resistance curves in strainsoftening materials. J Mech Phys Solids 34(6): 593-607.
190) Forth SC, Favrow LH, Keat WD, Newman JA (2003). Three-dimensional mixedmode fatigue crack growth in a functionally graded titanium alloy. Engrg Fract Mech
70: 2175-85.
191) Freund LB (1993). Stress distribution and curvature of a general compositionally
graded semiconductor layer. J Crystal Growth 132(1-2): 341-344.
192) Fujimoto T, Noda N (2001). Two crack growths in a functionally graded plate under
thermal shock. J Thermal Stresses 24: 847-62.
193) Fukui Y, Yamanaka N, Yoshihiro E (1997). Bending strength of an Al-Al3Ni
functionally graded material. Composites: B 28:37-43.
194) Gao H (1991). Fracture analysis of non-homogeneous materials via a moduliperturbation approach. Int J Sol Struct 27(13): 1663-82.
195) Garcia DE, Schicker S, Janssen R, Claussen N (1998). Nb- and Cr-Al2O3 Composites
with Interpenetrating Networks. J Euro Ceram Soc 18: 601-05.
196) Gasik MM (1998). Micromechanical modelling of functionally graded materials.
Comput Mater Sci 13: 42-55.
197) Geraghty RD, Hay JC, White KW (1999). Fatigue degradation of the crack wake
zone in monolithic alumina. Acta mat 47(4): 1345-53.
198) Gerasoulis A, Srivastav RP (1980). A Griffith crack problem for a nonhomogeneous
medium. Int J Engrg Sci 18: 239-47.

R-14

199) Geubelle PH, Knauss WG (1994). Crack propagation at and near bimaterial
interfaces: linear analysis. J Appl Mech 61: 560-66.
200) Gibson RE (1967). Some results concerning displacements and stresses in a
nonhomogeneous elastic half space. Geotechnique 17: 58-67.
201) Gilbert CJ, Dauskhardt RH, Ritchie RO (1997). Microstructural mechanisms of
fatigue crack propagation in grain-bridging ceramics. Ceram Int 23: 413-18.
202) Gilbert CJ, Ritchie RO (1998). On the Quantification of Bridging Tractions during
Subcritical Crack Growth under Monotonic and Cyclic Fatigue Loading in a GrainBridging Silicon Carbide Ceramic. Acta mat 46(2): 609-16.
203) Gilbert CJ, Han YS, Kim DK, Ritchie RO (2000). Anomalous cyclic fatigue-crack
propagation behavior of small cracks in monolithic, grain-bridging ceramics. Ceram.
Int 26: 721-25.
204) Goldstein RV, Salganik RL (1970). Plane problem of curvilinear cracks in an elastic
solid. Izv. An SSR, Mekhanika Tverdogo Tela 7(3): 69-82 (in Russian).
205) Goldstein RV, Salganik RL (1974). Brittle fracture of solids with arbitrary cracks. Int
J Fracture 10(4): 507-523.
206) Gosz M, Dolbow J, Moran B (1998). Domain Integral Formulation for Stress Intensity
Factor Computation along Curved Three-Dimensional Interface Cracks. Int J Sol
Struct 35(15): 1763-83.
207) Govindarajan S, Moore JJ, Mishra B, Olson DL, Ohno T, Disam J (1996). On the
possibility of tailoring a compositional gradient in thin films sputtered from a
MoSi/xSiC composite target. Surf Coat Technol 86-87: 33-40.
208) Greco F, Lonetti P, Zinno R (2002). An analytical delamination model for
delaminated plates including bridging effects. Int J Sol Struct 39: 2435-63.
209) Griffith AA (1920). The phenomenon of rupture and flow in solids. Phil Trans Royal
Soc 45: 251-266.
210) Grujicic M, Zhao H (1998). Optimization of 316L stainless steel/alumina functionally
graded material for reduction of damage induced by thermal residual stress. Mater.
Sci. Eng. A 252: 117-32.

R-15

211) Grujicic M, Zhang Y (1998). Determination of effective elastic properties of

functionally graded materials using a Voronoi cell finite element method. Mater Sci
Eng A 251: 6476.
212) Gu P, Asaro RJ (1997a). Cracks in Functionally Graded Materials. Int J Sol Struct
34(1): 1-17.
213) Gu P, Asaro RJ (1997b). Crack Deflection in Functionally Graded Materials. Int J Sol
Struct 34(24): 3085-98.
214) Gu P, Dao M, Asaro RJ (1999). A Simplified Method for Calculating the Crack-Tip
Field of Functionally Graded Materials Using the Domain Integral. J Appl Mech 66:
101-108.
215) Guagliano M, Vergani L. Mode I stress intensity factors for curved cracks in gears by
a weight functions method. Fatig Fract Eng Mat Struct 2001; 23: 41-52.
216) Guler MA, Erdogan F (2004). Contact mechanics of graded coatings. Int J Sol Struct
41(14): 3865-89.
217) Gumbsch P (2001). Modelling brittle and semi-brittle fracture processes. Mater Sci
Eng A 319321: 17.
218) Guo ZK, Kobayashi AS, Hay JC, White KW (1999). Fracture process zone modeling
of monolithic Al2O3. Engrg Fract Mech 63: 115-129.
219) Han B, Post D, Ifju P (2001). Moir interferometry for engineering mechanics:
current practices and future developments. J Strain Analysis 36(1): 101-117.
220) Hasegawa M, Zhu SJ, Kagawa Y, Evans AG (2003). Effect of metal layer thickness
on the decohesion of high purity coppersapphire interfaces. Acta mater 51: 5113-21.
221) Hashin Z (1983). Analysis of Composite Materials A Survey. J Appl Mech 50: 481505.
222) Hashin Z (1962). The Elastic Moduli of Heterogeneous Materials. J Appl Mech 29:
143-50.
223) Hashin Z, Shtrikman S (1963). A Variational Approach to the Theory of the Elastic
Behaviour of Multiphase Materials. J Mech Phys Solids 11: 127-40.
224) Hay JC, White KW (1993). Grain-bridging mechanisms in monolithic alumina. J Am
Ceram Soc 76: 1849-54.

R-16

225) Hay JC, White KW (1997). Microstructural interaction in the wake cohesive zone for
small crack-opening displacements. Acta mat 45(9): 3625-33.
226) Hayashi K, Nemat-Nasser S (1981). Energy-Release Rate and Crack Kinking under
Combined Loading. J Appl Mech 48: 520-524.
227) He M-Y, Hutchinson JW (1989). Kinking of a crack out of an interface. J Appl Mech
56: 270-78.
228) He M-Y, Hutchinson JW (1989b). Crack deflection at an interface between dissimilar
elastic materials. Int J Sol Struct 25(9): 1053-67.
229) He MY, Cao HC, Evans A (1990). Mixed-mode fracture: The four-point shear
specimen. Acta met mat 38(5): 839-46.
230) He MY, Evans AG, Curtin WA (1993). The ultimate tensile strength of metal and
ceramic-matrix composites. Acta metall mater 41(3): 871-8.
231) Henager Jr. CH, Brimhall JL, Brush LN (1995). Tailoring structure and properties of
composites synthesized in situ using displacement reactions. Mater Sci Eng A 195:
65-74.
232) Henshell RD, Shaw KG (1975). Crack tip finite elements are unnecessary. Int J Num
Meth Eng 9: 495-507.
233) Heritage K, Frisby C, Wolfenden A (1988). Impulse excitation technique for dynamic
flexural measurements at moderate temperature. Rev Sci Instr 59(6): 973-74.
234) Hill R (1965). A self-consistent mechanics of composite materials. J Mech Phys Sol
13: 213-22.
235) Hill MR, Carpenter RD, Paulino GH, Munir ZA, Gibeling JC (2002). Fracture testing
of a layered functionally graded material. In: Fracture Resistance testing of monolithic
and composite brittle materials. ASTM STP 1409 JA Salem, GD Quinn and MG
Jenkins (eds) ASTM, West Conshohocken, PA.
236) Hill MR, Panontin TL (1998). How residual stresses affect prediction of brittle
fracture. K.K. Yoon, S. Bhandari, D. Moinereau M.B. Ruggles, T. Shimakawa, S.
Kubo (eds).Proceedings, ASME/JSME Pressure Vessels & Piping Conf, San Diego.
237) Hillerborg A, Modeer M, Petersson PE (1976). Analysis of crack formation and
crack growth in concrete by means of fracture mechanics and finite elements, Cem.
Concr. Res. 6: 773-82.
R-17

238) Hirano T, Yamada T, Teraki J, Niino M, Kumakawa A (1988). A study on a

functionally graded material design system for a thrust chamber. Proc. 16th Int Symp
on Space Tech and Sci, Sapporo, Japan.
239) Hirano T, Wakashima K (1995). Mathematical Modelling and Design of FGMs.
MRS Bulletin 20(1): 40-42.
240) Hoffman M, Fiedler B, Emmel T, Prielipp H, Claussen N, Gross D, Rdel J (1997).
Fracture Behaviour in Metal Fibre Reinforced Ceramics. Acta mat 45(9): 3609-23.
241) Hoffman M, Skirl S, Pompe W, Rdel J (1999). Thermal Residual Strains and
Stresses in Al2O3/Al Composites with Interpenetrating Networks. Acta Mat 47(2):
565-77.
242) Hoffman M, Kidson L, Kelly D, Deneke C (2001). Effect of Crack Growth
Resistance upon Fracture of Ceramic/Polymer Graded Interfaces. In: Ravi-Chandar K,
Karihaloo BL, Kishi T, Ritchie RO, Yokobori AT, Yokobori T (eds). Advances in
Fracture Research, Proc. ICF10, Pergamon CDROM.
243) Hofinger I, Bahr H-A, Balke H, Kirchhoff G, Husler C, Wei H-J (1999). Fracture
Mechanical Modelling and Damage Characterisation of Functionally Graded Thermal
Barrier Coatings by Means of Laser Irradiation. Mater Sci Forum 308-311: 450-56.
244) Howard IC (1978). Simple approximate results for the stress intensity factors at the
tip of a kinked crack. Int J Fracture 14: R307-R310.
245) Hu X-Z, Wittmann FH (1989). Fracture process zone and Kr-curve of hardened
cement paste and mortar. In: Shah SP, Swartz SE, Barr B, editors. Fracture of concrete
and rock: Recent developments. Elsevier Applied Science Publishers, London/New
York, 307-16.
246) Hu X-Z, Wittmann FH (1990). Experimental method to determine extension of
fracture process zone. J Mater Civ Engrg 2: 459-64.
247) Hu X-Z, Mai Y-W (1992). Crack-bridging analysis for alumina ceramics under
monotonic and cyclic loading. J Am Ceram Soc 75(4): 848-53.
248) Hu SY, Li YL, Munz D, Yang YY (1998). Thermal stresses in coated structures. Surf
Coat Technol 99(1-2): 125-31.
249) Huang Y, Zhang HW (1995). The role of metal plasticity and interfacial strength in
the cracking of metal/ceramic laminates. Acta metall mater 43(4): 1523-30.
R-18

