Professional Documents
Culture Documents
06
Failure of Metals
A. PINEAU
Centre des Mate riaux Ecole des Mines de Paris, Evry Cedex, France
T. PARDOEN
Universite Catholique de Louvain, Louvain-la-Neuve, Belgium
2.06.1 INTRODUCTION 686
2.06.2 CLEAVAGE IN METALS 688
2.06.2.1 Introduction 688
2.06.2.2 Theories of Cleavage 689
2.06.2.2.1 Theoretical cleavage stress 689
2.06.2.2.2 Dislocation-based theories 690
2.06.2.3 Transgranular Cleavage of Ferritic Steels 692
2.06.2.3.1 Introduction 692
2.06.2.3.2 Multiple barrier models 693
2.06.2.3.3 Statistical aspects of cleavage fracture in steels 695
2.06.2.4 Transgranular Cleavage of Other Metals 700
2.06.2.4.1 Welds in HSLA steels: Influence of MA constituents 700
2.06.2.4.2 Cleavage fracture in other BCC metals 704
2.06.2.4.3 Cleavage fracture in HCP metals 704
2.06.2.4.4 Irradiation-induced embrittlement in ferritic steels 706
2.06.2.5 Intergranular Brittle Fracture in Ferritic Steels 708
2.06.3 DUCTILE FRACTURE IN METALS 709
2.06.3.1 Introduction: Two Classes of Failure Mechanisms 709
2.06.3.2 Plastic Localization Mechanisms in Homogeneous Medium 710
2.06.3.2.1 Necking under uniaxial tension 710
2.06.3.2.2 Plastic localization under biaxial loading conditions 713
2.06.3.3 Void Nucleation 715
2.06.3.3.1 Macroscopic evidences 715
2.06.3.3.2 Microscopic observations 715
2.06.3.3.3 Computational cell simulations 717
2.06.3.3.4 Void nucleation models 718
2.06.3.4 Void Growth 720
2.06.3.4.1 Macroscopic evidences 720
2.06.3.4.2 Microscopic observations 721
2.06.3.4.3 Void cell simulations 723
2.06.3.4.4 Void growth models 727
2.06.3.5 Void Coalescence 732
2.06.3.5.1 Macroscopic evidences 733
2.06.3.5.2 Microscopic observations 733
2.06.3.5.3 Void cell simulations 735
2.06.3.5.4 Models for the onset of void coalescence 737
2.06.3.5.5 Models for the coalescence process 740
2.06.3.6 Fracture Strain of Metals 743
2.06.3.6.1 Simple closed-form estimates of the fracture strain 743
2.06.3.6.2 More advanced predictions of the fracture strain 745
2.06.3.7 Fracture Toughness of Thick Ductile Metallic Components 751
2.06.3.7.1 Basics 752
2.06.3.7.2 Fracture initiation toughness 757
2.06.3.7.3 Ductile tearing resistance 759
684
2.06.3.8 Fracture Resistance of Thin Metallic Sheets 761
2.06.3.8.1 Introduction to the fracture mechanics of thin metallic sheets 761
2.06.3.8.2 The EWF method 763
2.06.3.8.3 Crack-tip necking work 764
2.06.3.8.4 Flat mode I fracture in thin plates 765
2.06.3.8.5 Competition between flat and slant fracture 768
2.06.3.8.6 General views about thickness dependence of fracture resistance 769
2.06.4 DBT IN FERRITIC STEELS 771
2.06.4.1 Introduction 771
2.06.4.2 DBT in fracture toughness tests 772
2.06.4.2.1 Introduction 772
2.06.4.2.2 A simplified approach 772
2.06.4.2.3 Advanced models 773
2.06.4.3 DBT under Charpy V impact testing 779
2.06.4.3.1 Introduction 779
2.06.4.3.2 Modeling Charpy V-notched impact test salient features 780
2.06.4.3.3 Other applications 781
2.06.5 CONCLUSIONS 782
2.06.6 REFERENCES 783
NOMENCLATURE
a crack length (defect size)
a
cs
critical largest crack arrest length
b equilibrium atomic spacing and
norm of the Burger vector
C cleavage crack length
C
0
thickness of the cementite platelet
CVN energy from Charpy V-notched
tests
d grain size
E Youngs modulus
f void volume fraction
f
0
initial void volume fraction
f
c
critical void volume fraction at
coalescence
f
f
critical void volume fraction at
final fracture
f
p
particle volume fraction
J J-integral
J
c
J-integral at cracking initiation
J
lc
J-integral at cracking initiation
under mode I plane strain
conditions
k elastic bulk modulus
k
y
Hall and Petch constant
k0
y
cleavage stress grain size-depen-
dent constant
K
c/f
Ia
critical stress intensity factor for
crack arrest at carbide/ferrite
interface
K
f/f
Ia
critical stress intensity factor for
crack arrest at ferrite/ferrite
interface
K
Imin
threshold stress intensity factor
for cleavage
K
Jc
critical stress intensity factor
determined from J
lc
under condi-
tions of large-scale yielding
L length of a pile-up
L
0x
initial void spacing along x
L
0z
initial void spacing along z
L
x
void spacing along x
L
z
void spacing along z
m Weibull shape factor
m9 rate sensitivity exponent
n. N strain-hardening exponent
(depends on the context of the
hardening law)
P(o) failure probability function
R
Cl
index for transition between
intergranular and cleavage fracture
R
0x
initial void radius along x
R
0z
initial void radius along z
R
x
void radius along x
R
z
void radius along z
s
ij
components of the deviatoric
stress tensor
S void shape parameter (=ln W)
S
0
initial void shape parameter
(=ln W
0
)
T stress triaxiality
U bonding energy
V
u
representative volume element
w
e
essential work of fracture
W void aspect ratio
W
0
initial void aspect ratio
W
p
particle aspect ratio
X
0
average distance between a crack
tip and the closest void
a crack advance
T
56J
transition temperature shift for a
56 Joules Charpy energy
T
K
lc.100
transition temperature shift for a
fracture toughness equal to
100 MPa m
1/2
relative void spacing
0
initial relative void spacing
p
initial relative particle spacing
d crack-tip opening displacement
Nomenclature 685
d
c
critical crack-tip opening displa-
cement at cracking initiation
e
c
void nucleation strain
e
ep
, e
p
e
effective (or equivalent) plastic
strain
e
en
effective (equivalent) strain in the
minimum section of a neck
e
f
fracture strain
e
p1
plastic strain in the direction of
the highest principal stress
e
p
y
average accumulated effective
plastic strain of the matrix
g
bas
plastic slip along basal plane
g
j
surface energy of a grain boundary
g
s
surface energy
g
int
s
breaking energy
G fracture energy per unit area of
crack advance
G
0
fracture energy per unit area of
crack advance associated to the
damage and material separation
in the fracture process zone
G
0
init
fracture energy per unit area of
crack advance at cracking initia-
tion associated to the damage and
material separation in the fracture
process zone
G
0
ss
steady-state fracture energy per
unit area of crack advance asso-
ciated to the damage and material
separation in the fracture process
zone
G
n
fracture energy per unit area of
crack advance associated to
crack-tip necking
G
n
init
fracture energy per unit area of
crack advance at cracking initia-
tion associated to crack-tip
necking
G
n
SS
steady-state fracture energy per
unit area of crack advance asso-
ciated to crack-tip necking
G
p
fracture energy per unit area of
crack advance associated to gross
plasticity
G
p
SS
steady-state fracture energy per
unit area of crack advance asso-
ciated to gross plasticity
m elastic shear modulus
Z geometric imperfection
l void distribution parameter
l
0
initial void distribution parameter
l
p
initial particle distribution
parameter
(m,k)CD ratio for the transition between
ductile and cleavage fracture
i Poissons ratio
s
0
yield stress
s
1
maximum principal stress
s
c
theoretical cleavage stress
o
d
fracture stress of a particle
s
e
equivalent (effective) von Mises
stress
s
f
cleavage stress
s
G
Griffith fracture stress
s
h
hydrostatic stress
s
ij
component of the Cauchy stress
tensor
s
th
threshold stress
s
u
cleavage stress parameter in
Weibull distribution analysis
s
w
Weibull stress
s
w min
minimum value of Weibull stress
for cleavage
s
y
current yield or flow stress
t shear stress
t
i
friction stress
z void packing geometric parameter
2.06.1 INTRODUCTION
The study of the micromechanisms of failure
plays a key role in the development of engineer-
ing metallic alloys, in manufacturing, and in the
assessment of the mechanical integrity of struc-
tures. For example, in the steel industry, the
development of new alloys occurs rapidly
since about one half of existing compositions
are replaced by new compositions every 5 years.
In todays automotive industry, dual-phase and
other multiphase steels are quickly becoming
one of the most popular and versatile materials.
Currently, these steels are most commonly used
in structural applications where they have
replaced more conventional high-strength low-
alloy (HSLA) steels. They offer a great
opportunity for weight reduction. The develop-
ment of these new compositions requires a
perfect knowledge of their deformability, in
particular their forming limit diagrams
(FLDs) and cracking resistance essential for
controlling crash worthiness. Another area
where the study of the micromechanisms of
failure is essential is the assessment of the
mechanical integrity of structures, in particular
flawed mechanical structures. These flaws
appear either during manufacturing or during
service conditions. Developing damage-
tolerant microstructures is thus essential in
many fields of engineering. For a long time,
these developments have remained essentially
empirical. However, more recently, new meth-
odologies have been introduced.
686 Failure of Metals
The first objective of this chapter is to present
an overview of the new methodologies which are
based on the investigation of the micromechan-
isms at a local scale and, through a multiscale
approach, on the transfer of these local informa-
tion to the macroscale. Indeed, nowadays, the
final goal is to develop predictive approaches
which can be used in finite element codes for
structural analysis or for simulation of forming
operations. Several excellent reviews and books
have already been written on the micromechan-
isms of failure in metals (see, e.g., Knott, 1973;
Besson 2004), but very few of them have
attempted to provide a comprehensive synthesis
of the state-of-the-art predictive approaches.
This is one of the goals of this chapter. In parti-
cular, in the present study, a special effort was
made to incorporate the most recent develop-
ments in the theoretical and numerical
modeling of both ductile and brittle fracture.
The second objective of this chapter is to
introduce the four tools available for microstruc-
tural/micromechanical investigations and to
show their complementarity. These tools include:
(1) macromechanical tests under various loading
conditions (iso- and non-isothermal tests, multi-
axial tests, etc.); (2) advanced characterization
methods (scanning and transmission electron
microscopy, electron back-scattered diffraction,
X-ray tomography, with an emphasis on in situ
testing where mechanical load is combined with
the characterization); (3) computational unit cell
calculations used to investigate the mechanical
response of elementary volume elements
(RVEs) or to simulate numerically crack initia-
tion and crack growth; and (4) theoretical models
which remain essential tools in structural analysis
and to formulate analytical expression, the best
for revealing the essence of the physics.
The third objective is to address the main
mechanisms of fracture based on the use of these
four tools. In this chapter we deal only with duc-
tile and brittle (cleavage and intergranular)
fracture. These two modes of failure are analyzed
separately. An attempt is made to cover a wide
range of materials including steels and other BCC
materials (Mo, Nb), HCP metals (Zn, Mg), alu-
minum alloys, titanium alloys, etc. Ductile
fracture by nucleation, growth, and coalescence
of voids, and by plastic localization are addressed.
The necessity of introducing a characteristic
length or volume when dealing with crack-tip
singularities is evident. Then the ductile-to-brittle
transition (DBT) in ferritic steels is reviewed. A
special emphasis is also laid on the effect of irra-
diation embrittlement in ferritic steels.
A fourth objective of this chapter is to cover
a wide range of metals and metallic alloys. In
steels, both ferritic and austenitic microstruc-
tures are considered. Multiphased and
quenched-tempered steels are also included.
An attempt has also been made to underline a
number of specific aspects encountered with
welds, in particular the mismatching effect and
the problems related to the variety of micro-
structures encountered in these welds. Many
examples dealing with cast and wrought alumi-
num alloys are given. A number of other metals
are also considered.
Many other topics related to the mechanisms
of failure in metals, such as fracture at high
temperature and environmentally assisted
cracking or most generally most of the coupling
effects with chemistry (see Volume 6), are not
covered in this chapter. It should also be added
that while an attempt has been made to give
credit through extensive references, many of
them have been omitted. Moreover, for the
sake of simplicity, many examples are often
extracted from our own researches.
These four objectives are those of what is now
called the local approach to fracture. The influ-
ence of crack-tip constraint and stress triaxiality
on ductile and brittle fracture is of major impor-
tance for the assessment of structural integrity.
This assessment is usually made by means of
linear and nonlinear fracture mechanics con-
cepts (for a review, see Chapter 2.03).
Constraint is a structural feature which inhibits
plastic flow and causes a higher triaxiality of
stresses. Local stress triaxiality promotes void
growth on the micromechanical level and thus
causes damage in the process zone located at
the crack tip. Constitutive equations that
account for damage as, for example, the
Gurson potential (Gurson, 1977), must hence
be able to describe the physical effect of cons-
traint on the tearing resistance in a natural
way. Compared with conventional fracture
mechanics concepts, micromechanical models
developed in the frame of the local approach to
fracture have the advantage that the correspond-
ing material parameters for ductile fracture can
be transferred in a more general way between
different specimen geometries. It is not even
necessary to regard specimens with an initial
crack as, of course, initially uncracked structures
also will break if the local degradation condi-
tions of the material have exceeded some critical
state. In the Gurson model, initiation and pro-
pagation of a crack are a natural result of the
local softening of the material due to the void
coalescence which starts when a critical void
volume fraction, f
c
, is reached over a character-
istic distance l
c
. The parameter f
c
can therefore
be determined from rather simple tests, for
example, tensile tests of smooth and notched
round bars in combination with numerical ana-
lyses of these steels, or from micromechanical
models. Similarly, the Weibull stress model
Introduction 687
originally proposed by Beremin (1983) provides
a framework to quantify their complex interac-
tion among specimen size, deformation level,
and material flow properties when dealing with
brittle cleavage (or intergranular) fracture. The
original Beremin model includes only two para-
meters which can be determined by testing
smooth or notched bars.