250) Huang Y, Hu KX, Chandra A (1994). A generalised self-consistent mechanics

method for microcracked solids. J Mech Phys Sol 42(8): 1273-91.
251) Huang Z-M (2001). Simulation of the mechanical properties of fibrous composites by
the bridging micromechanics model. Composites: A 32: 143-72.
252) Huang G-Y, Wang Y-S, Gross D (2003). Fracture analysis of functionally graded
coatings: plane deformation. Euro J Mech A 22: 535-44.
253) Hutchinson JW (1968). Plastic stress and strain fields at a crack tip. J Mech Phys Sol
16: 337-47.
254) Hutchinson JW, Suo Z (1992). Mixed mode cracking in layered materials. Adv Appl
Mech 29: 63-191.
255) Hutchinson JW, Evans AG (2000). Mechanics of materials: Top-down approaches to
fracture. Acta mat 48: 125-35.
256) Hutchinson JW, Mear ME, Rice JR (1987). Crack paralleling an interface between
dissimilar materials. J. Appl. Mech. 54: 828-832.
257) Hwu KL, Derby B (1999). Fracture of metal/ceramic laminates II. Crack growth
resistance and toughness. Acta mater 47(2): 545-63.
258) Ibijola EA (2002). On some fundamental concepts of Continuum Damage Mechanics.
Comput Meth Appl Mech Eng 191: 1505-20.
259) Ilschner B (1996). Processing-microstructure-property relationships in graded
materials. J Mech Phys Sol 44(5): 647-56.
260) Inglis CE (1913). Stresses in a plate due to the presence of cracks and sharp corners.
Trans Inst Nav Arch 55: 219-41.
261) Ioakimidis NI, Theocaris PS (1979). A System of Curvilinear Cracks in an Isotropic
Elastic Half-plane. Int J Fracture 15(4): 299-307.
262) Irwin GR (1957). Analysis of stresses and strains near the end of a crack traversing a
plate. J Appl. Mech. 1957, 24, 361-364.
263) Itoh Y, Kashiwaya H (1992). Residual stress characteristics of Functionally Graded
Materials. J Ceram Soc Jap 100: 481-6.
264) Jacobsen TK, Srensen BF (2001). Mode I intra-laminar crack growth in composites
modelling of R-curves from measured bridging laws. Composites: A 32: 1-11.

R-19

265) Jain N, Rousseau CE, Shukla A (2004). Crack-tip stress fields in functionally graded
materials with linearly varying properties. Theor Appl Fract Mech 42: 155-170.
266) Jedamzik R, Neubrand A, Rdel J (1999). Characterisation of electrochemically
processed graded tungsten/copper composites. Mater Sci Forum 308-311: 782-7.
267) Jedamzik R, Neubrand A, Rdel J (2000). Functionally graded materials by
electrochemical

processing

and

infiltration:

Application

to

tungsten/copper

composites. J Mater Sci 35(2): 477-86.

268) Jha M, Charalambides PG (1998). Crack-tip micromechanical fields in layered elastic
composites: crack parallel to the interfaces. Int J Sol Struct 35(1-2): 149-79.
269) Jiang F, Deng ZL, Zhao K, Sun J (2003). Fatigue crack propagation normal to a
plasticity mismatched bimaterial interface. Mater Sci Eng A 356: 258-66.
270) Jiang W, Nair R, Molian P (2005). Functionally graded mold inserts by laser-based
flexible fabrication: processing modeling, structural analysis, and performance
evaluation. J Mater Proc Technol (in publication).
271) Jin Z-H, Noda N (1994). Crack-Tip Singular Fields in Nonhomogeneous Materials. J
App Mech 61: 738-40.
272) Jin Z-H, Batra RC (1996a). Some Basic Fracture Mechanics Concepts in Functionally
Graded Materials. J Mech Phys Sol 44(8): 1221-34.
273) Jin Z, Batra RC (1996b). Stress intensity relaxation at the tip of an edge crack in
functionally graded materials subjected to a thermal shock. J Therm Str 19(4): 317-39.
274) Jin Z-H, Batra RC (1998). R-curve and strength behaviour of a functionally graded
material. Mater Sci Eng A 242: 70-6.
275) Jin Z-H, Paulino GH, Dodds RH (2002). Finite Element Investigation of Quasi-Static
Crack Growth in Functionally Graded Materials Using a Novel Cohesive Zone
Fracture Model. J Appl Mech 69: 370-9.
276) Jin Z-H, Dodds Jr. RH (2004). Crack growth resistance behavior of a functionally
graded material: computational studies Engrg Fract Mech 71: 1651-72.
277) Jivkov AP (1999). On Crack Growth in Functionally Graded Materials. PhD Thesis,
Solid Mech Div, Dept of Mech Engrg, Lule Univ Technol, Lule, Sweden.
278) Jivkov AP, Sthle P (2003). A model for graded materials with application to cracks.
Int J Fracture 124: 93105.
R-20

279) Joyce MR, Reed PAS, Syngellakis S (2003). Numerical modelling of crack shielding
and deflection in a multilayered material system. Mater Sci Eng A 342: 11-22.
280) Kachanov LM (1958). On the time to failure under creep conditions. Izv. AN SSSR,
Otd Tekhn Nauk 8, 26-31.
281) Kah Soh A, Bian LC (2001). Mixed mode fatigue crack growth criteria. Int J Fatigue
23(5): 427-39.
282) Kamiski M (2002). On probabilistic fatigue models for composite materials. Int J
Fatigue 24: 477-95.
283) Kamiya S, T Yamauchi, H Abe. Three-dimensional simulation of crack extension in
brittle polycrystalline materials. Engrg Fract Mech 66 (2000) 573-586
284) Kang K-J, Beom HG (2000). Plastic zone size near the crack tip in a constrained
ductile layer under mixed mode loading. Engrg Fract Mech 66: 257-68.
285) Karihaloo BL (1979). Discussion of: Howard IC (1978), Simple approximate results
for the stress intensity factors at the tip of a kinked crack. Int J Fracture 15: R131-2.
286) Karihaloo BL, Keer LM, Nemat-Nasser S (1980). Crack Kinking under
Nonsymmetric Loading. Engrg Fract Mech 13: 879-88.
287) Kassir MK (1972). Note on the twisting deformation of a nonhomogeneous shaft
containing a circular crack. Int J Fract Mech 8: 325-34.
288) Kawasaki A, Watanabe B (1999). Cyclic Thermal Fracture Behaviour and Spallation
Life of PSZ/NiCrAlY Functionally Graded Thermal Barrier Coatings. Mater Sci
Forum 308-11: 402-9.
289) Kfouri AP (1986). Some evaluations of the elastic T-term using Eshelby's method. Int
J Fracture 30: 301-15.
290) Khan SMA, Khraisheh MK (2000). Analysis of mixed-mode experiments crack
initiation angles under various loading conditions. Engrg Fract Mech 67(5): 397-419.
291) Khan SMA, Khraisheh MK (2004). A new criterion for mixed mode fracture
initiation based on the crack-tip plastic core region. Int J Plasticity 20: 55-84.
292) Kidson L (1998). Development and fracture behaviour of a ceramic/polymer
functionally graded interface. Honours Thesis, School of Materials Science &
Engineering, University of New South Wales.

R-21

293) Kieback B, Neubrand A, Riedel H (2003). Processing techniques for functionally

graded materials. Mater Sci Eng A 362: 81-105.
294) Kim AS, Suresh S, Shih CF (1997). Plasticity effects on fracture normal to interfaces
with homogeneous and graded compositions. Int J Sol Struct 34: 3415-32.
295) Kim AS, Besson J, Pineau A (1999). Global and local approaches to fracture normal
to interfaces. Int J Sol Struct 36: 1845-64.
296) Kim SJ (2000). An analysis of fatigue crack growth rate considering spatial variation
of fatigue crack growth resistance. Key Engrg Mater 183-187: 19-24.
297) Kim J-H, Paulino GH (2002a). Finite element evaluation of mixed mode stress
intensity factors in functionally graded materials. Int J Numer Meth Eng 53: 1903-35.
298) Kim J-H, Paulino GH (2002b). Isoparametric graded finite elements for
nonhomogeneous isotropic and orthotropic materials. J Appl Mech 69(4):502-14.
299) Kim J-H, Paulino GH (2003a). Mixed-mode J-integral Formulation and
Implementation

Using

Graded

Finite

Elements

for

Fracture

Analysis

of

Nonhomogeneous Orthotropic Materials. Mech Mater 35(1-2): 107-28.

300) Kim J-H, Paulino GH (2003b). Mixed-mode fracture of orthotropic functionally
graded materials using finite elements and the modified crack closure method. Engrg
Fract Mech 69(14-16): 1557-86.
301) Kim J-H, Paulino GH (2003c). An Accurate Scheme for Mixed-mode Fracture
Analysis of Functionally Graded Materials Using the Interaction Integral and
Micromechanics Models. Int J Numer Meth Eng 58(10): 1457-97.
302) Kim J-H, Paulino GH (2003d). T-stress, mixed-mode stress intensity factors, and
crack initiation angles in functionally graded materials: a unified approach using the
interaction integral method. Comput Methods Appl Mech Eng 192:1463-94.
303) Kim J-H, Paulino GH (2003e). The interaction integral for fracture of orthotropic
functionally graded materials: evaluation of stress intensity factors. Int J Sol Struct 40:
3967-4001.
304) Kim J-H, Paulino GH (2004). Simulation of crack propagation in functionally graded
materials under mixed-mode and non-proportional loading. Int J Mech Mat Design 1:
63-94.