The identification and determination of the
damage parameters in the Gurson or in the
Beremin model require a hybrid methodology of
combined testing and numerical simulation. The
description of this methodology is out of the
scope of this chapter. More details are given in
Chapter 7.05. Here, it is enough to say that,
different from classical fracture mechanics, the
local approach procedure is not subject to any
size requirements for the specimens as long as the
same fracture phenomena occur, that is, cleavage
fracture or the mechanisms of void nucleation,
and growth and coalescence.
This chapter is organized according to fail-
ure modes: cleavage, ductile fracture, and
DBT. In Section 2.06.2, the early theories
for this mode of failure are presented first.
Then these theories and more recent theoreti-
cal developments are applied to cleavage
fracture in ferritic steels and other metals
with either a BCC or an HCP microstructure.
Intergranular brittle fracture in steels is also
reviewed in this section. Two modes of ductile
fracture are distinguished. The first one
occurs in metal forming and is associated
with plastic localization mechanisms which
are briefly reviewed. The second one involves
void nucleation, growth, and coalescence.
This mode of ductile fracture is reviewed in
details. After presenting recent results on the
modeling of these three steps involved in duc-
tile fracture, an attempt is made to show how
it is possible to predict general trends in the
variation of fracture strain with mechanical
loading conditions, in particular stress triaxi-
ality, and with the parameters representing
the second-phase particles (shape, spacing,
volume fraction, etc.) responsible for cavity
initiation. The application of these theories
to fracture toughness of thick components
and fracture resistance of thin sheets is then
presented. Finally, Section 2.06.4 is devoted
to the study of the DBT observed in ferritic
steels. Simplified and more advanced models
accounting for this transition in fracture
modes are presented. Two extreme loading
conditions are considered: quasi-static similar
to those met during conventional fracture
toughness tests and Charpy V impact testing.
A number of applications of the methodology
presented in this chapter to the DBT, includ-
ing welds and irradiation embrittlement, are
given. However, most applications are given
in Chapter 7.05.
2.06.2 CLEAVAGE IN METALS
2.06.2.1 Introduction
Cleavage and intergranular cracking are the
most detrimental modes of fracture. Cleavage
fracture occurs preferentially over dense atomic
planes. Table 1 lists some cleavage planes that
have been observed experimentally. Two exam-
ples of cleavage fracture observed in a high-
strength ferritic steel (a,b) and in pure zinc (c)
are shown in Figure 1. These figures reveal that
the orientation of cleavage plane changes when
it crosses sub-boundaries, twin boundaries, or
grain boundaries, and steps appear on the frac-
ture surface to compensate for the local
misorientation. In the case of mechanical twins,
these steps look like indentation marks called
tongue (Figure 2). In order to maintain the
equilibrium of the crack front, the nearest steps
gather to form a single step of larger height
leading to the formation of rivers, as observed
in Figure 1b. These rivers align with the direction
of the local propagation of the cleavage cracks.
On a macroscopic scale the surfaces of the clea-
vage facets are normal to the maximum
principal stress. In fracture mechanics terminol-
ogy, this is called mode I fracture.
Another brittle mode of fracture observed in
polycrystalline metals corresponds to intergra-
nular fracture. If a crack forms along a grain
boundary having a surface energy,
j
, the
breaking energy, 2
s
int
, of the atomic bonds
must be reduced by
j
. Hence, one could think
that intergranular fracture will be easier than
transgranular cleavage. However, the aniso-
tropy in the surface energy within the crystals,
s
, must also be taken into account. The surface
energy
s
for a crystallographic cleavage plane
is always less than the surface energy of an
intergranular surface by typically a factor of
1.20. In order to characterize the transition
Table 1 Cleavage planes
Structure
Cleavage
plane
Some
materials
BCC {100} Ferritic steels, Mo; Nb, W
FCC {111} Very rarely observed
HCP {0002} Be, Mg, Zn
Diamond {111} Diamond, Si, Ge
NaCl {100} NaCl, LiF, MgO, AgCl
ZnS {110} ZnS, BeO
CaF
2
{111} CaF
2
, UO
2
, ThO
2
Francois, D., Pineau, A., and Zaoui, A. 1998. Mechanical
Behaviour of Materials. Kluwer, Dordrecht.
688 Failure of Metals
between intergranular and cleavage fracture,
the parameter R
CI
is defined as
R
CI
=
2g
int
s
g
j
2g
s
= 1.20
g
j
2g
s
[1[
Intergranular fracture is favored when R
CI
<1.
Cottrell (1989, 1990a, 1990b) has shown that
j
depends mainly on the shear modulus, j,
while
s
depends on the bulk modulus, k. This
means that the parameter R
CI
is a function of
the ratio j/k and can thus be written as
R
CI
= 1.20 am,k [2[
where a is a numerical constant close to 1. In
nickel, for instance, a was found to be equal to
0.95. Table 2 gives the values of R
CI
for a num-
ber of metals. The table shows that, in a large
number of pure metals including Fe, intergra-
nular fracture should be the preferential mode
of fracture. However, in many cases, the segre-
gation of impurities like carbon in iron tends to
suppress intergranular brittleness.
2.06.2.2 Theories of Cleavage
2.06.2.2.1 Theoretical cleavage stress
The normal stress, o
c
, theoretically needed to
fracture a crystal by cleavage can be easily
determined provided that the bonding energy,
U, between the atoms located across the clea-
vage plane is known. The force required to
separate cleavage planes is the derivative of
this energy with respect to distance. As the dis-
tance between the lattice planes increases, the
10 m
5 m
20 m
Figure 1 a, Cleavage microcracks observed on longitudinal sections in a low-alloy steel (Lambert-Perlade,
2001); b, fracture surface in a low-alloy steel (Lambert-Perlade, 2001) showing the presence of rivers; c, fracture
in polycrystalline zinc. Rivers originating from a grain boundary.
Figure 2 SEM micrograph of a fracture surface of a
low-alloy steel. The arrows indicate the presence of
tongues corresponding to the intersection of the
main (001) fracture plane with mechanical twins.
Cleavage in Metals 689
stress, which is zero at the distance b corre-
sponding to interatomic equilibrium, goes
through a maximum which corresponds to the
value of the cleavage stress, o
c
. Cleavage frac-
ture requires the energy of two new surfaces
associated with the formation of a pair of new
surfaces, 2
s
, per unit area of new surfaces.
Assuming that the variation of the force with
displacement of the crystallographic planes is
sinusoidal leads to (Franc ois et al., 1998)
s
c
= (Eg
s
,b)
1,2
[3[
where E is Youngs modulus. With the typical
values E=200GPa, b =0.3 nm,
s
~0.1jb
~1 J m
2
, eqn [3] leads to o
c
=26GPa ~E/10.
This theoretical value of o
c
~E/10 for the clea-
vage stress is much higher than the experimental
values found for classical metallic materials (typi-
cally 1 GPa in steels). However, for a number of
whiskers (i.e., small filaments free of disloca-
tions), the measured values are of the same
order as the theoretical value, providing a quali-
tative validation of theoretical calculation.
The reasons for the large difference between
the observed and the calculated values for o
c
are twofold. In crystalline ceramic materials in
which brittle fracture occurs under purely elas-
tic conditions, that is, without the nucleation
and the propagation of dislocations, the clea-
vage stress is related to the existence of defects
which are inherently present. In this case, the
fracture stress is given by the Griffith stress
s
G
= (Eg
s
,pa)
1,2
[4[
where a is the size of the large defects.
In crystalline metallic materials, the reason
for the difference between the theoretical value
for o
c
and the experimental cleavage stress is
different. Cleavage in these materials is always
accompanied by plastic deformation. In other
words, plastic deformation is a prerequisite to
initiate cleavage fracture. The dislocations pro-
duce stress concentrations which are sufficient
to reach locally the theoretical cleavage stress.
This is the basis of the theories which are briefly
presented in the following.
Before introducing these theories it is worth
noting that, in crystalline solids, cleavage
cracks blunt by the emission of dislocations.
Rice and Thomson (1974) investigated the con-
ditions under which this mechanism operates.
They derived a criterion for the transition
between pure and blunted cleavage. This transi-
tion occurs when the ratio between the shear
modulus, j, and the bulk modulus, k, reaches a
critical value given by
(m,k)CD = 10g
s
,bk [5[
The propensity for blunted cleavage increases
with increasing ratio (m,k)CD. A number of
values for this ratio are given in Table 2, which
shows that, in almost all metallic materials,
blunted cleavage is the rule. More recently, this
theoretical transition between blunted cleavage
and pure cleavage has been reanalyzed by Rice
et al. (1992). These authors used the Peierls con-
cept to analyze dislocation nucleation from a
crack tip. They showed that in most FCC metals,
except iridium, blunted cleavage should always be
observed. Conversely, in most BCC metals, pure
cleavage should be observed before the nucleation
of dislocations ahead of a crack tip. These theo-
retical calculations are useful to explain the
DBT in materials like silicon or pure chromium
which contain initially a very low density of dis-
locations but they do not apply to engineering
materials. In these materials, cleavage fracture is
explained by the existence of cleavage initiation
sites and the stress intensification produced by
plastic deformation, as detailed below.
A closing remark must be made in this intro-
duction devoted to the concept of cleavage
stress. The results of brittle fracture are inher-
ently scattered, and Section 2.06.2.3.3 focuses
on the statistical aspects of cleavage fracture. In
particular, the Weibull stress concept, which is
different from the cleavage stress concept, is
central to address this stochastic behavior.
2.06.2.2.2 Dislocation-based theories
(i) Initiation-controlled cleavage
The formation of slip bands and, under given
circumstances, of mechanical twins during defor-
mation are the sources of stress concentration.
This is illustrated in Figure 3 where a pile-up of
Table 2 Transition parameters for fracture: j/k is the ratio of the shear modulus to bulk modulus, R
CI
quantifies the risk of intergranular fracture (if smaller than 1) vs cleavage; (j/k)CD is the ratio required to for
the transition between cleavage and ductile fracture
Metal Au Ag Cu Pt Ni Rh Ir Nb Ta V Fe Mo W Cr
j/k 0.11 0.19 0.22 0.24 0.34 0.52 0.52 0.25 0.31 0.32 0.33 0.48 0.52 0.82
R
CI
1.09 1.02 0.99 0.97 0.87 0.71 0.70 0.97 0.91 0.89 0.88 0.75 0.71 0.42
(j/k)CD 0.32 0.43 0.57 0.38 0.49 0.39 0.32 0.59 0.55 0.65 0.56 0.35 0.45 0.68
Francois, D., Pineau, A., and Zaoui, A. 1998. Mechanical Behaviour of Materials. Kluwer, Dordrecht.
690 Failure of Metals
n dislocations is blocked by a grain boundary.
Many variants have been proposed for this ele-
mentary mechanism (see, e.g., Zener, 1949;
Stroh, 1954). For a pile-up of length 2L, the
normal stress at a distance r from the grain
boundary in a direction W is given by
s = (t t
i
)(L,2r)
1,2
f(W) [6[
where t and t
i
are the applied resolved shear
stress and the lattice friction stress, respectively.
It is assumed that cleavage is initiated when the
applied stress and thus the resolved shear stress,
t, reaches the theoretical critical value, o
c
, over
a sufficiently long distance r =X
c
. Equation [6]
leads to
t t
i
=
s
c
(X
c
,d)
1,2
f(W)
,
(E,10)(X
c
,d)
1,2
f(W)
[7[
Since the length, L, which characterizes the size
of the dislocation pile-up, is a linear function of
the grain size, d, that is, L , d/2, eqn [7] predicts
that the stress necessary to initiate cleavage frac-
ture varies like 1,
d
_
. Moreover, as the friction
stress, t
i
, is strongly temperature dependent, this
equation also shows that the stress for cleavage
initiationstrongly increases whenthe temperature
is decreased. In many cases, in particular in steels,
it has been shown that, within a first approxima-
tion, the cleavage stress does not depend on
temperature. This strongly suggests that cleavage
is not initiation controlled, otherwise a tempera-
ture dependence should be observed.
(ii) Growth-controlled cleavage
The above calculation of the stress necessary
to initiate a cleavage crack has not addressed
the question of whether the process is possible
on energy grounds. This problem was studied
by Cottrell (1958), who assumed that a {100}
cleavage was initiated in a BCC material by the
so-called self-blocking mechanism of two
{110} slip systems (Figure 4). When a cleavage
crack of length C appears, the dislocations in
the two pile-ups climb rapidly in the crack
which can thus be considered as a sessile dis-
location with a Burgers vector nb and a core
whose size, according to Cottrell, is C/2.
Simultaneously, two surfaces are created, and
a potential energy equal to (1 i
2
)o
2
C
2
/2E is
released (i is the Poisson ratio). Thus, the
change in energy is given by
U
t
=
m(nb)
2
4p(1 i)
log
2R
C
_ _
2g
s
C
s
2
pC
2
2E
1 i
2
_ _
[8[
Writing the conditions of instability
(cU
t
/cC<0), it is found that the critical value
for the growth of a cleavage crack is given by
s = s
f
=
2g
s
nb
[9[
As nb =(1 i)L(t t
i
)/j (Friedel, 1964),
eqn [9] leads to
s
f
(t t
i
) =
2mg
s
p(1 i)d
[10[
The effective stress (t t
i
) acting on the dis-
locations in the pile-ups is a function of grain
size according to Hall and Petch (Petch, 1953):
t t
i
= k
y
d
1,2
[11[
2L
r
n
D
is
lo
c
a
tio
n
s
o
u
r
c
e
=
K
Ia
1
2 c/f
D
=
K
Ia
1
2
D
f /f
Figure 6 Multiple barrier model. Three events are
schematically shown with their probability of
occurrence (Martin-Meizoso et al., 1994).
Table 3 Parameters of multiple barrier models
Literature data
Parameter Present study value Value Microstructural unit Reference
o
d
(MPa) 2112 Lambert-Perlade et al. (2004)
K
Ia
c/f
(MPa m
1/2
)
7.8
2.55.0 Carbides Martin-Meizoso et al. (1994)
2.5 Globular carbides Hahn (1984)
1.8 TiN particles Rodrigues-Ibabe (1998)
K
Ia
f/f
(MPa m
1/2
) CGHAZ-25 5.07.0 Bainite packets Martin-Meizoso et al. (1994)
28 7.0 Bainite packets Martin-Meizoso et al. (1994)
7.5 Ferrite grains Hahn (1984)
ICCGHAZ-25 4.8 Bainite packets Rodrigues-Ibabe (1998)
18 15.2 Bainite packets Rodrigues-Ibabe (1998)
694 Failure of Metals
Griffith criterion is located above all the curves
corresponding to the elastic viscoplastic mate-
rial. This situation may appear as being rather
counterintuitive since plastic flow is expected to
increase the resistance to crack growth.
However, as stated by the authors, the Griffith
curve is valid for a stationary crack, whereas in
these numerical simulations the crack has a
significant speed when it reaches the carbide
ferrite interface. Figure 8 shows also that the
stress levels required to arrest a microcrack
vary at a much lower rate with decreasing car-
bide size compared to the Griffith stress. This
implies that small carbides may play a more
prominent role in cleavage fracture of ferritic
steels than what might be expected from the
straightforward application of the static
Griffith criterion. The strain rate sensitivity
and the dynamic aspect of crack growth thus
come into play in the initiation and in the con-
tinued growth of a cleavage crack. Figure 8 also
suggests that if the stress triaxiality level is
decreased from 0.75 to 0.60, the critical stress
level, required to initiate a critical microcrack,
decreases by an amount approximately equal to
the initial yield stress of the ferrite matrix, o
0
.
This effect of stress triaxiality on cleavage frac-
ture initiation is opposite to what is normally
seen in cleavage fracture experiments, where a
decrease in crack-tip constraint leads to an
increase of the fracture toughness. However,
this last conclusion is not valid since, in this
study, the mechanism responsible for the initia-
tion of carbide cracking was not considered
because of a precrack being introduced in the
carbides. The deleterious effect of the increase
in stress triaxiality level on the nucleation of
microcracks from particles is described by eqn
[15]. Further results were obtained in the study
by Kroon and Faleskog (2005), such as a rela-
tionship between the applied axial stress S
z
and
the critical largest crack arrest length a
cs
, and
this relation is independent of the carbide size
and of the level of stress triaxiality.
In spite of all the researches devoted to the
study of cleavage micromechanisms in ferritic
steels, many questions remain open and both
experimental and theoretical investigations are
still necessary to reach full transferability to
real life engineering problems.
2.06.2.3.3 Statistical aspects of cleavage
fracture in steels
(i) Beremin model
Rather surprisingly, although the scatter in
cleavage stress measurements is well known
since a long time, it was only in the 1980s that
models have been proposed to account for this
scatter (for review papers, see, e.g., Wallin et al.,
1984, 1991a, 1991b). Nowadays, the most
r
z
r
2c
2R
0
2
H
0
2
h
(a) (b)
Figure 7 a, A cracked grain boundary carbide in ferrite. b, The axisymmetric model with a carbide embedded
in ferrite (Kroon and Faleskog, 2005).
7
6
5
4
3
2
0 2 4 6
c (m)
Griffith criterion
8 10 12
z
/
0
= 0.75
= 0.70
= 0.65
= 0.60
Figure 8 Overall stress S
z
/o
0
as a function of critical
largest carbide size c for four levels of stress
triaxiality, C=S
r
/S
z
. Comparison with Griffith
criterion (Kroon and Faleskog, 2005).
Cleavage in Metals 695
largely used models are those derived from the
work by Beremin (1983) (for a review of these
models, see Mudry, 1988; Pineau, 1992, 2006; see
also Evans, 1983; see Chapter 7.05). Assuming
that the material contains a population of micro-
defects (particles or grain-sized microcracks)
distributed according to a simple (power or
exponential) law, p(a), the weakest link theory
states that the probability to failure P(o) of a
representative volume element V
u
is given by
P(s) =
_
a
c
(s)
p(a) da [17[
where a
c
is simply given by eqn [4], that is,
a
c
=
2Eg
s
as
2
[18[
where a is a numerical constant. Knowing the
distribution p(a), it is therefore possible to cal-
culate the associated distribution P(o).
In a volume V which is uniformly loaded and
which contains V/V
u
statistically independent
elements, the probability to failure can thus be
expressed as
P
R
= 1 exp
V
V
u
P(s)
_ _
[19[
As a general rule the function p(a) is not
known. However, when the critical step for
cleavage fracture is the propagation of micro-
cracks initiated from particles, the distribution
p(a) can be determined experimentally. Two
types of laws are usually proposed:
v a power law
p(a) = ga
b
[20[
v an exponential law including, if necessary, a
cutoff parameter (see, e.g., Carassou et al.,
1998; Lee et al., 2002) such as the cumulated
probability is given by
p(size>a) = exp
a a
u
a
0
_ _
n
_ _
[21[
where a
u
and a
0
are parameters of the dis-
tribution. The simple power law leads to the
well-known Weibull expression
P
R
= 1 exp
V
V
0
s
s
u
_ _
m
_ _
[22[
with the Weibull shape factor
m = 2b 2 [23[
It should be noted that, within a first approx-
imation, m and o
u
are temperature
independent. Similarly, the exponential law
given by eqn [21] leads to (Tanguy et al., 2003)
P
R
=1 exp
V
V
0
(1,s
2
) (1,s
2
u
)
(1,s
2
c0
)
_ _
n
_ _ _ _
[24[
s
c0
= (2Eg
s
,aa
0
)
1,2
and s
u
= (2Eg
s
,aa
u
)
1,2
[25[
Equation [22] is a simplified expression since no
threshold is introduced. In three dimensions
and in the presence of smooth stress gradients,
this equation can be written as
P
R
= 1 exp
_
PZ
s
m
1
dV
s
m
u
V
u
_ _
[26[
where the volume integral is extended over the
plastic zone (PZ). This equation can be rewritten
as
s
w
=
_
PZ
s
m
1
dV
V
u
m
[27[
where o
w
is referred to as the Weibull stress.
A number of investigators have introduced a
threshold stress, o
th
, directly into eqn [27] (see,
e.g., Bakker and Koers, 1991; Xia and Cheng,
1997). One possible expression for the inte-
grand of eqn [26] has the form [(o
1
o
th
)/
(o
u
o
th
)]
m
. However, rational calibration pro-
cedures for o
th
remain an open issue. In order to
avoid these difficulties, Gao et al. (1999) pro-
posed a modified form of eqn [27] given by
P
R
= 1 exp
s
w
s
w min
s
u
s
w min
_ _
m
_ _
[28[
where o
wmin
represents the minimum value of
o
w
at which cleavage fracture becomes possible
(for a full discussion, see Gao et al., 1998a,
1998b; Gao and Dodds, 2000).
In the original Beremin model (Beremin,
1983), an implicit threshold Weibull stress was
also included, a point which is somewhat for-
gotten in the literature. In this model, it is
assumed that cleavage fracture cannot occur
in the absence of plastic deformation, that is,
below the yield strength, o
0
. This means that
ahead of a crack tip the PZ size must be larger
than a critical size, X
c
, or otherwise stated that
there exists a threshold for the stress intensity
factor, K
Imin
, below which cleavage fracture
cannot occur. This threshold is given by
K
Imin
~ s
0
3pX
c
_
[29[
It should also be noted that the exponential
law with a cutoff parameter (eqn [21]) used to
describe the particle distribution leads also to a
threshold stress given by eqn [24].
The original Beremin model contains only
two independent parameters, m and the pro-
duct o
u
m
V
u
. A number of studies performed on
quenched and tempered steels used in reactor
696 Failure of Metals
pressure vessels (RPVs) have shown that m ~
20 when no threshold is explicitly used, that is,
when the Weibull stress is calculated using eqn
[27] (see, e.g., Beremin, 1983). Similar values
have been reported on structural steels with a
yield strength between 490 and 685 MPa
(Minami et al., 2002). The strategy used for
the calibration of Weibull parameters has been
described in details elsewhere (Chapter 7.05).
Values of m lower than 20 are found when a
threshold is introduced (see, e.g., Gao and
Dodds, 2000; Petti and Dodds, 2005a).
In the Beremin model, when the value of V
u
is
fixed, the value of o
u
reflects the resistance of
the material to brittle cleavage fracture. This is
illustrated in Figure 9 where the fracture tough-
ness temperature dependence of two
low-alloyed steels used in the fabrication of
pressurized water reactors was studied, that is,
A 508 steel and a modern 2Cr1Mo steel
(Bouyne et al., 2001). These materials have
similar yield strength at room temperature but
different microstructures.
As observed in Figure 10, the A 508 steel has
an upper bainite microstructure, while 2Cr
1% steel shows a lower bainite microstructure.
Figure 9 shows that the ductile-to-brittle transi-
tion temperature (DBTT) is lower in the
2 Cr1Mo steel than in the A 508 steel.
Tests on notched specimens showed that the
same ranking was observed when the values of
o
u
were measured adopting the same value for
V
u
(50 mm)
3
. In A 508 steel, it was found that
m=22 and o
u
=2600 MPa (Beremin, 1983),
while in 2 1/4 Cr1Mo steel, it was found that
m=20 and o
u
=3500 MPa (Bouyne, 1999).
These values for o
u
and m are typically those
encountered in quenched and tempered steels.
The Beremin model has also been applied to
multiple barrier models. As stated previously,
this model is essentially based on the descrip-
tion of the propagation of an existing critical
defect belonging to a single population. This is
a simplification which might explain why in the
application of the model over a wide range of
temperatures, a number of investigators have
reported that it was necessary to assume that
the normalizing stress, o
u
, was an increasing
function of temperature (see, e.g., Tanguy
et al., 2005a, 2005b). This might simply reflect
the existence of different critical steps depend-
ing on temperature, as indicated earlier.
Multiple barrier models would therefore
appear more satisfactory to account for the
variation of cleavage fracture toughness over
a wide temperature range. In particular, mul-
tiple barrier models addressing the fracture
process schematically shown in Figure 11
have been proposed (Martin-Meizoso et al.,
1994; Lambert-Perlade et al., 2004). These
models are also based on the weakest link
theory. The nature of these barriers depends
on temperature. Application of these models
requires the knowledge of a certain number of
metallurgical factors including the nucleating
particle size distribution and the grain
(packet) size distribution. These factors were
measured in one specific steel in which the
brittle particles were formed by MA constitu-
ents (Lambert-Perlade et al., 2004), that is,
mixed MA particles. The application of
these models also necessitates the knowledge
of the local fracture toughness, K
Ia
c/f
and
K
Ia
f/f
, and that of the cleavage fracture stress
of particles, o
d
. There are very few results in
the literature. However, a number of results
400 A 508
2 Cr
300
200
100
K
I
c
(
M
P
a
m
1
/
2
)
0
200 150 100 50
Temperature (C)
0 50
Ring
CT25
CT25
CT20
CT20
CT50
CT50
CT195
CT100
Figure 9 Fracture toughness of two pressure vessel steels (A 508 and 2Cr1Mo steels) as a function of test
temperature. The DBT temperature in 2Cr1Mo steel with a lower bainite microstructure is much lower than
that of A 508 steel with an upper bainite microstucture.
Cleavage in Metals 697
(c)
25 m
50 m
(a)
(b)
(d)
Figure 10 Microstructures of pressure vessel steels: a, b, A 508 steel; c, d, 2Cr1Mo steel. A 508 steel
exhibits an upper bainite microstructure containing relatively coarse carbide particles, while 2Cr1Mo steel
has a lower bainite microsctructure with smaller carbides.
(a) (b)
(d) (c)
Figure 11 Schematic representation of the role of microstructural barriers on fracture micromechanisms. The
crack is assumed to nucleate from an intragranular particle: a, undamaged material; b, microcrack initiation
and propagation in the particle; c, microcrack propagation across the particle/matrix interface; and
d, microcrack propagation across a grain boundary or a bainite high-angle packet boundary leading to final
fracture (Lambert-Perlade et al., 2004).
698 Failure of Metals
are reported in Table 3. In the study devoted
to a bainitic steel containing MA particles, it
was assumed that the local values of fracture
toughness, K
Ia
c/f
and K
Ia
f/f
, were not temperature
dependent, which is a crude approximation.