R-22

305) Kishimoto H, Ueno A, Okawara S, Kawamoto H (1994). Crack propagation behavior

of polycrystalline alumina under static and cyclic load. J Am Ceram Soc 77(5): 132428.
306) Kitagawa H, Yuuki R, Ohira T (1975). Crack Morphological Aspects in Fracture
Mechanics. Engrg Fract Mech 7: 515-29.
307) Knechtel M, Prielipp H, Meullejans H, Claussen N, Rdel J (1994). Mechanical
properties of Al/Al2O3 and Cu/Al2O3 composites with interpenetrating networks. Scr
met mater 31:1085-90.
308) Knehans R, Steinbrech R (1982). Memory effect of crack resistance during slow
crack growth in notched Al2O3 bend specimens. J Mat Sci Lett 1: 327-9.
309) Kohnle C, Mintchev O, Schmauder S (2002). Elastic and plastic fracture energies of
metal/ceramic joints. Comput Mater Sci 25: 272-77.
310) Kolednik O, Predan J, Shan GX, Simha NK, Fischer FD (2005). On the fracture
behavior of inhomogeneous materialsA case study for elastically inhomogeneous
bimaterials. Int J Sol Struct 42: 605-20.
311) Konda N, Erdogan F (1994). The Mixed Mode Crack Problem in a Nonhomogeneous Elastic Medium. Engrg Fract Mech 47(4): 533-45.
312) Knigshofer R, Eder A, Lengauer W, Dreyer K, Kassel D, Daubb H-W, van den Berg
H (2004). Growth of the graded zone and its impact on cutting performance in highpressure nitrogen modified functionally gradient hardmetals. J Alloys Comp 366: 22832.
313) Kotoul M, Urbis R (2001). On the stability of cracks with bridged kinks. Engrg Fract
Mech 68: 89-105.
314) Kouzeli M, Dunand DC (2003). Effect of Reinforcement Connectivity on the ElastoPlastic Behavior of Aluminum Composites containing sub-micron Alumina Particles.
Acta mat 51: 6105-21.
315) Krajcinovic D (2000). Damage mechanics: accomplishments, trends and needs. Int J
Sol Struct 37: 267-77.
316) Kreher W, Pompe W (1989). Internal Stresses in Heterogeneous Solids. 1st ed.,
Akademie Verlag, Berlin.

R-23

317) Krstic VD (1983). On the fracture of brittle-matrix/ductile-particle composites. Phil

Mag A 48(5): 695-708.
318) Krumova M, Klingshirn C, Haupert F, Friedrich K (2001). Microhardness studies on
functionally graded polymer composites. Compos Sci Tech 61(4): 557-63.
319) Kruzic JJ, McNaney JM, Cannon RM, Ritchie RO (2004). Effects of plastic
constraint on the cyclic and static fatigue behavior of metal/ceramic layered
structures. Mech Mater 36: 57-72.
320) Lacarac V, Smith DJ, Pavier MJ, Priest M (2000). Fatigue crack growth from plain
and cold expanded holes in aluminum alloys. Int J Fatigue 22: 189-203.
321) Lakshminarayanan R, Shetty DK, Cutler RA (1996). Toughening of Layered
Ceramic Composites with Residual Surface Compression. J Am Ceram Soc 79(1): 7987.
322) Lambros J, Abanto-Bueno J (2002). Investigation of crack growth in functionally
graded materials using digital image correlation. Engrg Fract Mech 69: 1695-1711.
323) Lamon J (2001). A micromechanics-based approach to the mechanical behavior of
brittle-matrix composites. Compos Sci Tech 61 2259-72.
324) Lange FF, Velamakanni BV, Evans AG (1990). Method for processing metalreinforced ceramic composites. J Am Ceram Soc 73(2): 388-93.
325) Larsson SG, Carlsson AJ (1973). Influence of non-singular stress terms and specimen
geometry on small scale yielding at crack tips in elastic plastic materials. J Mech Phys
Sol 21: 263-77.
326) Lawn BR (1993). Fracture of Brittle Solids. 2nd ed., Cambridge University Press,
Cambridge, UK.
327) Lee Y-D, Erdogan F (1995). Residual/thermal stresses in FGM and laminated thermal
barrier coatings. Int J Fracture 69: 145-65.
328) Lee CS, De Jonghe LC, Thomas G (2001). Mechanical properties of polyptoidally
joined Si3N4-Al2O3. Acta mater 49: 3767-73.
329) Lee KH (2004). Characteristics of a crack propagating along the gradient in
functionally gradient materials. Int J Solids Structures 41: 2879-98.
330) Leevers PS, Radon JC, Culver LE (1976). Fracture trajectories in a biaxially stressed
plate. J Mech Phys Sol 24: 381-390.
R-24

331) Lele P, Dong M, Schmauder S (1999). Self-consistent matricity model to simulate

the mechanical behaviour of interpenetrating microstructures. Comput Mat Sci 15:
455-65.
332) Leck KM (2002). Finite Element Analysis of Residual Stress Effect on Crack
Propagation in Functionally Graded Materials. Masters Thesis, School of Biomedical
Engineering, University of New South Wales.
333) Lee W, Howard SJ, Clegg WJ (1996). Growth of interface defects and its effect on
crack deflection and toughening criteria. Acta mater 44(10): 3905-22.
334) Lee W (2002). Crack propagation within corrugated interfaces in ceramic laminates.
Scripta mater 47: 295-300.
335) Leguillon D, Tariolle S, Martin E, Chartier T, Besson JL (2005). Prediction of crack
deflection in porous/dense ceramic laminates. J Europ Ceram Soc (to be published).
336) Lekhnitskii SG. Anisotropic Plates, translated from 2nd Russian edition, Gordon and
Breach, New York, 1968, 78-84.
337) Lengauer W, Dreyer K (2002). Functionally graded hardmetals. J Alloys Comp 338:
194-212.
338) Le Pen E, Baptiste D, Hug G (2002). Multi-scale fatigue behaviour modelling of Al
Al2O3 short fibre composites Int J Fatigue 24: 205-14.
339) Li Z, Schmauder S, Dong M (1999). A simple mechanical model to predict fracture
and yield strengths of particulate two-phase materials. Comput Mat Sci 15: 11-21.
340) Li H, Lambros J, Cheeseman BA, Santare MH (2000). Experimental investigation of
the quasi-static fracture of functionally graded materials. Int J Sol Struct 37: 3715-32.
341) Li C, Zou Z, Duan Z (2000). Multiple Isoparametric Finite element method for
nonhomogeneous media. Mech Res Comm 27(2): 137-142.
342) Li J, Ngan AHW, Gumbsch P (2003). Atomistic modeling of mechanical behavior.
Acta mat 51: 5711-42.
343) Li J, Zhang X-B, Recho N (2004). JMp based criteria for bifurcation assessment of a
crack in elasticplastic materials under mixed mode III loading. Engrg Fract Mech
71: 329-343.
344) Lim IL, Johnston IW, Choi SK (1996). A finite element code for fracture propagation
analysis within elasto-plastic continuum. Engrg Fract Mech 53(2): 193-211.
R-25

345) Lim YM, Park YJ, Yun HY, Hwang KS (2002). Functionally graded Ti/HAP coatings
on Ti6Al4V obtained by chemical solution deposition. Ceram Int 28: 37-41.
346) Lin T, Evans AG, Ritchie RO (1986). A statistical model of brittle fracture by
transgranular cleavage. J Mech Phys Sol 34: 47797.
347) Lin JS, Miyamoto Y (1999). Internal Stress and Fracture Behaviour of Symmetric
Al2O3/TiC/Ni FGMs. Mater Sci Forum 308-311: 855-860.
348) Lin XB, Smith RA (1999). Finite element modelling of fatigue crack growth of
surface cracked plates Part I: The numerical technique. Engrg Fract Mech 63: 503-22.
349) Lindhagen JE, Jekabsons N, Berglund LA (2000). Application of bridging-law
concepts to short-fibre composites 4. FEM analysis of notched tensile specimens.
Compos Sci Tech 60: 2895-2901.
350) Liu B, Kimoto H, Kitagawa H (1996). Elastic-plastic analysis of a crack parallel to
the interface. Engrg Fract Mech 53(4): 607-23.
351) Liu GR, Han X, Hu YG, Lam KY (2001). Material Characterisation of functionally
graded material by means of elastic waves and a progressive-learning neural network.
Compos Sci Tech 61(10): 1401-11.
352) Liu H, Maekawa T, Patrikalakis NM, Sachs EM, Cho W (2004). Methods for featurebased design of heterogeneous solids. Computer-Aided Design 36(12): 1141-59.
353) Lo KK (1978). Analysis of Branched Cracks. J Appl Mech 45: 797-801.
354) Lorentzon M, Eriksson K. A path independent integral for the crack extension force
of the circular arc crack. Engrg Fract Mech 2000;66:423-39.
355) Love AEH (1927). A Treatise on the Mathematical Theory of Elasticity. 4th ed,
Dover, New York.
356) Low IM (1998). Physical characteristics of an in-situ layered and graded
alumina/aluminium titanate composite. Mater Res Bull 33: 1475-81.
357) Lugovy M, Orlovskaya N, Berroth K, Kbler J (1999). Macrostructural engineering
of ceramic-matrix layered composites. Compos Sci Tech 59: 1429-37.
358) Lugovy M, Orlovskaya N, Slyunyayev V, Gogotsi G, Kbler J, Sanchez-Herencia AJ
(2002). Crack bifurcation features in laminar specimens with fixed total thickness.
Compos Sci Tech 62: 819-30.

R-26

359) Lugovy M, Slyunyayev V, Orlovskaya N, Blugan G, Kbler J, Lewis M (2005).