In spite of this approximation, a good agree-
ment was found between the experimental
values of the fracture toughness and those
inferred from this multiple barrier model, as
shown later.
(ii) Effect of plastic strain
The Weibull expression in eqn [27], which is
the product of a stress function and a volume
(PZ), is based on the assumption that there
exists a population of microcracks nucleated
at the onset of plastic deformation. These
microcracks remain active during the entire
loading history. This is an oversimplification
which requires further discussion.
Detailed investigations on a number of steels
have shown that the number of microcracks
nucleated from carbides was an increasing func-
tion of plastic strain and was increasing with
decreasing temperature (see, e.g., Kaechele and
Tetelman, 1969; Mc Mahon and Cohen, 1965).
More recently in a study devoted to the initiation
of microcracks from niobium carbides and tita-
nium nitrides in an Ni-based superalloy (In 718),
it was shown that the probability to failure of
these particles could be written as
P
nucl
= 1 exp
s
1
l(s
e
s
0
)
s
uN
_ _
N
a
[30[
where o
uN
is a normalizing stress and N
a
is a
material constant (Alexandre et al., 2005).
Similar studies are lacking for a proper descrip-
tion of the nucleation of microcracks from
carbides and nonmetallic inclusions in ferritic
steels. This is why the recent modifications to the
Beremin model presented below are essentially
based on phenomenological considerations.
The original Beremin model has been
recently modified by Bordet et al. (2005a,
2005b) in order to include the effect of plastic
strain on the nucleation of microcracks and the
deactivation of the latter if they are not imme-
diately propagated. At a material point located
within the PZ, the probability of cleavage frac-
ture is expressed as
P
cleav
(t) =
_
t
0
P
prop
dP
nucl
[31[
where t represents the loading time, while
P
prop
and cP
nucl
represent the probability of
crack propagation and the increment of prob-
ability to nucleate a microcrack. In Bordets
model, cP
nucl
is written as
dP
nucl
aNs
y0
(T. _ e
ep
)de
ep
= N
0
(1 P
nucl
)s
0
de
ep
[32[
where e
ep
is the equivalent plastic strain, N and
N
0
are respectively the remaining and the initial
number of cleavage initiation sites. Faleskog
et al. (2004) have similarly assumed that the
increment of the probability of microcrack
nucleation was linearly related to plastic strain.
However, these authors did not include the
effect of yield strength on microcrack nuclea-
tion. Equation [32] can be easily integrated.
Normalizing the yield strength by a stress o
y0
and the plastic strain by e
ep0
which are depen-
dent on the material, eqn [32] leads to
P
nucl
= 1 exp
s
0
s
y0
e
ep
e
ep0
_ _
[33[
which bears some similarity with eqn [30].
However, there exists a significant difference
between those two expressions since eqn [33]
does not include the effect of stress state which
is well known to affect the nucleation of micro-
cracks from particles (see Section 2.06.3.3).
As for the Beremin model, assuming that the
nucleated microcracks can be treated as
Griffith flaws, which are assumed to be distrib-
uted according to a power law, and invoking
the weakest link principle over the PZ volume,
the overall fracture probability can then be
expressed as
P
R
= 1 exp
_
PZ
_
t
0
P
prop
dN
nucl
dV
_ _
= 1 exp
s
*
w
s
*
u
_ _
m
_ _
[34[
where o
u
*
is a scaling parameter and o
w
*
is a
modified Weibull stress defined as
s
*
w
_ _
m
=
_
PZ
(s
1
>s
th
)
_
e
ep
0
s
0
s
y0
s
m
1
s
m
th
_ _
(1P
nucl
(t))
de
ep
e
ep0
_ _
dV
V
u
[35[
where V
u
is a reference volume as in eqn [19].
The similarities between the models of
Bordet and Faleskog for the nucleation of
microcracks have already been underlined.
There exists, however, a difference in the
expression for the probability to propagate
microcracks, since in Faleskogs model it was
implicitly assumed that these microcracks were
distributed according to an exponential law
similar to the expression introduced previously
(see eqns [21] and [24]). A modification of the
original Beremin model has been also proposed
by Bernauer et al. (1999) to account for clea-
vage fracture in the transition region of a
ferritic steel. These authors have emphasized
that in the transition region the number
Cleavage in Metals 699
of available microcracks is reduced by the num-
ber of particles around which a void has been
formed. They assumed that the number of
omitted carbides was proportional to the
total number of voids formed at second-phase
particles. The nucleation rate of cavities was
assumed to follow the law proposed by Chu
and Needleman (1980) (see Section 2.06.3.3).
Sto ckl et al. (2000) have applied the
Bernauers model to interpret results on warm
prestress effect.
In these modifications or enhancements of
the original Beremin model, plastic strain
appears as detrimental since the microcracks
are nucleated from carbide particles or non-
metallic inclusions due to plastic deformation.
On the contrary, a number of other observa-
tions have also shown that the effect of a
predeformation can be beneficial (see, e.g.,
Groom and Knott, 1975; Knott, 1966, 1967;
Beremin, 1981). This is why in the Beremin
model a strain correction factor was intro-
duced to account for the effect of plastic
strain. The probability to failure was written as
P
R
= 1 exp
_
PZ
s
m
1
exp(me
p1
,a) dV
s
m
u
V
u
_ _
[36[
where e
p1
is the plastic strain in the direction of
the highest principal stress and a is a constant
close to 2.
To conclude this part devoted to ferritic steels,
it appears that, in spite of its simplifications, the
Beremin model has largely contributed to a bet-
ter description of the brittle cleavage fracture in
ferritic steels. A full account of the application of
this model to predict the fracture toughness of a
number of ferritic steels is given in Chapter 7.05.
The main criticism to this model, which is the
absence of a threshold, is not really acceptable
since this threshold is present although it does
not explicitly appear in the expression giving the
probability to fracture. The original Beremin
model applied to the prediction of fracture
toughness tends to underestimate the variation
of fracture toughness with temperature, in parti-
cular in the transition regime. This has led a
number of authors to assume that the scaling
stress, o
u
, was an increasing function of
temperature.
2.06.2.4 Transgranular Cleavage of
Other Metals
In this section, cleavage fracture in welds
made of HSLA steels is reviewed because of
its technological importance. The emphasis is
laid on the influence of MA constituents which
may be formed in those welds. Then cleavage
fracture in other BCC metals, especially Mo
and Nb, is briefly considered. Finally, cleavage
fracture in two HCP metals (Zn and Mg) is
examined.
2.06.2.4.1 Welds in HSLA steels: Influence of
MA constituents
HSLA steels are now widely used for struc-
tural applications. These materials combine
excellent tensile strength and DBT properties.
However, this combination of high-strength
and high-fracture toughness usually deterio-
rates after welding thermal cycles. The
degradation of the fracture toughness of
HSLA steels after welding is attributed to the
formation of local brittle zones in the welded
joint (see, e.g., Davis and King, 1994).
Significant embrittlement can be encountered
in the coarse-grained heat-affected zone
(CGHAZ) and, in particular, in the intercriti-
cally reheated CGHAZ (ICCGHAZ) of
multipass welded joints (see, e.g., Toyoda,
1988; Kenney et al., 1997; Zhou and Liu, 1998).
An example of such embrittlement effects is
shown in Figure 12 which refers to a recent study
performed on the micromechanisms and model-
ing of cleavage fracture in simulated HAZ
microstructures obtained in an HSLA steel
(o
0
=430 MPa at room temperature) (Lambert-
Perlade et al., 2004). In this study, a Gleeble
simulator was used to apply thermal cycles
representative of those encountered during mul-
tipass welding. The maximum temperature of
the first cycle was T
p1
=1250
C. Cooling times
from 800 to 500
C (Dt
8/5
) were chosen to be
100 s (CGHAZ-100) and 500 s (CGHAZ-500)
corresponding to a medium and high-input
welding energy, respectively. Intercritical heat-
ing of the CGHAZ-100 microstructure
(ICCGHAZ) at maximum temperature
(T
p2
=775
C. Tensile tests on
notched specimens showed that the cleavage
stress was reduced as compared to the ferrite
perlite microstructure (base metal), in particular
in the ICCGHAZ-100 and CGHAZ-500 micro-
structures. This clearly illustrates the deleterious
effect of the presence of MA constituents on
cleavage fracture in welds.
Fracture toughness tests were carried out on
specimens which were simulated with other
(Dt
8/5
) cooling conditions. Figure 14 shows
that the transition temperature measured at
K
J
=100 MPa m
1/2
is much higher in the
700 Failure of Metals
simulated microstructures than in the base
metal since shifts as large as 150
C are mea-
sured. Cleavage crack initiation observed on a
fracture toughness specimen tested at 20
C is
shown in Figure 15: cleavage was initiated from
an MA constituent (arrow), revealed after che-
mical etching of the fracture surface.
An attempt was made to model the observed
variations of the fracture toughness with tem-
perature shown in Figure 14 using the Beremin
model. The results are reported on the same
graphs while the values of the Beremin parameter,
m and o
u
, are given in Table 4. These values were
fitted by using elasticplastic finite element calcu-
lations on notched specimens tested in the brittle
temperature range, followed by post-processing
calculation of both the Weibull stress, o
w
, and the
probability to fracture, P
R
. Figure 14 shows a
good agreement with the experimental results
for the simulated HAZmicrostructures. Asimilar
conclusion was reached by Tagawa et al. (1993).
In Figure 14, we have also included the predic-
tions obtained from the master curve approach
(Wallin, 1991a, 1991b, 1993; ASTM E 1921-02,
2002; for further details, see Lambert-Perlade
et al., 2004). Tagawa et al. (1993) also used
the Beremin model for low carbon steel HAZ
microstructures. They found smaller values for
m (1018) instead of 2027. From a theoretical
point of view, the values of the parameter m
should be related to the defect size power-law
distribution (p(a)) by eqn [20]. The values of the
b exponent in eqn [20] were determined experi-
mentally. Using these values of b and the
relationship m=2b 2 leads to m~5 which is
much smaller than the values (m=2027) used
to draw the curves shown in Figure 14. They are
even smaller than those reported by Tagawa
et al. (1993). This probably arises from the fact
that cleavage fracture occurs in several steps
which theoretically implies the use of a multiple
barrier model as stated previously.
The multiple barrier model was also applied
to the results of fracture toughness measure-
ments obtained in various simulated HAZ
microstructures. The details are given elsewhere
(Lambert-Perlade et al., 2004). The effect of
temperature on fracture toughness can be
described by dividing the problem into three
temperature ranges:
0
20
40
60
80
200 150 100 50
CGHAZ-100 s
ICCGHAZ-100 s
CGHAZ-500 s
Temperature (C)
Base metal
S
t
r
a
i
n
t
o
f
a
i
l
u
r
e
0 50
Figure 12 Variation of the strain to failure measured on notched specimens as a function of test temperature in
a low-alloy steel with four microstructures (Lambert-Perlade, 2001). Reproduced with permission from
Lambert-Perlade, A., Gourgues, A. F., Besson, J., Sturel, T., and Pineau, A. 2004. Mechanisms and
modeling of cleavage fracture in simulated heat-affected zone microstructure of a high-strength low alloy
steel Metall. Mater. Trans. A 35, 10391053.
3
2
1
Figure 13 SEM micrograph of the microstructure
of an HAZ in a weld. (1) Residual austenite located
at lath grain boundaries; (2) MA mixed constituent
at a former austenite grain boundary; and (3) bainitic
packet boundary (Lambert-Perlade, 2001).
Cleavage in Metals 701
1. At very low temperature, the critical step
in cleavage fracture was assumed to correspond
to the nucleation of cleavage microcracks from
MA particles.
2. At somewhat higher temperatures, micro-
cracks initiate at particles and stop at the
particle/matrix interface. The critical step is
then the propagation of these particle-sized
microcracks.
300
(a) (b)
(d) (c)
250
200
150
100
50
0
200 150 100
Temperature (C)
K
J
c
(
M
P
a
m
1
/
2
)
Experiments
Pr = 10%
Pr = 90%
50 0 50
300
250
200
150
100
50
0
200 150 100
Temperature (C)
K
J
c
(
M
P
a
m
1
/
2
)
50 0 50
Experiments
Pr = 10%
Pr = 90%
300
250
200
150
100
50
0
200 150 100
Temperature (C)
K
J
c
(
M
P
a
m
1
/
2
)
50 0 50
Experiments
Pr = 10%
Pr = 90%
300
250
200
150
100
50
0
200 150 100
Temperature (C)
K
J
c
(
M
P
a
m
1
/
2
)
Experiments
Pr = 10%
Pr = 90%
50 0 50
Figure 14 Brittle-to-ductile toughness transition curves in a low-alloy steel and in simulated HAZ
microstructures. a, Base metal; b, CGHAZ-25; c, ICCGHAZ-25; d, CGHAZ-120 microstructures. Solid lines
(respectively, dotted lines) show fracture probabilities of 10% and 90% given by the Beremin model and the
master curve approach. Numerical values used in the Beremin model are given in Table 4 (Lambert-Perlade,
2001; Lambert-Perlade et al., 2004). Reproduced with permission from Lambert-Perlade, A., Gourgues, A. F.,
Besson, J., Sturel, T., and Pineau, A. 2004. Mechanisms and modeling of cleavage fracture in simulated heat-
affected zone microstructure of a high-strength low alloy steel. Metall. Mater. Trans. A 35, 10391053.