Apparent fracture toughness of Si3N4-based laminates with residual compressive or
tensile stresses in surface layers. Acta mater 53: 289-96.
360) Luo J, Bowen P (2003). A probabilistic methodology for fatigue life prediction. Acta
mat 51: 3537-50.
361) Ma F, Sutton MA, Deng X (2001). Plane strain mixed mode crack-tip stress fields
characterized by a triaxial stress parameter and a plastic deformation extent based
characteristic length. J Mech Phys Sol 49: 2921-53.
362) Ma L, Wu L-Z, Zhou Z-G, Zeng T (2004). Crack propagating in a functionally graded
strip under the plane loading. Int J Fracture 126: 39-55.
363) Ma J, Wang H, Weng L, Tan GEB (2004b). Effect of porous interlayers on crack
deflection in ceramic laminates. J Euro Ceram Soc 24: 825-31.
364) Magill MA, Zwerneman FJ (1997). An analysis of sustained mixed mode fatigue
crack growth. Engrg Fract Mech 56(1): 9-24.
365) Mai YW, Lawn BR (1987). Crack-Interface Grain Bridging as a Fracture Resistance
Mechanism in Ceramics: II, Theoretical Fracture Mechanics Model. J Am Ceram Soc
70(4): 289-94.
366) Majumdar BS, Newaz GM (1995). Constituent damage mechanisms in metal matrix
composites under fatigue loading and their effects on fatigue life. Mater Sci Eng A
200: 114-129.
367) Markworth AJ, Ramesh KS, Parks WP Jr (1995). Review: Modelling Studies applied
to functionally graded materials. J Mat Sci 30: 2183-93.
368) Marshall DB, Ratto JJ, Lange FF (1991). Enhanced Fracture Toughness in Layered
Microcomposites of Ce-ZrO2 and Al2O3. J Am Ceram Soc 74(12): 2979-87.
369) Marur PR, Tippur HV (2000). Numerical Analysis of Crack Tip Fields in
Functionally Graded Materials with a Crack Normal to the Elastic Gradient. Int J Sol
Struct 37: 5353-70.
370) Mataga PA (1989). Deformations of Crack-Bridging Ductile Reinforcements in
Toughened Brittle Materials. Acta met 37(12): 3349-59.
371) Matterson J, Reimanis IE, Berger J (2004). Fracture in Nb/Al2O3 particulate
composites. Ceram Trans (in publication).
R-27

372) Mattern A, Huchler B, Staudenecker D, Oberacker R, Nagel A, Hoffmann MJ

(2004). Preparation of Interpenetrating Ceramic-Metal-Composites. J Euro Ceram Soc
24(12) 3399-3408.
373) Maxwell JC (1873). Treatise on Electricity and Magnetism. 1st ed., Clarendon Press,
Oxford, p. 365.
374) McNaney JM, Cannon RM, Ritchie RO, 1994. Near-interfacial crack trajectories in
metal-ceramic layered structures. Int J Fracture 66(3): 227-40.
375) McNaney JM, Cannon RM, Ritchie RO (1996). Fracture and fatigue crack growth
along aluminium-alumina interfaces. Acta mater 44(12): 4113-28.
376) Meguid SA, Wang XD, Jiang LY (2002). On the dynamic propagation of a finite
crack in functionally graded materials. Engrg Fract Mech 69: 1753-1768.
377) Melin S (2002). The influence of the T-stress on the directional stability of cracks. Int
J Fracture 114: 259-65.
378) Mintchev Or, Rohde J, Schmauder S (1998). Mesomechanical simulation of crack
propagation through graded ductile zones in hardmetals. Comput Mat Sci 13: 81-89.
379) Miranda ACO, Meggiolaro MA, Castro JTP, Martha LF, Bittencourt TN (2003).
Fatigue life and crack path predictions in generic 2D structural components. Engrg
Fract Mech 70: 1259-1279.
380) Mishnaevsky Jr. L, Dong M, Hnle S, Schmauder S (1999). Computational
mesomechanics of particle-reinforced composites. Comput Mater Sci 16: 133-143.
381) Mishnaevsky Jr. L, Lippmann N, Schmauder S (2003). Computational modeling of
crack propagation in real microstructures of steels and virtual testing of artificially
designed materials. Int J Fracture 120: 581600.
382) Mishnaevsky Jr. L (2004). 3D finite element study of damage evolution in a particle
composite. Proc. Euro Conf Fract, Stockholm.
383) Mos N, Dolbow J, Belytschko T (1999). A Finite Element Method for Crack Growth
without Remeshing. Int J Numer Meth Engrg 46: 131-150.
384) Moon HJ, Earmme YY. Calculation of elastic T -stresses near interface crack tip
under in-plane and anti-plane loading. Int J Fracture 91: 179-195, 1998.

R-28

385) Moon RJ, Bowman KJ, Trumble KP, Rdel J (2001). Fracture Resistance Curve
Behaviour

of

Multilayered

Alumina-Zirconia

Composites

Produced

by

Centrifugation. Acta Mat 49: 995-1003.

386) Moon RJ, Hoffman M, Hilden J, Bowman KJ, Trumble KP, Rdel J (2002a). R-Curve
Behaviour in Alumina-Zirconia Composites with Repeating Graded Layers. Engrg
Fract Mech 69: 1647-1665.
387) Moon RJ, Hoffman M, Hilden J, Bowman KJ, Trumble KP, Rdel J (2002b). A
Weight Function Analysis on the R-Curve Behaviour of Multilayered AluminaZirconia Composites. J Am Ceram Soc 85(6): 1505-11.
388) Moon RJ, Tilbrook MT, Hoffman M, Neubrand A (2004). Al-Al2O3 Composites with
Interpenetrating Network Structures: Composite Modulus Estimation. J Am Ceram
Soc (in publication).
389) Moon RJ (2004). Fracture of graded and non-graded alumina-aluminium and
alumina-copper composites (unpublished work).
390) Moro A, Kuroda A, Kusaka K (2002). Development status of the reusable highperformance engines with functionally graded materials. Acta Astronautica 50(7) 42732.
391) Mortensen A, Suresh S (1995). Functionally graded metals and metal-ceramic
composites: Part 1 Processing, Int Mat Rev 40(6): 239-65.
392) Movchan BA, Yakovchuk KY (2004). Graded thermal barrier coatings, deposited by
EB-PVD. Surf Coat Technol 188189: 85-92.
393) MSEA (2003): Special issue of Materials Science and Engineering A 362.
394) Mujul S (2000). Crack propagation in bimaterial multilayered periodically
microcracking composite media. Compos Sci Tech 60(12-13): 2213-21.
395) Mller E, Draar C, Schilz J, Kaysser WA (2003). Functionally graded materials for
sensor and energy applications. Mater Sci Eng A 362: 17-39.
396) Munro RJ (2001). Effective medium theory of the porosity dependence of bulk
moduli. J Am Ceram Soc 84(5): 1190-92.
397) Muskhelishvili NI (1953). Some Basic Problems of Mathematical Theory of
Elasticity. 1st ed., Noordhoff, Groningen, Netherlands.

R-29

398) Nadeau JC, Ferrari M (1999). Microstructural optimization of a functionally graded

transversely isotropic layer. Mech Mater 31: 63751.
399) Nagata F (1999). Intelligent Modelling Mechanisms and Design Concepts of FGMs
in Natural Composites. Mat Sci Forum 308-311: 331-37.
400) Nairn JA (2001). Fracture mechanics of composites with residual stresses, imperfect
interfaces, and traction-loaded cracks. Compos Sci Tech 61: 2159-2167.
401) Nakagaki M, Wu Y (2001). LRM constitutive model for describing functionally
graded material with fast propagating crack. In: Ravi-Chandar K, Karihaloo BL, Kishi
T, Ritchie RO, Yokobori AT, Yokobori T (eds). Advances in Fracture Research, Proc.
ICF10, Pergamon CDROM.
402) Nakamura T, Parks DM (1992). Determination of elastic T-stress along threedimensional crack fronts using an interaction integral. Int J Sol Struct 29: 1597-611.
403) Nakamura T, Wang T, Sampath S (2000). Determination of Properties of Graded
Materials by Inverse Analysis and Instrumented Indentation. Acta mat 48: 4293-4306.
404) Nakamura T, Wang T (2001). Simulations of crack propagation in porous materials.
J Appl Mech 68: 242-251.
405) Nee AYC, Fuh JYH, Miyazawa T (2001). On the improvement of the
stereolithography process. J Mater Proc Tech 113: 262-268.
406) Needleman A (1987). A continuum model for void nucleation by inclusion
debonding. J Appl Mech 54: 52531.
407) Nemat-Alla M (2003). Reduction of thermal stresses by developing two-dimensional
functionally graded materials Int J Sol Struct 40: 7339-56.
408) Neubrand A, Rdel J (1997). Gradient Materials: An Overview of a Novel Concept.
Zeit Metall 88: 358-71.
409) Neubrand A, Chung TJ, Rdel J, Steffler ED, Fett T (2002). Residual stresses in
functionally graded plates, J Mater Res 17: 2912-2920.
410) Neubrand

(2002).

Private

communication.

Fraunhfer

Institut

fr

WerkstoffeMechanik, Freiburg, Germany.

411) Newman Jr. JC (1998). The merging of fatigue and fracture mechanics concepts: a
historical perspective. Progress in Aerospace Sciences 34: 347390 (From: ASTM
STP 1321Fatigue and Fracture Mechanics, 28th Volume, copyright American
R-30

Society for Testing and Materials, 100 Barr Harbor Drive, West Conshohocken, PA
19428-2959, USA.)
412) Niino M, Hirai T, Watanabe R (1987). The functionally gradient materials. J Jap Soc
Compos Mat 13: 257-64.
413) Noda N-A, Oda K (1993). Effect of curvature at the crack-tip on the stress intensity
factor for curved cracks. Int J Fracture 64: 239-249.
414) Noda N, Oda K, Ishi K (1994). Analysis of Stress Intensity Factors for Curved
Cracks. JSME Int J A 37(4): 360-365.
415) Nuismer RJ (1975). An energy release rate criterion for mixed mode fracture. Int J
Fracture 11: 245-50.
416) ODowd NP, Shih CF (1991; 1992). Family of Crack-Tip Fields Characterized by a
Triaxiality Parameter: Part I-Structure of Fields. J Mech Phys Sol 39: 989-1015; Part
II-Fracture Applications. J Mech Phys Sol 40: 939-63.
417) Oh TS, Rdel J, Cannon RM, Ritchie RO (1988). Ceramic/metal interfacial crack
growth: Toughening by controlled microcracks and interfacial geometries. Acta met
36(8): 2083-93.
418) Ohtsuka K (2002). Comparison of criteria on the direction of crack extension. J
Comput Appl Math 149: 335-39.
419) Ortiz M, Pandolfi A (1999). Finite-deformation irreversible cohesive elements for
three-dimensional crack-propagation analysis. Int J Numer Meth Eng 44: 1267-82.
420) Ostoja-Starzewski M (1998). Random field models of heterogeneous materials. Int J
Sol Struct 35(19): 2429-55.
421) Ozturk M, Erdogan F (1996). Axisymmetric Crack Problem in Bonded Materials
with a Graded Interfacial Region. Int J Sol Struct 33(2): 193-219.
422) Palaniswamy K, Knauss WG (1978). On the problem of crack extension in brittle
solids under general loading. S Nemat-Nasser ed., Mech Today 4, Pergamon Press.
423) Parameswaran V, Shukla A (1999). Crack-tip stress fields for dynamic fracture in
functionally graded materials. Mech Mater 31:579-96.
424) Parameswaran V, Shukla A (2000). Processing and characterisation of a model
functionally graded material. J Mater Sci 35: 21-9.

R-31

425) Parsons WB (1939). Engineers and engineering in the renaissance. Williams and
Wilkers, Baltimore, MD.
426) Patankar R, Ray A, Lakhtakia A (1998). A state-space model of fatigue crack growth
Int J Fracture 90: 235-49.
427) Paul B (1960). Prediction of Elastic Constants of Multiphase Materials. Trans Met
Soc AIME 218: 36-41.
428) Paulino GH, Fannjiang AC, Chan Y-S (1999). Gradient Elasticity Theory for a Mode
III Crack in a Functionally Graded Material. Mater Sci Forum 308-311: 971-6.
429) Paulino GH, Jin Z-H (2001). Fracture Mechanics of Viscoelastic Functionally Graded
Materials. Ceram Trans 114: 715-22.
430) Paulino GH (2002). Fracture in Functionally Graded Materials. Engrg Fract Mech 69:
1519-20.
431) Paulino GH, Kim J-H (2004). A New Approach to Compute T-stress in Functionally
Graded Materials Using the Interaction Integral Method. Engrg Fract Mech 71(13-14):
1907-50.
432) Pawliska P, Richard RH, Diekmann P (1993). The behavior of cracks in elastic
plastic materials under plane normal and shear loading. Int J Fracture 62: 43-54.
433) Pedersen TO (1998). Remeshing in analysis of large plastic deformations. Comput
Struct 67: 279-88.
434) Peixiang H, Ziran L, Changchun W (2001). An element-free Galerkin method for
dynamic fracture in functional graded material. In: Ravi-Chandar K, Karihaloo BL,
Kishi T, Ritchie RO, Yokobori AT, Yokobori T (eds). Advances in Fracture Research,
Proc. ICF10; Pergamon CDROM.
435) Peng HX, Fan Z, Evans JRG (2001). Bi-Continuous Metal Matrix Composites. Mater
Sci Eng A 303: 37-45.
436) Peng XQ, Liu G, Wu L, Liu GR, Lam KY (1998). A stochastic finite element method
for fatigue reliability analysis of gear teeth subjected to bending Comput Mech 21:
253-61.
437) Petroski HJ, Achenbach JD (1978). Computation of the weight function from a stress
intensity factor. Engrg Fract Mech 10; 257-66.

R-32

438) Pettit RG, Chen C-S, Ingraffea AR, Hui CY (2001). Process zone size effects on
naturally curving cracks. Engrg Fract Mech 68: 1181-205.
439) Pezzotti G (1999). In Situ Study of Fracture Mechanisms in Advanced Ceramics
using Fluorescence and Raman Microprobe Spectroscopy. J Raman Spectr 30: 86775.
440) Pezzotti G, Sbaizero O, Sergo V, Muraki N, Maruyama K, Nishida T (1998). In-situ
measurements of frictional bridging stresses in alumina using fluorescence
spectroscopy. J Am Ceram Soc 81: 187-92.
441) Pezzotti G, Ichimaru H, Ferroni LP, Hirao K, Sbaizero O (2001). Raman Microprobe
Evaluation of Bridging Stresses in Highly Anisotropic Silicon Nitride. J Am Ceram
Soc 84(8): 1785-90.
442) Pezzotti G, Okuda H, Muraki N, Nishida T (1999). In situ determination of bridging
stresses in Al2O3/Al2O3-platelet composites by fluorescence spectroscopy. J Euro
Ceram Soc 19: 601-8.
443) Pfeiffer S (2003). Fatigue of nanocrystalline copper produced by equal-channel
angular extrusion. Honours Thesis, School of Materials Science & Engineering,
University of New South Wales.
444) Pindera M-J, Aboudi J, Arnold SM (2002). Analysis of spallation mechanism in
thermal barrier coatings with graded bond coats using the higher-order theory for
FGMs. Engrg Fract Mech 69: 1587-606.
445) Pintsuk G, Brnings SE, Dring J-E, Linke J, Smid I, Xu L (2003). Development of
W/Cu/functionally graded materials. Fusion Eng Design 66-68: 237-40.
446) Pippan R, Riemelmoser FO (1998). Fatigue of bimaterials. Investigation of the plastic
mismatch in case of cracks perpendicular to the interface. Comput Mater Sci 13:
1080-116.
447) Pippan R, Flechsig K, Riemelmoser FO (2000). Fatigue crack propagation behavior
in the vicinity of an interface between materials with different yield stresses. Mater
Sci Eng A 283: 225-33.
448) Pitakthapanaphong S, Busso EP (2002). Self-consistent elastoplastic stress solutions
for functionally graded material systems subjected to thermal transients. J Mech Phys
Sol 50: 695-716.
R-33

449) Plank R, Kuhn G (1999). Fatigue crack propagation under non-proportional mixed
mode loading. Engrg Fract Mech 62(2-3): 203-29.
450) Plevako VP (1972). On the theory of elasticity in inhomogeneous media. PMM 35(5):
853-60.
451) Pompe W, Worch H, Epple M, Friess W, Gelinsky M, Greil P, Hempel U,
Scharnweber D, Schulte K (2003). Functionally graded materials for biomedical
applications. Mater Sci Eng A 362: 4060.
452) Post D, Han B, Ifju P (1994). High Sensitivity Moir: Experimental analysis for
mechanics and materials. Springer, Berlin, Germany.
453) Poursatip A, Ashby MF, Beaumont PWR (1982). Damage Accumulation during
Fatigue of Composites. Scr met 16: 601-6.
454) Preuss M, Rauchs G, Doel TJA, Steuwer A, Bowen P, Withers PJ (2003).
Measurements of fibre bridging during fatigue crack growth in Ti/SiC fibre metal
matrix composites. Acta mat 51: 1045-57.
455) Prielipp H, Knechtel M, Claussen N, Streiffer SK, Mllejans H, Rhle M, Rdel J
(1995). Strength and fracture toughness of aluminium/alumina composites with
interpenetrating networks. Mater Sci Eng A 197: 19-30.
456) Put S, Vleugels J, Van der Biest O (2003). Microstructural engineering of
functionally graded materials by electrophoretic deposition. J Mater Proc Tech 143-4:
572-7.
457) Pyrz R, Bochonek B (1998). Topological disorder of microstructure and its relation to
the stress field. Int J Sol Struct 35(19): 2413-27.
458) Qian J, Fatemi A (1996). Mixed mode fatigue crack growth: a literature survey.
Engrg Fract Mech 55(6): 969-90.
459) Quinn GD, Morrell R (1991). Design Data for Engineering Ceramics: A Review of
the Flexure Test. J Am Ceram Soc 74(9): 2037-66.
460) Raddatz O, Schneider GA, Mackens W, Vo H, Claussen N (2000). Bridging
Stresses and R-Curves in Ceramic/Metal Composites. J Euro Ceram Soc 20(13):
2261-73.
461) Rahman S (2001). Probabilistic fracture mechanics: J-estimation and finite element
methods. Engrg Fract Mech 68: 107-125.
R-34

462) Rahman S, Kim JS (2001). Probabilistic fracture mechanics for nonlinear structures.
Int J Pres Ves Pip 78: 261-9.
463) Raju IS (1987). Calculation of strain-energy release rates with high order and singular
finite elements. Engrg Fract Mech 28(3): 251-74.
464) Raju IS, Shivakumar KN (1990). An equivalent domain integral method in the twodimensional analysis of mixed mode crack problems. Engrg Fract Mech 37(4): 70725.
465) Rao BN, Rahman S (2003). Mesh-free analysis of cracks in isotropic functionally
graded materials. Engrg Fract Mech 70: 1-27.
466) Raob KJ, Varmab KBR, Ramaswamya P, Seetharamua S (1997). Al2O3-ZrO2
composite coatings for thermal barrier applications. Compos Sci Tech 57(1): 81-9.
467) Rashid MM (1997). A computational procedure for simulation of crack advance in
arbitrary two-dimensional domains. Comput Mech 20: 133-8.
468) Rashid MM (1998). The arbitrary local mesh replacement method: an alternative to
remeshing for crack propagation analysis. Comput Meth Appl Mech Eng 154: 133-50.
469) Rashid MM, Tvergaard V (2003). On the path of a crack near a graded interface
under large scale yielding. Int J Sol Struct 40: 2819-31.
470) Rabach S, Lehnert W (1999). Modelling of the deformation behaviour of FGM by
fuzzy-logic. Comput Mater Sci 16: 167-75.
471) Rauchs G, Thomason PF, Withers PJ (2002). Finite element modelling of frictional
bridging during fatigue crack growth in fibre-reinforced metal matrix composites.
Comput Mat Sci 25(1-2): 166-73.
472) Rauchs G, Withers PJ (2002). Computational assessment of the influence of load
ratio on fatigue crack growth in fibre-reinforced metal matrix composites. Int J
Fatigue 24: 1205-11.
473) Ravichandran KS (1992). The Mechanics of Toughness Development in Ductile
Phase Reinforced Brittle Matrix Composites. Acta met mat 40(5): 1009-22.
474) Ravichandran KS (1994). Elastic Properties of Two-Phase Composites. J Am Ceram
Soc 77: 1178-1193.
475) Ravichandran KS (1995). Thermal residual stresses in a Functionally Graded
Material system. Mat Sci Eng A 201: 269-76.
R-35