Fatigue precrack
First cleavage facet
10 m
Figure 15 Cleavage crack initiation after
interrupted test of a fracture mechanics specimen.
ICCGHAZ-25 microstructure. The cleavage crack
is initiated from an MA constituent (indicated by
an arrow) (Lambert-Perlade et al., 2004).
Reproduced with permission from Lambert-
Perlade, A., Gourgues, A. F., Besson, J., Sturel, T.,
and Pineau, A. 2004. Mechanisms and modeling of
cleavage fracture in simulated heat-affected zone
microstructure of a high-strength low alloy steel.
Metall. Mater. Trans. A 35, 10391053.
Table 4 HSLA steel. Parameters of the Beremin
model (unit volume V
u
=(100 mm)
3
)
Microstructure Beremin
model
Experimental values
of T*
100
(
C)
o
u
(MPa) m
Base metal 2158 27 140
CGHAZ-25 2670 20 55
ICCGHAZ-25 2351 20 20
CGHAZ-120 2085 20 10
The parameter T
*
100
provides the value of temperature for
which K
Jc
=100 MPa m
1/2
.
702 Failure of Metals
3. At higher temperatures, microcracks are
arrested at high-angle bainitic packet bound-
aries. The critical step is then the propagation
of these arrested grain-sized microcracks.
The input parameters of the multiple barrier
model are, therefore, as follows:
1. The fracture probability p(c) of an MA
particle of size c. In the absence of statistical
measurements it was assumed that this initia-
tion process occurred for a single value of the
critical stress (see eqn [15]). The value of
o
d
=2100 MPa was assumed using the inclusion
theory (Eshelby, 1957) (see Table 3).
2. The distribution functions f
c
(C) and f
g
(D)
for MA particles and bainitic packet size were
experimentally determined and represented by
lognormal functions.
3. The critical size for cracked MA particles
and cracked bainitic packets were calculated
using eqn [16]. For the sake of simplicity, it
was assumed that the values of the local frac-
ture toughness, K
Ia
c/f
and K
Ia
f/f
, were independent
of temperature, as indicated previously. This
means that, as in the Beremin model, the tem-
perature dependence of the fracture toughness
arises mainly from the variation of the yield
strength with temperature.
The numerical values of these input para-
meters are given in Table 3. The results
showing the application of the multiple barrier
model to one specific condition (ICCGHAZ-25)
are reported in Figure 16. A good agreement is
observed between the theory (solid lines) and the
experiments. In particular the model is able to
reproduce the dispersion, which is not trivial
since the calculated scatter derives directly from
the experimental size distribution of second-
phase particles and bainitic packets. In these
tests, acoustic emission was also used to detect
the number of events occurring before fracture.
Interrupted tests showed that one event corre-
sponded to the nucleation and crack growth of a
microcrack which was arrested at high-angle
grain boundaries (Lambert-Perlade et al.,
2004). In Figure 16, it is observed that the lowest
value of fracture toughness (open symbols) cor-
responding to the first detection of microcrack
events detected by acoustic emission occurs for
stress intensity factors equal to about
3040 MPa m
1/2
. These values agree with the
calculated probability for a cleavage microcrack
to propagate across the particle/matrix bound-
ary which is shown by dotted lines. These lines
were drawn using the values of K
Ia
c/f
and K
Ia
f/f
given in Table 3 and the results of finite element
calculations, assuming that the fracture tough-
ness specimens were tested under plane strain
conditions. Figure 16 successfully demonstrates
that the multiple barrier model is able to predict
not only the evolution of the fracture toughness
with temperature but also the value of the criti-
cal stress intensity factor associated to the
development of temporarily stable grain sized
microcracks. However, the difficulty with the
multiple model is that it requires the knowledge
of many microstructural parameters.
The main advantage of this type of model is
that it captures the fracture toughness, in the
upper part of the DBT curve, is related to the
propagation of microcracks which are tempora-
rily arrested at grain boundaries in ferritic steels
or at packet boundaries in bainitic steels. Recent
studies have shown that in bainitic steels the
effective packet boundaries are those with a
high-angle misorientation (see, e.g., Bouyne
et al., 1998; Gourgues et al., 2000; Lambert-
Temperature (C)
T
o
u
g
h
n
e
s
s
(
M
P
a
.
m
1
/
2
)
200
ICCGHAZ-25
150
100
100 50 0 50 150 200
50
0
Figure 16 Results of the multiple barrier model. Open circles denote microcrack events detected by acoustic
emission; solid circles correspond to final fracture. Solid (respectively, dotted) lines represent 10%, 50%, and
90% probabilities for the specimen to fracture (respectively, for a cleavage microcrack to propagate across a
particle/matrix boundary) as given by the multiple barrier model (Lambert-Perlade et al., 2004). Reproduced
with permission from Lambert-Perlade, A., Gourgues, A. F., Besson, J., Sturel, T., and Pineau, A. 2004.
Mechanisms and modeling of cleavage fracture in simulated heat-affected zone microstructure of a high-
strength low alloy steel. Metall. Mater. Trans. A 35, 10391053.
Cleavage in Metals 703
Perlade et al., 2004). This suggests that the frac-
ture toughness of these materials can be
improved by the development of such favorable
boundaries. This links with a new research field
sometimes called interface engineering or grain
boundary engineering. Fundamental studies in
this field must therefore be strongly encouraged.
2.06.2.4.2 Cleavage fracture in other
BCC metals
While the plastic deformation and brittle clea-
vage fracture behavior of iron and ferritic steels
has, for technological reasons, received consider-
able attention, other BCC metals have been
relatively neglected. This is partly due to the
expected similarity in behavior with a iron, and
partly due to the small use in key engineering
applications of the high melting refractory metals
(V, Nb, Ta, Cr, Mo, W) in industry. However,
the increased importance of these metals, in par-
ticular in gas turbines and nuclear energy sectors,
pushes for more detailed investigation of their
mechanical properties. Commercially available
group V-A refractory metals (V, Nb, Ta) are
considerably more ductile and have considerably
lower transition temperature than commercially
available group VI-A refractory metals (Cr, Mo,
W). This mainly results from that the solubility
of impurities in the V-A metals is much higher
than in the VI-A metals.
Molybdenum, which is representative of the
metals of group VI-A, has been investigated
(see, e.g., Briggs and Campbell, 1972; Koval
et al., 1997). Niobium, which is representative
of the metals in group V-A, has been investigated
in much more detail (Briggs and Campbell, 1972;
Samant and Lewandowski, 1997a, 1997b; Pahdi
and Lewandowski, 2004). Lewandowski and his
co-workers have studied pure Nb and NbZr
solid solutions. These authors have investigated
the effect of grain size (,60 and 165 mm). The
cleavage stress, o
f
, was measured using either
blunt notched specimens or fatigue precracked
fracture toughness specimens. In blunt notched
specimens, the Griffith and Owen solution was
used to calculate the cleavage stress (Griffiths
and Owen, 1971). In fracture toughness speci-
mens, the Ritchie, Knott, and Rice (RKR)
model (Ritchie et al., 1973) was used in conjunc-
tion with either the HRR field (Hutchinson,
1968; Rice and Rosengren, 1968; see Section
2.06.3.7.1) or the Tracey solution (Tracey,
1976). The cleavage stress was found to be a
decreasing function of grain size. The identified
value for o
f
appeared to be larger for fracture
toughness specimens than when identified based
on blunt notched specimens (typically 1700MPa
compared to 1400/1500 MPa in 60mm grain size
Nb). This suggests that the sampling volumes
which are quite different in these two geometries
may also affect the values of o
f
. Additional tests
and analyses are required to show if this size
effect can be interpreted by using the Beremin
theory.
2.06.2.4.3 Cleavage fracture in HCP metals
When compared to crystal systems like BCC,
HCP metals exhibit a wider variety of deforma-
tion modes, including slip and twinning systems.
Historically, HCP metals have been categorized in
terms of c/a ratio. For metals with c,a<
3
_
(e.g.,
beryllium, titanium, zirconium, and magnesium),
the {1012} twinning mode is activated by com-
pression along the c-axis and <ca> slip is
active. All these metals have a preferential clea-
vage plane which is the basal plane {0002}.
However, in metals like titanium and zirconium,
pure cleavage fracture along the {0002} plane has
not been observed, except under stress corrosion
conditions which are outside the scope of this (see,
e.g., for Zr alloys: Kubo et al., 1985; Schuster and
Lemaignan, 1989a, 1989b; Cox, 1990). This is
why in this section only cleavage fracture of zinc
and magnesium are considered.
(i) Cleavage fracture of zinc
Zinc shows brittle cleavage fracture along the
basal plane and along the prismatic planes. An
example of the fracture surface of a pure poly-
crystalline specimen broken at 196
C is shown
in Figure 17. The importance of prismatic clea-
vage in zinc and that of the accommodation
required at a grain boundary as a crack propa-
gates from grain to grain has been underlined
recently (Hughes et al., 2005). Cleavage on basal
plane has been studied more thoroughly, in par-
ticular on zinc single crystals (Gilman, 1958;
Deruyttere and Greenough, 1956). More
Figure 17 SEM micrograph of a polycrystalline
zinc specimen fractured at 196
C.
704 Failure of Metals
recently, cleavage fracture on basal plane has
also been studied in detail on hot dip galvanized
steel sheets (Parisot et al., 2004a, 2004b). In stu-
dies devoted to the analysis of single crystals, it
was shown that the critical cleavage stress mea-
sured by the stress normal to the {0002} cleavage
plane is largely reduced when the amount of
plastic strain in the basal plane,
bas
, is increased
(see Figure 18). Similar observations have been
made on crack propagation behavior in strongly
textured zinc sheets (Lemant and Pineau, 1981).
The cleavage criterion proposed by Gilman
(1958) involved the product o
n
i
bas
where
i
bas
is
the amount of plastic slip on the basal slip sys-
tem, i. The reasonfor the decrease of the cleavage
stress when basal plastic strain is increased is
likely related to the existence of local stress con-
centrations associated with basal dislocation
pile-ups that can locally reach the theoretical
cleavage stress (see Section 2.06.2.2). A similar
conclusion was reached in the study devoted to
zinc coatings on hot-dip galvanized steel sheets,
except that thresholds for the cleavage stress and
the basal plastic strain were introduced, that is,
s
f
= s
th
k
g
0
g
bas
ep
[37[
where o
th
, k, and
0
are material parameters. In
eqn [37] the theoretical cleavage stress is, in the
absence of basal slip, equal to o
th
k/
0
.
A measure of the total basal slip activity of the
three basal slip systems was defined as
g
bas
ep
= [g
1
[
bas
[g
2
[
bas
[g
3
[
bas
[38[
Equation [37] introduces an asymptotic value
o
th
for zinc crystals formed at the free surface of
galvanized steel sheets and undergoing a very
large amount of basal slip, which is not the case
in the original Gilmans criterion. This may be
related to the purity of the materials used in
those studies.
(ii) Cleavage fracture of magnesium
A major problem facing HCP metals, such
as magnesium and zinc, is their limited low-
temperature ductility. The dominant slip
mode in all HCP metals involves the Burgers
vector a =1/3<1120>whether the primary
slip plane is basal (e.g., cadmium, zinc, or mag-
nesium) or prismatic (e.g., titanium or
zirconium). Even if both slip planes operate,
there is still no way to accommodate strains
along the c-axis. Deformation twinning can
help alleviating this problem but it is often
insufficient to provide a large ductility.
This problem is encountered not only in zinc
but also in high-purity magnesium. Magnesium
also cleaves easily along the basal plane. The
cleavage fracture stress of high-purity magnesium
is independent of temperature, but highly depen-
dent upon grain size. The addition of lithium to
magnesium alloy has an interesting effect on low-
temperature fracture behavior in decreasing the
DBTT. Several reasons have been given to
explain this effect which is known for quite a
long time (Hauser et al., 1956). Many reports
have been published regarding the beneficial
effect of lithium on the ductility of magnesium
(see, e.g., Raynor, 1960; Saito et al., 1997). In
many of these studies, the lithium additions
resulted in a substantial volume fraction of the
soft Li-rich BCC b-phase which is probably one
part of the explanation for the better ductility.
The increase in ductility has also been observed
even in MgLi a-solid-solution alloys. Another
reason for the ductility improvement might be
related to the reduction of the stress for prismatic
slip relative to that for basal slip. More recently,
Agnew et al. (2001, 2002) have shown the
45
40
35
30
25
20
15
10
5
0
0 10 20
(%) at fracture
n
(
M
P
a
)
a
t
f
r
a
c
t
u
r
e
= 89
= 82
= 75
= 60
= 45
= 30
=
30
Figure 18 Fracture in zinc. Normal stress to the basal
plane at fracture as a functionof basal glide for different
misorientations
between the c-axis and the tensile
direction (Deruyttere and Greenough, 1956; Gilman,
1958). Reproduced with permission of Parisot, R.,
Forest, S., Pineau, A., Nguyen, F., Demonet, X., and
Mataigne, J. M. 2004b. Deformation and damage
mechanisms of zinc coatings on hot-dip galvanized
steel. Part II: Damage modes. Metall. Mater. Trans.
A 35, 813823.
Cleavage in Metals 705
existence of 1/3<1123>{1122} pyramidal slip in
MgLi solid-solution alloys, while pure magne-
sium exhibits only basal and pyramidal slip.
Nevertheless, the structure of<ca>
dislocations, in particular their dissociation
and decomposition, remains an open issue. Very
much remains thus to be done before the theories
of cleavage fracture in HCP metals reach a devel-
opment similar to those used for BCC metals.
2.06.2.4.4 Irradiation-induced embrittlement
in ferritic steels
Althoughit is out of the scope toreviewindetail
the micromechanisms of embrittlement produced
by irradiation, it is worth mentioning a number of
relevant mechanisms causing this mode of embrit-
tlement (for a review, see for instance Gurovich
et al., 1997; Odette et al., 2003).
RPVs of commercial nuclear power plants are
subjected to embrittlement due to the exposure to
high-energy neutrons from the core. The current
methods used to determine the effect of the degra-
dation by irradiation on the mechanical behavior
of RPVs rely on tensile tests and impact Charpy
tests. For a given fluence (f (#/cm
2
)) of neutrons
(energy>1MeV) and irradiation temperature,
irradiation-induced embrittlement is strongly
dependent on material chemical composition,
especially on Cu, Ni, and P contents (Haggag,
1993). The Cu content plays an important role in
the hardening-induced embrittlement due to the
irradiation-induced precipitation of Cu-rich
nanoparticles (see, e.g., Buswell et al., 1995;
Auger et al., 1995; Akamatsu et al., 1995). Other
fine-scale microstructural modifications influence
the macroscopic behavior of the irradiated mate-
rial. Two types of influence can be distinguished
(Nikolaev et al., 2002; Wagenhofer et al., 2001):
(1) modification of the plastic properties and
(2) embrittlement. One of the most well-known
and well-described embrittlement effects is asso-
ciated with phosphorus segregation at grain
boundaries (Miller et al., 1995; Faulkner et al.,
1996). This embrittlement effect is considered in
the following section devoted to intergranular
fracture. Hardening mechanisms include matrix
and precipitation hardening. Matrix hardening is
due to irradiation-produced point defects and
dislocation loops. The deformation is then
concentrated along channels, producing local
stress concentrations which facilitate the initiation
of cleavage fracture, as described previously.
Precipitation hardening, as already stated, is asso-
ciated with the irradiation-enhanced formation of
Cu-rich precipitates. These two hardening
mechanisms cause anincrease of the yieldstrength
and, usually, a decrease of the work-hardening
exponent.
The effect of irradiation embrittlement mea-
sured with Charpy V-notch tests is depicted
schematically in Figure 19. A shift of the
DBTT, DT, is observed. Similar effects on frac-
ture toughness, K
Ic
, are observed. In many cases,
the upper shelf energy (USE) determined from
Charpy specimens is lowered, as shown schema-
tically in Figure 19. The yield strength measured
in RPV steels with a low to medium content in
impurities (Cu, P) increases monotonically with
the fluence, c, while the work-hardening cap-
ability measured by (UTSo
0
) tends to
decrease (Tanguy et al., 2006; EDF, 2003). The
variation in yield strength can be represented by
the following empirical expression:
Ds
0
= 32.46f
0.51
[39[
where Do
0
is expressed in MPa and c in #/cm
2
.
The shift of the transition temperature, DT,
can be qualitatively explained using the theories
for cleavage fracture which have been presented
Temperature
C
h
a
r
p
y
e
n
e
r
g
y
;
f
r
a
c
t
u
r
e
t
o
u
g
h
n
e
s
s
T
Irradiated
Non-irradiated
(USE)
Figure 19 Schematic effect of irradiation effect on the DBT curves in ferritic steels.
706 Failure of Metals
in the previous section. More quantitative
details can be found elsewhere (Al Mundheri
et al., 1989; Tanguy et al., 2006). Assuming that
the cleavage stress is not affected by irradiation,
the increase in yield strength o
0
provides a sim-
ple explanation for the shift of the DBTT. The
Beremin theory (Section 2.06.3.3) applied to the
crack-tip region predicts that, when small-scale
yielding (SSY) conditions are fulfilled, the
probability to failure, P
R
, can be written as
P
R
= 1 exp
K
4
Ic
Bs
m4
0
Cm(n)
V
u
s
m
u
_ _
[40[
where B is the specimen thickness and Cm(n) is a
numerical factor which is an increasing function
of the work-hardening exponent o=ke
n
of the
material; for further details, see, for example,
Chapter 7.05. An increase of the yield strength
due to irradiation will thus produce a reduction
of the fracture toughness, K
Ic
, for a given test
temperature and a given specimen thickness.
Very recently, a similar approach has been used
to predict the Charpy V temperature shift in
A508PVR steel by Tanguy et al. (2006). These
authors have simulated Charpy V specimens by
the finite element method using a material model
integrating a description of viscoplasticity, duc-
tile damage, and cleavage fracture. They also
assumed that irradiation affects only the yield
strength and/or the work-hardening coefficient
but not the cleavage stress, o
u
. Using empirical
correlation relating the increases in yield
strength Do
0
(in MPa) to the irradiation fluence
c given by [39], Tanguy et al. (2006) showed that
the increase in the DBTT with c can be reason-
ably well predicted with the Beremin theory. In
particular, the low values of the Charpy energy
are properly captured. These authors also
showed that the shift in fracture toughness mea-
surements, K
Ic
, could be predicted using the
Beremin theory. Their results are provided in
Table 5, which reports the values of the DBTT
corresponding to a fracture toughness equal to
100 MPa m
1/2
, the value used in many
approaches based on the ASTM E 1921 Master
Curve Standard for measuring the fracture
toughness in the cleavage transition (ASTM E
1921-02, 2002; Markle et al., 1998). In Table 5
we have also included the values of the tempera-
ture shift, DT
56J
, for a Charpy energy of 56J,
which is a typical value used in many standards.
These values of DT
56J
were also predicted using
sophisticated numerical finite element calcula-
tions in combination with the Beremin theory.
These variations in DT can also be deter-
mined using simpler analytical calculations
provided that the temperature dependence of
the yield strength is known. Several expressions
have been proposed for this dependence.
A polynominal function given by Rathbun
et al. (2006) can be used:
s
0
(T) = 0.0085T
2
0.4402T 481.51 [41[
where o
0
is expressed in MPa while T is
expressed in
C. It is assumed that this function,
valid for non-irradiated materials, can also be
applied to the irradiated material. It is also
assumed that the slight variation in the work-
hardening capability of the material with irra-
diation does not influence the value of the
coefficient Cm appearing in the theoretical eqn
[40]. It might be useful to remember that this
equation is strictly valid for a stationary crack
under SSY conditions (Chapter 7.05). This the-
oretical expression indicates that for a given
probability to failure, the product K
Ic
o
0
((m/4)1)
should remain constant and is independent of
the irradiation conditions. Differentiating this
expression leads to
DK
Ic
K
Ic
=
m
4
1
_ _
Ds
0
s
y
((m,4) 1)
[42[
This expression relates the variation in fracture
toughness, DK
Ic
, with the increase in yield
strength, Do
0
, produced by neutron irradiation.
The increase of the flow strength Do
0
is given by
eqn [39]. The variation of yield-strength equa-
tion [39] can also be differentiated to predict the
temperature shift necessary to obtain, after given
irradiation conditions, the same value of the
fracture toughness, for instance, 100 MPa m
1/2
.
One easily obtains that the variation in yield
strength, Do
0
, is simply related to the tempera-
ture shift, DT, by the following expression:
Ds
0
(MPa) = [0.017T
0
(
C) 0.4402[DT(
C) [43[
Assuming that, for instance, T
0
=60
C, eqn
[41] predicts that, in the non-irradiated condi-
tion, o
0
(60
C)
DT
K
Ic,100
numerical
(
C)
DT
K
Ic,100
analytical
(
C)
1.90 45 30 49 31
7.07 88 56 73 60.3
20.10 150 95 104 103
Cleavage in Metals 707
value for A508 steel. The above expression and
eqn [39] simply leads to
Ds
0
(MPa) = 1.46DT(
C) = 32.46f
0.51
[44[
where c is expressed in 10
19
#/cm
2
. This expres-
sion can thus be used to assess the values of the
DBTT shift, DT, corresponding to a reference
fracture toughness equal to 100MPam
1/2
. The
results are reportedinTable 5 where a comparison
with those obtained by numerical finite element
calculations (Tanguy et al., 2006) can be made.
These results show that the simplified analytical
approach leads to values for the temperature shift
close to those inferred from sophisticated numer-
ical computations. Abetter agreement is observed
for large values of the fluence. This is likely
because the basic assumption of SSY conditions
behind the theoretical eqn [40] is much better
satisfied for large increases in the yield strength,
Do
0
, due to very large values for the fluence. The
values of the DT shift predicted from the analyti-
cal approach are closer to those numerically
calculated for the shift in Charpy tests. No simple
explanation can be given to this observation
which is likely purely fortuitous.
This brief overview of the irradiation-induced
embrittlement suggests that the main source of
embrittlement is related to the irradiation-hard-
ening effect. As a matter of fact, the problem is
likely far more complex. In particular, as already
stated, a modification in the failure mode from
transgranular cleavage to intergranular has been
reported (Miller et al., 1995; Faulkner et al.,
1996; Gurovich et al., 1999, 2000). Moreover,
the irradiation-hardening effect cannot simply
explain the decrease of the USE depicted in
Figure 19. Many theories for ductile fracture
depict that an increase in yield strength should
produce an elevation of the USE (see Section
2.06.3.7.2), when other factors are maintained.
The observed decrease in the USE may be due
to the reduction in the work hardenability of the
material which is evidenced at high fluences. A
reduction in the work-hardening exponent leads
to strain localization which is detrimental to the
ductility of the material (see Section 2.06.3.2) and
to the fracture toughness (see Section 2.06.3.7.2).
Irradiation-induced segregation of impurities like
P at the matrix/precipitate interface may also
contribute to the reduction in the ductility of
the material. Such segregation effects seem to
have been observed by Gurovich et al. (2000).
These difficulties related to impurities contri-
bute to the fact that the predictions of
irradiation-induced embrittlement remain
largely empirical. However, in this section, an
attempt has been made to show that, at least in
relatively clean steels, in which the segregation
phenomena are limited, the shift in the DBTT
due to irradiation effects can reasonably be well
predicted using the theoretical local approach
to cleavage fracture. Clearly, this is another
research area which should deserve more
research effort.
2.06.2.5 Intergranular Brittle Fracture
in Ferritic Steels
As stated previously theoretical calculations
suggest that intergranular fracture should be
observed preferentially in many polycrystalline
metals instead of transgranular cleavage fracture
(see Table 2). However in ferritic steels, brittle
fracture occurs at low temperature by transgra-
nular cleavage. This is usually attributed to the
reinforcement effect of a number of elements
segregated along the grain boundaries, in parti-
cular carbon. Conversely, the segregation of
other impurities, for example, phosphorus,
along the grain boundaries can change the brittle
fracture mode from cleavage to intergranular.
This is the situation observed for instance in
low-alloy steels, such as A508 Cl3 steel used for
the fabrication of pressurized water reactors
(PWRs). In these thick components the presence
of small areas of low toughness, referred to as
ghost lines, may be an important source of
scatter in fracture toughness values (see, e.g.,
Tavassoli et al., 1983, 1989; Kantidis et al.,
1994). These ghost lines can initiate intergranu-
lar brittle fracture due to temper-embrittlement
effect. These lines were identified as segregated
zones containing increased amounts of impuri-
ties (P, S, etc.) and alloying elements (C, Mn, Ni,
Mo, etc.), as compared to the base material.
Intergranular brittle fracture is therefore of
great practical importance.
Very few detailed studies have been made to
determine quantitatively the variation of the
critical intergranular fracture stress o
CI
with
test parameters using procedures similar to
those used for the measurement of the cleavage
stress. Most often, the effect of temper embrit-
tlement has been investigated by determining
the shift in the DBTT, using Charpy V-notched
specimens. However, the interesting work by
Kameda and Mc Mahon (1980) should be men-
tioned. These authors showed that the critical
intergranular fracture stress was directly related
to the amount of impurity (Sb) segregated on
grain boundaries. Similarly in an RPV steel
based on A508 steel composition, it was
shown that o
CI
was decreasing linearly with
the amount of phosphorus segregated along
grain boundaries (Naudin, 1999; Naudin
et al., 1999). In this material the critical stress
(,2300 MPa for cleavage fracture) was reduced
by almost 30% when the phosphorus
708 Failure of Metals
monolayer grain-boundary coverage reached
approximately 40%. In another recent study
devoted to the statistical aspect of intergranular
fracture in a low-alloy steel, it was shown that
after a heat treatment leading to temper embrit-
tlement, the intergranular fracture stress was of
the order of 1400 MPa while before applying
this heat treatment the cleavage stress was
close to 1560 MPa (Wu and Knott, 2004).
These authors showed also that o
CI
was inde-
pendent of temperature within a first
approximation and was distributed according
either a normal or a Weibull law.