476) Reimanis IE, Dalgleish BJ, Brahy M, Rhle M, Evans AG (1990). Effects of
plasticity on the crack propagation resistance of a metal/ceramic interface. Acta met
mat 38(12): 2645-53.
477) Reiter T, Dvorak GJ, Tvergaard V (1997). Micromechanical models for graded
composite materials. J Mech Phys Sol 45(8): 1281-1302.
478) Reynaud P (1996). Cyclic Fatigue of Ceramic-Matrix Composites at Ambient and
Elevated Temperatures. Compos Sci Tech 56: 809-14.
479) Rice JR, Sih GC (1965). Plane Problems of Cracks in Dissimilar Media. J Appl Mech
32: 418-23.
480) Rice JR (1968). Path-independent integral and the approximate analysis of strain
concentration by notches and cracks. J Appl Mech 35(2): 379-86.
481) Rice JR, Rosengren GF (1968). Plane strain deformation near a crack tip in a powerlaw hardening material. J Mech Phys Sol 16:1-12.
482) Rice JR (1974). Limitations to the small scale yielding approximation for crack tip
plasticity. J Mech Phys Sol 22: 17-26.
483) Rice JR (1988). Elastic Fracture Mechanics Concepts for Interfacial Cracks. J Appl
Mech 55: 98-103.
484) Richard HA (2001). Major aspects of Mixed-Mode problems.. In: Ravi-Chandar K,
Karihaloo BL, Kishi T, Ritchie RO, Yokobori AT, Yokobori T (eds). Advances in
Fracture Research, Proc. ICF10, Pergamon CDROM.
485) Ritchie RO, Rice JR, Knott JF (1973). On the relationship between critical stress and
fracture toughness in mild steel. J Mech Phys Sol 21: 395410.
486) Ritchie RO, Cannon RM, Dalgleish BJ, Dauskardt RH, McNaney JM (1993).
Mechanics and Mechanisms of Crack Growth at or Near Ceramic-Metal Interfaces:
Interface Engineering Strategies for Promoting Toughness. Mater Sci Eng A 166:
221-35.
487) Ritchie RO (1999). Mechanisms of fatigue crack propagation in ductile and brittle
solids. Int J Fracture 100: 55-83.
488) Ritchie RO, Gilbert CJ, McNaney JM (2000). Mechanics and mechanisms of fatigue
damage and crack growth in advanced materials. Int J Sol Struct 37: 311-29.

R-36

489) Rocha MM, Schueller GI (1996). A probabilistic criterion for evaluating the goodness
of fatigue crack growth models. Engrg Fract Mech 53(5): 707-31.
490) Rodopoulos CA, Yates JR, de los Rios ER (2001). Micro-mechanical modelling of
fatigue damage in titanium metal matrix composites. Theor Appl Fract Mech 35:5967.
491) Rdel J (1992). Crack Closure Forces in Ceramics: Characterisation & Formation. J
Eur Cer Soc 9: 323-34.
492) Rdel J (1992). Interaction between Crack Deflection and Crack Bridging. J Euro
Ceram Soc 10: 143-50.
493) Rdel J, Kelly JF, Lawn BR (1990). In Situ Measurements of Bridged Crack
Interfaces in the Scanning Electron Microscope. J Am Ceram Soc 73(11): 3313-8.
494) Rdel J, Prielipp H, Claussen N, Sternitzke M, Alexander KB, Becher PF, Schneibel
JH (1994). Ni3Al/Al2O3 composites with interpenetrating networks. Scr met mat 33:
853-6.
495) Rose LRF (1987). Crack reinforcement by distributed springs. J Mech Phys Sol 35(4):
383-405.
496) Rousseau CE, Tippur HV (2000). Compositionally Graded Materials with Cracks
Normal to the Elastic Gradient. Acta mat 48: 4021-33.
497) Rousseau CE, Tippur HV (2001a). Dynamic fracture of compositionally graded
materials with cracks along the elastic gradient: experiments and analysis. Mech
Mater 33: 403-21.
498) Rousseau CE, Tippur HV (2001b). Influence of elastic gradient profiles on
dynamically loaded functionally graded materials: cracks along the gradient. Int J Sol
Struct 38: 7839-56.
499) Rousseau C-E, Tippur HV (2002a). Evaluation of crack-tip fields and stress intensity
factors in functionally graded materials: Cracks parallel to elastic gradient. Int J
Fracture 114: 87-111.
500) Rousseau CE, Tippur HV (2002b). Influences of elastic variations on crack initiation
in functionally graded glass-filled epoxy. Engrg Fract Mech 69: 1679-93.
501) Roy YA, Dodds Jr RH (2001). Simulation of ductile crack growth in thin aluminum
panels using 3-D surface cohesive elements. Int J Fracture 110: 21-45.
R-37

502) Rozenburg K, Berger J, Tilbrook MT, Reimanis IE (2005). Experiments of cracks

near an interface with moir interferometry. Acta Materialia (in preparation).
503) Rubinstein AA (2003). Computational aspects of crack path development simulation
in materials with non-linear process zone. Int J Fracture 119: L15-20.
504) Rhle M, Claussen N, Heuer AH (1986). Transformation and microcrack toughening
as complementary processes in ZrO2-toughened Al2O3. J Am Ceram Soc 69: 195-9.
505) Russ JC (1998). Image Analysis Handbook. 3rd ed., CRC Press, Boca Raton, Florida.
506) Rutgers L (2004). Development and Fracture Behaviour of Graded Ceramic/Polymer
Joins. Project Report, School of Materials Science & Engineering, UNSW.
507) Ruys AJ, Popov EB, Sun D, Russell JJ, Murray CCJ (2001). Functionally graded
electrical/thermal ceramic systems. J Euro Ceram Soc 21: 2025-9.
508) Rybicki EF, Kanninen MF (1977). A finite element calculation of stress intensity
factors by a modified crack closure integral. Engrg Fract Mech 9: 931-8.
509) Santare MH, Lambros J (2000). Use of Graded Finite Elements to Model the
Behaviour of Nonhomogeneous Materials. J Appl Mech 67: 819-22.
510) Santare MH, Pastrama SD, de Castro PMST (2002). A practical fracture analysis for
functionally graded materials. J Mech Behav Mater 13(2): 17-32.
511) Sarkar BK, TGJ Glinn (1969). Impact fatigue of an alumina ceramic. J Mater Sci
4(11): 951-4.
512) Schapery RA (1968). Thermal Expansion Coefficients of Composite Materials Based
on Energy Principles. J Compos Mater 2(3): 380-404.
513) Schijve J (2003). Fatigue of structures and materials in the 20th century and the state
of the art. Int J Fatigue 25: 679-702.
514) Schmauder S, Weber U, Hofinger I, Neubrand A (1999). Modelling the Deformation
Behaviour of W/Cu Composites by a Self-Consistent Matricity Model. Techn Mech
19(4): 313-20.
515) Schollmann M, Fulland M, Richard HA (2003). Development of a new software for
adaptive crack growth simulations in 3D structures. Engng Fract Mech 70: 249-68.
516) Schulz U, Peters M, Bach Fr-W, Tegeder G (2003). Graded coatings for thermal,
wear and corrosion barriers. Mater Sci Eng A 362: 61-80.

R-38

517) Schwarzwalder K, Sommers AV (1963). Method of Making Porous Ceramic

Articles. US Patent 3.090.094.
518) Selvarathinam AS (1998). Fracture in off-axis unidirectionally reinforced ceramic
composites. Int J Fracture 90:209-34.
519) Sendeckyj GP (1974). Elastic Behaviour of Composites. Ch 3, GP Sendeckyj ed.,
Composite Materials, Vol 2: Mechanics of Composite Materials, Academic Press.
520) Sevostianov I, Sookay NK, von Klemperer CJ, Verijenko VE (2003). Environmental
degradation using functionally graded material approach. Compos Struct 62: 417-21.
521) Shabana YM, Noda N (2001). Thermo-elastic-plastic stresses in functionally graded
materials subjected to thermal loading taking residual stresses of the fabrication
process into consideration. Composites: B 32: 111-21.
522) Shah P, Jakus K, Ritter JE (2001). Impact Damage in Monolithic and Functionally
Graded Alumina. Ceram Trans; 114: 651-8.
523) Sham T-L (1991). The determination of the elastic T-term using higher order weight
functions. Int J Fracture 48: 81-102.
524) Sham TL, Li J, Hancock JW (1999). A family of plane strain crack tip stress fields for
interface cracks in strength-mismatched elastic/perfectly plastic solids. J Mech Phys
Sol 47: 1963-2010.
525) Shaw MC, Marshall DB, Dadkhah MS, Evans AG (1993). Cracking and damage
mechanisms in ceramic/metal multilayers. Acta metall mater 41(11): 3311-22.
526) Shaw MC (1998). The effects of strength probabilistics on the fracture mode of
ceramic/metal multilayers. Engrg Fract Mech 61: 49-74.
527) Sheikh AK, Younas M (1995). A reliability model for fatigue life characterization. Int
J Fatigue 17(2): 121-8.
528) Shih CF (1974). Small-scale yielding analysis of mixed mode plane-strain crack
problems. Fracture Analysis ASTM STP 560, ASTM; 187-210.
529) Shih CF (1981). Relationships between the J-integral and the crack opening
displacement for stationary and extending cracks. J Mech Phys Sol 4: 305-26.
530) Shih CF, Asaro RJ (1988). Elastic-plastic analysis of cracks on bimaterial interfaces:
Part I - small scale yielding. J Appl Mech 55: 299-316.