More detailed studies have been devoted to
the effect of intergranular fracture on the frac-
ture toughness of A508 steel due to the
importance of this mode of failure on the frac-
ture assessment of RPV components (Yahya
et al., 1998; Kantidis et al., 1994; Raoul et al.,
1999). In these studies, intergranular fracture
was favored by applying a step-cooling heat
treatment after tempering. It was shown that
the Weibull statistics (eqn [22]) was able to
describe the scatter in test results on notched
bars and fracture mechanics specimens. It was
also shown that the Beremin theory for clea-
vage fracture should be slightly modified to
account for the effect of test temperature,
since o
CI
was found to be an increasing function
of temperature. Kantidis et al. (1994) showed
that the variation of fracture toughness with
temperature could be well represented using
the original Beremin theory, provided that the
Weibull stress included a temperature depen-
dence, that is,
s
m
WW
=
_
PZ
s
m
1
[1 l(T T
0
)[
m
dV
V
0
[45[
where o
WW
is the modified Weibull stress, l >0
is a material parameter, and T
0
a reference
temperature. Equation [45] is very similar to
the original definition of the Weibull stress
given by eqn [27]. The expression of the mod-
ified Weibull stress indicates that when the
temperature is lower than T
0
, intergranular
fracture occurs at lower stresses, as compared
to the Weibull stress calculated without any
temperature correction. In these studies
devoted to A508 steel, it was also shown that
the value of the shape factor, m, was much
lower (,10) for intergranular fracture than for
cleavage where typically m,20. The reasons
for this difference in m values, which have
strong practical implications since the value of
m controls the scatter in the results, have not yet
been discussed in detail. Clearly brittle intergra-
nular fracture also requires further detailed
investigations.
2.06.3 DUCTILE FRACTURE IN METALS
2.06.3.1 Introduction: Two Classes
of Failure Mechanisms
Ductile fracture is the most common room-
temperature mechanism of failure in metals.
We will reserve the term ductile fracture for
the process of damage nucleation followed by a
phase of damage growth and coalescence driven
by plastic deformation. A good understanding
of ductile fracture relies thus first on proper
appraisal of the mechanisms and theory of
plasticity, that is, physics of dislocations, of
hardening and strain-hardening mechanisms,
crystal plasticity and plastic anisotropy con-
cepts. Note also that, in some high-strength
alloys, dirty metals, or metal matrix compo-
sites, the process of nucleation, growth, and
coalescence of voids can take place very rapidly
and lead to very low ductility, sometimes as
small as 1%. Nevertheless, in our terminology,
these materials fail by a ductile fracture
mechanism to contrast with a cleavage-type
mechanism.
Even if ductile fracture is defined as the result
of a damage process by plasticity-controlled
void nucleation, growth, and coalescence, it is
important, for practical purposes, to distin-
guish between two modes of ductile fracture.
In the first mode, damage occurs more or less
homogenously in homogenously deformed
regions up to the final fracture point (see
Figure 20). Only at the very end of the process,
when void coalescence takes place and a crack
starts propagating in the solids, the deforma-
tion can become highly heterogenous. In the
second mode, plastic localization occurs
before or early in the damaging process through
z mean
z mean
or
Figure 20 Schematic representation of the process
of nucleation, growth, and coalescence of voids
nucleated on second-phase particles inside an
idealized representative volume element of the
microstructure, and the relationship with the
macroscopic loading evolution.
Ductile Fracture in Metals 709
the development of localized necks or shear
bands, usually with no or only limited coupling
with the damage evolution. In that case, the
practical failure condition is the onset of plastic
localization. This is the common situation in
low stress triaxiality metal-forming applica-
tions in which the FLD concept has been
developed in order to provide guidelines for
avoiding plastic localization. Of course, the
regular damage process still takes place within
the localization band which involves large plas-
tic strains, and is ultimately responsible for the
fracture. The increasing stress constraint in the
band accelerates the damage process. In terms
of the macroscopic loading, the final fracture
follows rapidly after the onset of localization.
The details of the failure process within the
band and the prediction or measurement of
the fracture strain locus are of limited practical
interest. This division between plastic localiza-
tion-controlled failure and damage-controlled
failure motivates the organization of the section
in two parts. The first part consists in an intro-
duction to plastic localization. This part, made
of a single subsection, is kept relatively short
because it deviates from the main message of
this chapter which focuses on the fracture
micromechanisms in metals. The second part
is made of several subsections devoted to the
nucleation, growth, and coalescence of voids
and to the ductility and fracture resistance of
thick and thin metallic plates.
2.06.3.2 Plastic Localization Mechanisms
in Homogeneous Medium
Plastic deformation can be localized because
the material is not homogeneous at the level of
the microstructure. Such type of localizations,
when present, is usually a primary factor con-
trolling the ductility and fracture toughness of
metallic alloys. The most typical situation is a
microstructure or mesostructure that involves
hard and soft regions or phases. Examples of
such type of microstructure-induced localiza-
tions will be repetitively illustrated when
dealing with damage mechanisms.
The second type of plastic localizations takes
place in homogeneous materials under loading.
Advanced solid mechanics is necessary in order
to formulate a sound framework for addressing
plastic localization from general perspective.
Excellent reviews have been written on that
topic which goes outside the scope of this sec-
tion (e.g., several papers in Koistinen and
Wang, 1978; Rice, 1976; Hutchinson, 1979;
Semiatin and Jonas, 1984; Needleman and
Tvergaard, 1992; Perzyna, 1998; Forest and
Lorentz, 2004). We will start here from the
simple problem of necking under uniaxial ten-
sion and then move to the formulation of
plastic localization conditions in 2-D plane
stress.
2.06.3.2.1 Necking under uniaxial tension
Let us start by considering a long perfect bar
of a material whose plastic behavior is charac-
terized by the evolution of the flow stress as a
function of strain and strain rate. A generic
behavior is assumed for which the variation
of the strain-hardening rate as a function of
strain e is expressed by the function n(e)
defined as
n =
0lns
0lne
=
e
s
0s
0e
[46[
Similarly, the variation of the strain rate sensi-
tivity as a function of strain can be expressed by
the function m9(e) defined as
m9 =
0lns
0ln_ e
=
_ e
s
0s
0_ e
[47[
The specimen is loaded under uniaxial ten-
sion along direction z. Let us assume that this
specimen presents a infinitesimally small pertur-
bation of the cross-sectional area A(z) in the close
neighborhood of a point of coordinate z. By
imposing force equilibrium along the specimen
length and neglecting elastic strains, one can
demonstrate that at the beginning of straining, a
region of the specimen that would locally present
a smaller cross-sectional area (with for instance
dA/dz <0 at a particular point) will undergo a
lower rate of cross-sectional area reduction than
the other regions (i.e., d/dz(dA/dt) >0 at that
particular point). This means that the tensile
deformation of the specimen is then stable:
deformation tends to distribute uniformly
along the specimen. Stability toward section
reduction is allowed by the strain-hardening
capacity of the material. At larger strains, a
situation is attained where the strain-hardening
capacity is not sufficient anymore to compen-
sate for the section reduction and the
perturbation then amplifies while the rest of
the specimen starts to unload elastically. The
condition for dA/dz <0 d/dz(dA/dt) >0 or
dA/dz >0 d/dz(dA/dt) <0 can be written,
after some algebra, in the form
e _
n(e)
1 m9(e)
[48[
In principle, beyond this point, plastic strain-
ing in tension becomes unstable: a region with
lower cross-sectional area undergoes a larger
rate of contraction than the rest of the speci-
men, which is the phenomenon of necking. As
shown in Figure 21, the role of the rate
710 Failure of Metals
sensitivity to help stabilizing plastic localization
is in fact much stronger than what is predicted
by eqn [48]. A good example is provided in
Figure 22 showing the engineering stress strain
curves of a fully b titanium alloy at room tem-
perature. Although the strain rate sensitivity
seems quite low, m9 ~0.03, the effect on stabi-
lizing the necking process is already quite
significant.
Conversely, the presence of material or geo-
metrical imperfections leads to a significant
drop of the resistance to plastic localization
with respect to what is predicted by eqn [48]
(e.g., Hosford and Caddell, 1993). It is worth
noting that accounting for imperfections in the
modeling of the necking condition leads to a
much better description of the effect of the rate
sensitivity. Considering an initial imperfection
j defined as a fractional reduction of the initial
cross-sectional area, the following expression
provides a good estimate of the necking strain
when m9 is close to 0 and when the imperfection
is small (Hutchinson and Neale, 1977):
e = n 1
2Z
n
_
_ _
[49[
while a good estimate when n is close to 0 and
when the imperfection is small (but not equal to 0)
is given by
e = m9 ln 1 (1 Z)
1,m9
_ _
[50[
These two expressions provide a fairly good
indication of the effect of both m9 and j on
the onset of necking. For instance, an imperfec-
tion of 1% in a material sample with n =0.2 and
m9 =0 will cause a 30% drop of the necking
strain. Unfortunately, it is difficult to provide
analytical expressions for the onset of necking
involving simultaneously the effect of the rate
sensitivity and the effect of the presence of
imperfection except in some asymptotic limits.
More general parameter analysis is provided in
Hutchinson and Neale (1977).
After the onset of localization, the necking
region develops with the height scaling with the
width of the specimen (or the diameter for a
cylindrical specimen). Up to that point the
shape of the section is unimportant. The shape
of the section, that is, rectangular section versus
cylindrical, and the aspect ratio of the section,
0.001
0.01
0.1
1
Total elongation (%)
Data collected by
Woodford (1969) :
- Fe base + Cr and Mo
- Ni base
- Mg base + Zr
- Pu base
- Pb Sn
- Ti base + Al and Sn
- Ti base + Al and V
- Zircaloy
S
t
r
a
i
n
r
a
t
e
s
e
n
s
i
t
i
v
i
t
y
,
m
1 10 100 1000 10
4
Trend lines for the
condition > n /(1m)
Figure 21 Variation of the total elongation as a
function of strain rate sensitivity (from data
collected by Woodford, 1969; see also Hutchinson
and Neale, 1977).
1200
1000
800
600
400
0 0.05 0.1 0.15 0.2 0.25
l /l
0
e
n
g
(
M
P
a
)
= 10
2
s
1
= 10
2
s
1
LCB Ti alloy low cost beta
First observable necking
Maximum load ( approximately equal to n/(1m))
= 1 s
1
.
.
.
Figure 22 Engineering stress strain curves for a full beta titanium alloy for different applied strain rates.
Courtesy of N. Cle ment, Universite Catholique de Louvain.
Ductile Fracture in Metals 711
that is, small versus large ratio of width over
thickness, affects the geometrical evolution of
the neck (Zhang et al., 1999), the evolving stress
state within the neck, the possibility for shear
banding to take place within the necking
region, and the evolution of damage inside the
neck. This aspect is discussed in Section
2.06.3.6.1.
Figure 23 shows the evolution of the ratio a of
the height of the neck divided by the minimum
thickness (for a plane strain tension specimen)
or divided by the diameter of the minimum cross
section (for cylindrical specimens) as a function
of the effective strain within the minimum sec-
tion of the neck e
en
, for various strain-hardening
exponents n. These results were obtained by
finite element simulations assuming the isotro-
pic J2 flow theory for the response of the
material (see Pardoen et al., 2004). The strain-
hardening exponent was defined here through a
Swift-type representation:
s
e
s
0
= (1 ke
ep
)
n
[51[
where o
e
is the effective stress, e
ep
is the effective
plastic strain, o
0
is the yield stress, and k is a
parameter that is usually much larger than 1
(here, it is equal to E/o
0
). The ratio a =h/t (see
Figure 23) is equal to 1 for perfectly plastic
materials (Onat and Prager, 1954; Hosford
and Atkins, 1990). A good empirical fit for
these results is given by
a
axisymmetric
= 1.105 1.422
n
e
1.725
en
[52[
a
plane strain tension
= 0.012 4.353
n
e
1.37
en
[53[
All a values are larger than 1 (the rigid-perfectly
plastic value), and decrease with straining.
The development of the neck is not affected by
the value of the ratio o
0
/E but the shape of the
necking zone is very much influenced by the
strain-hardening capacity. A smaller hardening
exponent leads to a more localized neck and
thus smaller a. The fact that the neck is sharper
when the strain-hardening exponent is smaller
has a direct effect on the evolution of the stress
triaxiality (see definition in Section 2.06.3.4.1)
inside the neck, as shown in Figure 24, which
raises more slowly with shallower necks. The
variation of a with straining is important and
cannot be neglected in the analysis of necking.
These types of numerical simulations of
necking (as initiated by Needleman (1972a),
using the finite element method, or Norris
et al. (1978), using the finite difference method)
are very useful for the determination of the true
stress strain response of the material after the
onset of necking, through an inverse modeling
procedure (see also Pardoen and Delannay,
1998a; Zhang et al., 1999). In the same vein,
Zhang et al. (2001) have carefully analyzed the
h t
(a)
n =0.2
n= 0.3 n = 0.4
n = 0.5
n = 0.1
n =0.01
0.5 0
0
1
2
3
4
1 1.5
en
(c)
Axisymmetric conditions n =0.2
n= 0.3
n =0.1
4
3
2
1
0
0 0.5 1 1.5
en
n = 0.4
n=0.5
h
/
e
f
0
=0.2%, W
0
=1,
0
/E=0.001
n= 0.05
n = 0.2
n= 0.1
1/3 =pure uniaxial tension
1.2
0.9
0.6
0.3
0
0 0.5
1
1.5
e
Figure 24 Variation of the stress triaxiality in the
center of the minimum cross-sectional area as a
function of the average effective strain in the
minimum section.
712 Failure of Metals
effect of the anisotropy of the material on the
evolution of the necking process.
Note finally about necking under uniaxial
tension conditions that:
1. a short specimen length delays the onset of
the localization process when compared to con-
dition [48] (Hutchinson, 1979);
2. other localizationmodes involving multiple
necks or surface instabilities are also predicted by
the theory (e.g., Hutchinson, 1979) but they
require very specific boundary conditions rarely
observed in practice; and
3. shear band formation is unlikely to occur
under uniaxial loading conditions with smooth
yield surfaces but is predicted with cornered
yield surfaces (Needleman and Rice, 1978) or
after some amount of necking (then the stress
state is not uniaxial anymore).