R-39

531) Shimojima K, Yamada Y, Mabuchi M, Saito N, Nakanishi M, Shigematsu I,

Nakamura M, Asahina T, Igarashi T (1999). Optimisation Method of FGM
Compositional Distribution Profile by Genetic Algorithm. Mater Sci Forum 308-11:
1006-11.
532) Shin K-H, Natu H, Dutta D, Mazumder J (2003). A method for the design and
fabrication of heterogeneous objects. Mater Design 24: 339-53.
533) Siber B, Rettenmayr M, Mller C, Exner HE (1999). Concentration Gradients in
Aluminium Alloys Generated by Directional Solidification and their Effects on
Fatigue Crack Propagation. Mater Sci Forum 308-311: 211-6.
534) Sigl LS, Mataga PA, Dalgleish BJ, McMeeking RM, Evans AG (1988). On the
toughness of brittle materials reinforced with a ductile phase. Acta met 36(4): 945-53.
535) Sih GC, Paris PC, Erdogan F (1962). Crack-Tip Stress Intensity Factors for Plane
Extension and Plate Bending Problems. J Appl Mech 29: 306-12.
536) Sih GC (1974). Strain energy density factor applied to mixed mode crack problems.
Int J Fracture 10: 305-21.
537) Sih GC, Chen EP (1981). Cracks in Composite Materials. Sih GC ed., Mechanics of
Fracture, Vol 6.
538) Simha NK, Fischer FD, Kolednik O, Chen CR (2003). Inhomogeneity effects on the
crack driving force in elastic and elastic-plastic materials. J Mech Phys Sol 51: 20940.
539) Sladek J, Sladek V (1997). Evaluation of T-stresses and stress intensity factors in
stationary thermoelasticity by the conservation integral method. Int J Fracture 86:
199-219.
540) Smith DJ, Ayatollahi MR, Pavier MJ (2001). The role of T-stress in brittle fracture
for linear elastic materials under mixed-mode loading. Fatig Fract Eng Mat Struct 24:
137-50.
541) Sohn KS, Lee S, Baik S (1995). Analytical Modelling for Bridging Stress Function
Involving Grain Size Distribution in a Polycrystalline Alumina. J Am Ceram Soc
78(5): 1401-5.
542) Sokolnikoff IS (1956). Mathematical Theory of Elasticity, 2nd ed., McGraw-Hill, New
York.
R-40

543) Soma T, Masuda M, Matsui M, Oda I (1988). Cyclic Fatigue Testing of Ceramic
Materials. Int J Hi Tech Ceram 4: 289-99.
544) Srensen BF, Gamstedt EK, Jacobsen TK (2000). Equivalence of J Integral and
Stress Intensity Factor Approaches for Large Scale Bridging Problems. Int J Fracture
104: L31-6.
545) Souchet R (2003). Concerning the fictitious continuum in damage mechanics. Int J
Engrg Sci 41: 1975-88.
546) Spearing SM, Evans AG (1992). The Role of Fibre Bridging in the Delamination
Resistance of Fibre-Reinforced Composites. Acta met mat 40(9): 2191-9.
547) Stech M, Rdel J (1996). Method for Measuring Short-Crack R-Curves without
Calibration Parameters: Case Studies on Alumina and Alumina/Aluminium
Composites. J Am Ceram Soc 79(2): 291-7.
548) Steffler ED (2001). Applications of Phase Shifted Moir Interferometry. Ceram Trans
114: 595-602.
549) Steinbrech RW, Reichl A, Schaarwachter W (1990). R-Curve Behaviour of Long
Cracks in Alumina. J Am Ceram Soc 73(7): 2009-15.
550) Stellbrink KKU (1996). Micromechanics of Composites: Composite Properties of
Fibre and Matrix Constituents, Hanser, Dortmund.
551) Sternitzke M, Knetchel M, Hoffman M, Broszeit E, Rdel J (1996). Wear and
friction of aluminium oxide/aluminium composites with interpenetrating networks. J
Am Ceram Soc 79(1): 121-8.
552) Stinchcomb WW, Reifsnider KL (1979). Fatigue Damage Mechanisms in Composite
Materials: A Review. Fatigue Mechanisms: Proceedings of ASTM Symposium,
ASTM STP 675: 762-787.
553) Stone TJ, Babuska I (1998). A numerical method with a posteriori error estimation
for determining the path taken by a propagating crack. Comput. Methods Appl. Mech.
Engrg. 160: 245-71.
554) Stroud D (1998). Effective Medium Approximation: an update. Superlattices &
Microstructures 23(3/4): 567-73.
555) Sugimura Y, Grondin L, Suresh S (1995a). Fatigue crack growth at arbitrary angles
to bimaterial interfaces. Scripta met mat 33: 2007-12.
R-41

556) Sugimura Y, Lim P, Shih CF, Suresh S (1995b). Fracture normal to a bimaterial
interface: effects of plasticity on crack-tip shielding and amplification. Acta met 43:
1157-69.
557) Sumi Y, Nemat-Nasser S, Keer LM (1983). On crack branching and curving in a
finite body. Int J Fracture 21: 67-79.
558) Sumi Y, Wang ZN (1998). A finite-element simulation method for a system of
growing cracks in a heterogeneous material. Mech Mater 28: 197-206.
559) Suo Z, Bao G, Fan B (1992). Delamination R-curve phenomena due to damage. J
Mech Phys Sol 40: 1-16.
560) Suresh S, Shih CF (1986). Plastic near-tip fields for branched cracks. Int J Fracture
30: 237-59.
561) Suresh S (1985). Fatigue crack deflection and fracture surface contact:
Micromechanical models. Met Trans A 16: 249-60.
562) Suresh S, Mortensen A (1997). Functionally graded metals and metal-ceramic
composites: Part 2 Thermomechanical behaviour. Int Mat Rev 42(3): 85-113.
563) Suresh S (1998). Fatigue of Materials. 2nd ed, Cambridge University Press,
Cambridge, UK.
564) Suresh S (2001). Graded materials for resistance to contact deformation and damage.
Science 292: 2447-51.
565) Sutton MA, Deng X, Ma F, Newman JC, James M (2000). Development and
application of a crack-tip opening displacement-based mixed mode fracture criterion.
Int J Sol Struct 37: 3591-618.
566) Swanson PL, Fairbanks CJ, Lawn BR, Mai Y-W, Hockey BJ (1987). Crack-Interface
Grain Bridging as a Fracture Resistance Mechanism in Ceramics: I, Experimental
Study on Alumina. J Am Ceram Soc 70(4): 279-89.
567) Tada H, Paris PC, Irwin GR (1986). The Stress Analysis of Cracks Handbook, Del
Research, St Louis, USA.
568) Tamura I, Tomota Y, Ozawa H (1973). Strength and ductility of FeNiC alloys
composed of austenite and martensite with various strength. In: Proceedings of the
Third International Conference on Strength of Metals and Alloys, vol. 1. Cambridge:
Institute of Metals; 6115.
R-42

569) Tan CL, Wang X (2003). The use of quarter-point crack-tip elements for T stress
determination in boundary element method analysis. Engrg Fract Mech 70: 2247-52.
570) Taylor D, Zhou W, Ciepalowicz AJ, Devlukia J (1999). Mixed-mode fatigue from
stress concentrations: an approach based on equivalent stress intensity. Int J Fatigue
21(2): 173-8.
571) Taylor D (2002). Modelling of fatigue crack growth at the microstructural level.
Comput Mater Sci 25: 228-36.
572) Teoh SH (2000). Fatigue of biomaterials: a review. Int J Fatigue 22: 825-37.
573) Timoshenko T, Goodier JN (1970). Theory of Elasticity. 3rd ed. McGraw-Hill, New
York.
574) Tohgo K, Suzuki T, Araki H, Ishii H (2001). Evaluation of Fracture Toughness of
Ceramic/Metal Functionally Graded Material by Three-Point-Bending Test. In: RaviChandar K, Karihaloo BL, Kishi T, Ritchie RO, Yokobori AT, Yokobori T (eds).
Advances in Fracture Research, Proc. ICF10, Pergamon CDROM.
575) Tong J (2002). T-stress and its implications for crack growth. Engrg Fract Mech 69:
1325-37.
576) Torquato S, Yeong CLY, Rintoul MD, Milius DL, Aksay IA (1999). Elastic
Properties and Structure of Interpenetrating Boron Carbide/Aluminum Multiphase
Composites. J Am Ceram Soc 82(5): 1263-8.
577) Torquato S (1991). Random heterogeneous media: Microstructure and improved
bounds on the effective properties. Appl Mech Rev 44: 37-76.
578) Torquato S (1998). Morphology and effective properties of disordered heterogeneous
media. Int J Sol Struct 35: 2385-406.
579) Torquato S, Hyun S (2001). Effective-medium approximation for composite media:
Realizable single-scale dispersions. J Appl Phys 89(3): 1-4.
580) Trdegrd A, Nilsson F, stlund S (1998). FEM-remeshing technique applied to
crack growth problems. Comput Meth Appl Mech Eng 160: 115-131.
581) Tsukamoto H (2003). Analytical method of inelastic thermal stresses in a
functionally graded material plate by a combination of micro- and macro-mechanical
approaches. Composites: B 34: 561-8.

R-43

582) Tuchinskii LI (1983). Elastic Constants of Pseudoalloys with a Skeletal Structure.

Porosh Metall 7(247): 85-92. (Russian) Translated in: Powder Metallurgy, Plenum.
583) Tuchinskii LI (1983). Thermal expansion of composites with a skeletal structure.
Poroshkovaya Metallurgiya; 8(248): 73-80. (Russian) Translated in: Powder
Metallurgy; 659-664.
584) Tvergaard V, Hutchinson JW (1993). The influence of plasticity on mixed mode
interface toughness. J Mech Phys Sol 41: 1119-1135.
585) Tvergaard V, Hutchinson JW (1994). Toughness of an interface along a thin ductile
layer joining two elastic solids. Phil Mag A 70: 641-656.
586) Tvergaard V (2002). Theoretical investigation of the effect of plasticity on crack
growth along a functionally graded region between dissimilar elastic-plastic solids.
Eng Fract Mech 69: 1635-1645.
587) Tvergaard V (1999). Effect of plasticity on cleavage crack growth resistance at an
interface. J Mech Phys Sol 47: 1095-1112.
588) Tvergaard V (2003). Influence of plasticity on interface toughness in a layered solid
with residual stresses. Int J Sol Struct 40: 5769-5779.
589) Ueda S (2001). Elastoplastic analysis of W-Cu functionally graded materials
subjected to a thermal shock by micromechanical model. J Therm Str 24: 631-649.
590) Uzun H, Lindley TC, McShane HB, Rawlings T (2001). Fatigue crack growth
behaviour of 2124/SiC/10p functionally graded materials. Metall Mat Trans A 32:
1831-9.
591) Van Siclen CD (2003). Effective scalar properties of the critical region in functionally
graded materials. Physica A 322: 5-12.
592) Vasudevan AK, Sadananda K, Glinka G (2001). Critical parameters for fatigue
damage. Int J Fatigue 23; S39S53.
593) Vaughan DAJ, Guiu F, Dalmau MR (1987). Indentation fatigue of alumina. J Mat Sci
Lett 6(6): 689-91.
594) Vitale Brovarone C, Verne E, Krajewski A, Ravaglioli A (2001). Graded coatings
on ceramic substrates for biomedical applications. J Eur Ceram Soc 21: 2855-2862.
595) Vitek V (1977). Plane strain stress intensity factors for branched cracks. Int J Fracture
13(4): 481-501.
R-44

596) Walpole LJ (1969). On the overall elastic moduli of composite materials. J Mech
Phys Sol 17: 235-251.
597) Wang Y, Pan J (1999). Use of fracture mechanics to predict fatigue initiation life
under multiaxial loading conditions. Int J Fatigue 21(1): 173-180.
598) Wang BL, Mai Y-W, Noda N (2002). Fracture Mechanics Analysis Model for
Functionally Graded Materials with Arbitrarily Distributed Properties. Int J Fracture
116(2): 161177.
599) Wang YS, Gross D (2000). Analysis of a Crack in a Functionally Gradient Interface
Layer under Static and Dynamic Loading. Key Engrg Mater 183-187: 331-336.
600) Wang Z, Nakamura T (2004). Simulations of crack propagation in elasticplastic
graded materials. Mech Mater 36: 6010-6022.
601) Wegner LD, Gibson LJ (2000a). The mechanical behaviour of interpenetrating phase
composites I: modelling. Int J Mech Sci 42: 925-942.
602) Wegner LD, LJ Gibson (2000b). The mechanical behaviour of interpenetrating phase
composites II: a case study of a three-dimensionally printed material. Int J Mech Sci
42: 943-964.
603) Wegner LD, LJ Gibson (2001a). The mechanical behaviour of interpenetrating phase
composites III: resin-impregnated porous stainless steel. Int J Mech Sci 43: 1061-72.
604) Wegner LD, LJ Gibson (2001b). The fracture toughness behaviour of interpenetrating
phase composites. Int J Mech Sci 43: 1771-1791.
605) Weibull W (1939). A statistical theory of the strength of materials. Proceedings,
Royal Swedish Academy of Engineering Sciences, Stockholm, 151-158.
606) Westergaard HM (1939). Bearing pressures and cracks. J Appl Mech 61: A49-53.
607) Wheeler OE (1972). Spectrum loading and crack growth. J Basic Engrg 94: 181-186.
608) Willenborg J, Engle RM, Wood H (1971). A crack growth retardation model using an
effective stress intensity factor concept. Technical Report TFR 71-701. Rockwell, Los
Angeles.
609) Williams ML (1957). On the stress distribution at the base of a stationary crack. J
Appl Mech 24: 109-114.
610) Williams ML (1959). The Stresses Around a Fault or Crack in Dissimilar Media. Bull
Seismol Soc America 49:199-204.
R-45

611) Williams JG, Ewing PD (1972). Fracture under complex stress the angled crack
problem. Int J Fract Mech 8: 441-46.
612) Williamson RL, Rabin BH, Drake JT (1993). Finite Element Analysis of Thermal
Residual Stresses at Graded Ceramic-Metal Interfaces. Part 1: Model description and
geometrical effects. J App Phys 74(2): 1310-20.
613) Williamson RL, Rabin BH, Byerly GE (1995). FEM study of the effects of interlayers
and creep in reducing residual stresses and strains in ceramic-metal joints. Compos
Engrg 5: 851-863.
614) Willis JR (1977). Bounds and Self-Consistent Estimates for the Overall Properties of
Anisotropic Composites. J Mech Phys Sol 25: 185-202.
615) Wittmann FH, Hu X (1991). Fracture process zone in cementitious materials. Int J
Fract Mech 51: 3-18.
616) Winter AN, Steffler ED, Reimanis IE (2000). Deformation & Fracture of Graded NiAl2O3 Composites captured with Phase-shifted Moir Interferometry. Ceram Eng Sci
Proc 21(3): 691-96.
617) Wnuk MP (1998). Constitutive modelling of damage accumulation and fracture in
multiphase materials. Comput Meth Appl Mech Engrg 151: 587-591.
618) Whler A (1860). Versuche uber die Festigkeit der Eisenbahnwagenachsen.
Zeitschrift fur Bauwesen; 10, English summary: Engineering (1867) 4; 160-161.
619) Wu TT (1960). The effect of inclusion shape on the elastic moduli of a two-phase
material. Int J Sol Struct 2: 1-8.
620) Wu C-C, He P, Li Z (2002). Extension of J integral to dynamic fracture of functional
graded material and numerical analysis. Comput Struct 80: 411-416.
621) Xu H, Ostertag CP (1999). Crack closure stresses in fiber-reinforced brittle matrix
composites. J Eur Ceram Soc 19: 591-599.
622) Xu Q, Yu S, Kang Y (1999). Residual Stress Analysis of Functionally Gradient
Materials. Mech Res Comm 26(1): 55-60.
623) Xu FM, Zhu SJ, Zhao J, Qi M, Wang FG, Li SX, Wang ZG (2003a). Fatigue crack
growth in SiC particulate- reinforced Al matrix graded composite, Mater Sci Eng A
360: 191-196.

R-46

624) Xu FM, Zhu SJ, Zhao J, Qi M, Wang FG, Li SX, Wang ZG (2003b). Comparison of
the fatigue growth behaviour in homogeneous and graded SiC particulate reinforced
Al composite. J Mater Sci Lett 22: 899-901.
625) Xu FM, Zhu SJ, Zhao J, Qi M, Wang FG, Li SX, Wang ZG (2004). Effect of stress
ratio on fatigue crack propagation in a functionally graded metal matrix composite.
Compos Sci Tech 64(12): 1799-1803.
626) Xu L, Tippur HV, Rousseau CE (1999). Measurement of contact stresses using realtime shearing interferometry. Optic Engrg 38(11): 1932-1937.
627) Yamashita K, Watanabe C, Kumai S, Kato M, Sato A, Watanabe Y (2000). Cyclic
Deformation and Development of Dislocation Structures in a Centrifugally Cast AlAl3Ti Functionally Graded Material. Mater Trans JIM 41(10):1322-28.
628) Yan A-M, Marechal E, Nguyen-Dang H (2001). A finite-element model of mixedmode delamination in laminated composites with an R-curve effect. Compos Sci Tech
61: 1413-1427.
629) Yang YY, Munz D, Sckuhr MA (1997). Evaluation of the plastic zone in an elasticplastic dissimilar materials joint. Engrg Fract Mech 56(5): 691-710.
630) Yang QD, Thouless MD, Ward SM (1999). Numerical simulations of adhesivelybonded beams failing with extensive plastic deformation. J Mech Phys Sol 47(6):
1337-53.
631) Yang B, Ravi-Chandar K (1999). Evaluation of elastic T-stress by the stress
difference method. Engrg Fract Mech 64: 589-605.
632) Yang S, Yuan F-G (2000). Kinked cracks in anisotropic bodies. Int J Sol Struct 37:
6635-6682.
633) Yang B, Mall S, Ravi-Chandar K (2001). A cohesive zone model for fatigue crack
growth in quasibrittle materials. Int J Sol Struct 38: 3927-44.
634) Yau JF, Wang SS, Corten HT (1980). A Mixed-Mode Crack Analysis of Isotropic
Solids Using Conservation Laws of Elasticity. J Appl Mech 47: 335-341.
635) Zavalangios A (1997). Microstructure and residual stresses in FGMs. Composites and
Functionally Graded Materials; ASME, Materials Division, MD 80: 57-64.

R-47

636) Zhai P-C, Zhang Q-J, Yuan R-Z (1999). A Random Microstructure Finite Element
Method for Effective Properties of Functionally Graded Materials. Mat Sci Forum
308-311: 995-999.
637) Zhang Q-J, Zhai P-C, Liu L-S, Yuan R-Z, Moriya S-I, Niino M (2001). Recent
Development in the Computational Micromechanics, Thermal Damage Model and
Impact Response of Functionally Graded Materials. Ceram Trans 114: 659-66.
638) Zhang J, Stang H, Li VC (2001). Crack bridging model for fibre reinforced concrete
under fatigue tension. Int J Fatigue 23: 655-670.
639) Zhao FM, Okabe T, Takeda N (2000). The estimation of statistical fiber strength by
fragmentation tests of single-fiber composites. Compos Sci Technol 60 (10) 19651974.
640) Zhou W, Hu W, Zhang D (1999). Metal-matrix interpenetrating phase composite and
its in situ fracture observation. Mater Lett 40: 156-160.
641) Zhu J, M Gotoh (1999). Automatic remeshing of 2D quadrilateral elements and its
application to continuous deformation simulation: part I. remeshing algorithm. J
Mater Proc Technol 87: 165178.
642) Zienkiewicz OC, Taylor RL (2000). The Finite Element Method. Volume 1: The
Basis. 5th ed., Butterworth Heinmann, New York.
643) Zou ZZ, Wu SX, Li CY (2000). On the Multiple Isoparametric Finite Element
Method and Computation of Stress Intensity Factor for Cracks in FGMs. Key Engng
Mat 183-87: 511-16.
644) Zywicz Z, Parks DM (1992). Small scale yielding interfacial crack tip. J Mech Phys
Sol 40: 511-36.

R-48