2.06.3.2.2 Plastic localization under biaxial
loading conditions
The localization of plasticity, that is, a transi-
tion from a uniform to a nonuniform mode of
deformation while the loading remains uniform,
also occurs under a multiaxial state of strain and
stress. As illustrated in Figure 25, one then fre-
quently observes so-called shear bands, that is,
bands oriented in the direction of maximum
shear. The width of a shear band varies depend-
ing on the material microstructure, for example,
on the grain size or on the dislocation cell size. In
general, the orientation of the band depends on
the stress state and the geometry of the speci-
men. Necking is said to be a geometric diffuse
mode of plastic localization. (Diffuse means
that there is a progressive transition between
the necking zone and the rest of specimen,
while geometric means that the geometry of
the neck is imposed only by the geometry of
the specimen and not by microstructural features
of the material.) When referring to shear bands,
one usually refers to bands showing a very sharp
transition between the localized band (or loca-
lized neck) and the rest of the material. As
suggested in Figure 25, it is common, in homo-
genous thin plates, to observe first the onset of
necking and later the emergence of shear bands
within the diffuse neck (e.g., the detailed experi-
mental study by Carlson and Bird (1987)). In
complex thin plate geometry, non-homogeneous
deformation conditions can prevent the appear-
ance of necking but cannot generally prevent the
occurrence of shear bands.
Plastic localization is an important limiting
factor in plate-forming operations, such as in
deep drawing. It is common to map the condi-
tions of occurrence of plastic localization using
so-called FLDs. The FLD is intensively used in
the metal-forming industry. A schematic exam-
ple of an FLD is presented in Figure 26. The
curve represents the locus of the in-plane prin-
cipal strains, e
1
and e
2
, corresponding to the
onset of a plastic instability (by construction,
the FLD is symmetrical with respect to the
mirror line e
1
=e
2
). Figure 26 shows that the
value of the biaxiality ratio , =e
2
/e
1
, signifi-
cantly affects the onset of plastic instability.
A forming process is safe if the state of defor-
mation in the plate never reaches the forming
limit during the loading history.
In the case e
2
<0 and e
1
>0 (the left-hand
part of the FLD), it can be shown that, assum-
ing the response of the material under uniaxial
stress is isotropic and can be described by a
Hollomon representation (o =Ke
n
) and that
the strain biaxiality ratio , =e
2
/e
1
keeps a con-
stant value during loading, the condition for the
formation of these bands is
Diffuse neck Localized shear band
Figure 25 Diffuse neck and localized band.
Major strain
1
Minor strain
2
C
om
pression
Slope 1
Slope 2
W
r
i
n
k
l
i
n
g
n
Slope 1/2
2n
Diffuse necking
S
h
e
a
r
f
r
a
c
t
u
r
e
S
h
e
a
r
b
a
n
d
s
S
h
e
a
r
b
a
n
d
s
S
h
e
a
r b
a
n
d
s
F
r
a
c
t
u
r
e
F
r
a
c
t
u
r
e
Figure 26 Failure locus, i.e., FLDs, for thin sheets
under biaxial loading conditions.
Ductile Fracture in Metals 713
e
1
e
2
= n [54[
while the orientation 0 of the band with respect
to the x
2
axis is given by
tan y =
e
2
e
1
_ _
1,2
[55[
In the case of tensile testing (e
2
=e
1
/2), a loca-
lized band of deformation would thus appear at
an angle 0 =54.7
when e
1
=2n. The localized
band will thus develop inside the diffuse neck
which started to form at e
1
=n.
In anisotropic materials characterized by a
Lankfordcoefficient l =e
width
/e
thickness
whenload-
ing in the long direction, the orientation 0 of the
band depends on the degree of anisotropy. In
uniaxial tension, the orientationof the bandwrites
tan y =
l
1 l
_
[56[
which generalizes eqn [55].
A condition for diffuse necking in isotropic
materials under biaxial straining condition has
also been worked out by Swift (1952):
e
1
=
2n(1 r r
2
)
(r 1)(2r
2
r 2)
[57[
The prediction of plastic localization for
e
2
>0 and e
1
>0 (right-hand side of the FLD)
is much more complex and requires either an
imperfection-type of analysis (see Marciniak
and Kuszinsky, 1967) or a bifurcation analysis
(e.g., Rudnicki and Rice, 1975). In both
approaches numerical analysis is necessary.
These analyses show specific features of the
plastic flow response such as vertices on the
yield surfaces. The presence of porosity can
also have serious effect on the onset of shear
banding (e.g., Rice, 1976; Needleman and Rice,
1978; Needleman and Tvergaard, 1992;
Hosford and Caddell, 1993). When the strain
path is highly nonradial, significant departures
from the localization strains predicted by the
FLDs are observed, as shown in Figure 27 for
an aluminum 2008-T6 strained first under equi-
biaxial tension and then under plane strain
tension.
An arresting example of nonradial loading
effects affecting plastic localization is provided
by systems made of thin metal layers deposited
on elastomers (see Lacour et al., 2003; Li and
Suo, 2005; Li et al., 2005). If the Youngs mod-
ulus of the elastomer, E, is noted and if a
Hollomon fit (o =Ke
n
) is used for the stress
strain curve, three specific mechanisms can be
predicted theoretically for plane strain
conditions:
1. If the elastomer is very compliant, that is,
E/K is small (e.g., E/K<0.2), then the metal
film forms a neck at small strain as it was a
freestanding film.
2. If the elastomer has an intermediate com-
pliance (e.g., E/K=1), then the metal film
forms multiple necks and deforms very much
beyond the bifurcation point.
3. If the elastomer is stiff, that is, E/K is large
(e.g., E/K>2), then the metal film deforms
uniformly to large strains.
As a matter of fact, the substrate stabilizes the
localization process as it does not want to
undergo large local elongation. These phenom-
ena have been observed experimentally.
Nowadays, advanced multiscale physics-
based constitutive models coupled to numer-
ical simulation tools allow predicting quite
accurately the localization locus of metallic
alloys, for radial or nonradial loadings, by
incorporating plastic anisotropy effects
phenomenologically or through crystal or
polycrystal plasticity theory, kinematic hard-
ening, phase transformation or second phases,
information about the dislocation cell struc-
ture evolution, etc. (e.g., Peirce et al., 1982;
Hiwatashi et al., 1998; Hill, 2001; Inal et al.,
2002; Knockaert et al., 2002; Yao and Cao,
2002; Chien et al., 2004; Wu et al., 2005; He
et al., 2005). The capacity of constitutive
models to properly predict localization is an
excellent way to assess their validity. These
models constitute very important tools for
accelerating the development and optimiza-
tion of new forming operations. The current
challenge with these models is to incorporate
internal lengths in computationally efficient
ways while keeping the physics right. At this
time, the lack of internal length gives rise to
strong mesh dependency effects when simulat-
ing the development of plastic localizations
(see, for a detailed discussion, Forest and
Lorentz, 2004; Niordson and Redanz, 2004).
No
prestrain
= 0.07
= 0.12
= 0.17
= 0.04
1
0.30
0.20
0.10
0.00
0.30 0.10 0.00 0.10 0.20
Figure 27 Example of nonradial loading effects on
FLDs for aluminum alloy 2008-T6 (based on Graf
and Hosford, 1993; see also Hosford and Duncan,
1999).
714 Failure of Metals
As a matter of fact, in metals, the width of
the localization band is set by the microstruc-
ture (typically the grain size). Localization
results thus from a complex competition
between material hardening versus material
softening, geometry and loading configuration
effects.
2.06.3.3 Void Nucleation
2.06.3.3.1 Macroscopic evidences
Void nucleation is usually not detected
from the overall mechanical response of
metals. In most metals, the initial second-
phase content is indeed (and fortunately) too
small to bring about significant amount of
initial porosity: the initial porosity in metals
ranges typically between 10
5
and 10
2
.
Furthermore, voids do not nucleate at the
same time on all particles. A smart method,
but not applicable to all metals, to determine
the average overall nucleation strain consists
of prestraining samples to various levels of
deformation, heat-treating them to restore
the strain-hardening capacity and loading
them again up to fracture. The idea is that
the fracture strain will be independent of the
level of prestraining as long as it does not
lead to void nucleation. The level of pre-
straining which affects the fracture strain
after heat treatment is an indication of the
nucleation strain. This method has been used
on aluminum (Le Roy et al., 1981) and on
copper (Pardoen and Delannay, 1998b).
Ultrasonic detection of the nucleation events
can be used to estimate the nucleation strain
(e.g., Montheillet and Moussy, 1986). Note
also that all the techniques that are pres-
ented in the next section for measuring
damage growth macroscopically can be used
to quantify an overall nucleation strain or
stress by extrapolating the data to zero
damage.
2.06.3.3.2 Microscopic observations
Void nucleation is usually associated to the
presence of second-phase particles and inclu-
sions, located either within the grains or along
the grain boundaries (see the classical papers
by Puttick, 1959; Argon and Im, 1975; Argon
et al., 1975; Argon, 1976; Goods and Brown,
1979; Fischer and Gurland, 1981; Beremin,
1981; Wilsdorf, 1983; Van Stone et al., 1985).
For illustration, the two micrographs of
Figure 28 have been obtained during in situ
tensile tests on 6XXX aluminum alloys.
Similar observations have already been
described in Section 2.06.2.3.3 when dealing
with Inco 718 alloy (see Alexandre et al.,
2005). The AlFeSi particles, located along the
grain boundaries, present an elongated platelet
shape. These particles break into several frag-
ments when aligned with respect to the main
loading axis while interface separation occurs
when their long axis is orthogonal to the main
loading direction (Lassance et al., 2006a). This
dependence of the void nucleation mode on the
orientation of the second phases with the load-
ing configuration has been studied in details in
steels involving long MnS inclusions (see
Montheillet and Moussy, 1986, for references).
Babout et al. (2004a, 2004b) have used 3-D
tomography to determine the mode of void
nucleation in ideal composites made of elastic
ceramic particles in an aluminum matrix. As
shown in Figure 29, a soft matrix (pure Al)
favors particle decohesion while a hard matrix
(precipitate hardening 2124 Al) leads to particle
cracking. A soft matrix prevents the stress in the
particle to attain the critical stress required for
the particle to crack, while the accumulation of
plastic strain on the interface between the par-
ticle and the matrix allows the progressive
opening of the submicron interface defects.
Void nucleation in a perfect lattice by local
cleavage or specific dislocation accumulation
is not common. However, these micromechan-
isms of void nucleation have been observed in
(a) (b)
7 m 10 m
Figure 28 SEM micrographs of damage nucleation in a 6060Al alloys during in situ uniaxial testing:
a, particle/matrix decohesion and b, particle fracture, as indicated by an arrow (Lassance et al., 2006a).
Ductile Fracture in Metals 715
very pure single-phase metals, like Ti alloys
(e.g., Thompson and Williams, 1977).
The recognition of the heterogeneous nature
of void nucleation is essential for understanding
many ductile fracture problems. Void nuclea-
tion is inherently a discontinuous process made
of a succession of discrete nucleation events.
Several studies have reported that voids nucle-
ate first on the largest inclusions, which involve
probably the largest of internal or interfacial
defects, and that void nucleation becomes
increasingly difficult with decreasing particle
sizes (Gurland, 1972; Garrison et al., 1997;
Lewandowski et al., 1989; Dighe et al., 2002).
Some materials involve different families of sec-
ond phases and inclusions. Among others, the
existence of two populations of particles, one
with limited resistance to void nucleation and
another, usually of a much smaller size, invol-
ving a much better resistance to void
nucleation, has been repetitively illustrated in
the literature (Cox and Low, 1974; Hahn and
Rosenfield, 1965; Marini et al., 1985; Li et al.,
1989; Haynes and Gangloff, 1997; Bron et al.,
2004; Asserin-Lebert et al., 2005). The effect of
this second population of voids can, in some
circumstances, be a dominant feature of the
damage process and will be discussed in
Section 2.06.3.5. The inhomogeneity can also
be due to a statistical distribution of the
matrix/particle cohesion stress about a mean
interfacial stress (see, e.g., Kwon and Asaro,
1990) or to local microplasticity effects in rela-
tion with crystallographic details (see, e.g.,
Bugat et al., 1999). The existence of different
modes of void nucleation (interface vs particle
fracture) also participates to the inherently het-
erogenous nature of the nucleation process. The
inhomogeneity in the particle distributions
causes local stress concentrations and is cer-
tainly an important reason of heterogenous
void nucleation (see, Lewandowski et al.,
1989; Dighe et al., 2002; Gammage et al.,
2004). For instance, the inhomogeneity in
local nucleation rate was thoroughly investi-
gated in duplex stainless steels (Pineau and
Joly, 1991; Devillers-Guerville et al., 1997;
Joly et al., 1990) using interrupted tests, as
shown in Figure 30 where clusters of cavities
represented by Vorono cells are clearly
observed. The histogram of the cell sizes
shows that a very small fraction of the surface
area in Figure 30 is leading to large local nuclea-
tion rates compared to the mean nucleation
rate. This clustering effect, particularly pro-
nounced in this material, is another feature of
cavity nucleation which should be kept in mind
when modeling ductile rupture.
Matrix: 2124 alloy
Matrix: pure aluminum
100 m
100 m
(a) (b)
Figure 29 Reconstructed images from in situ 3-D X-ray tomography for metal matrix composites involving
4% of ZrO
2
/SiO
2
spherical particles embedded inside: a, a pure aluminum matrix (the strain level is equal to
0.27); b, a 2124T6 aluminum matrix (the strain level is equal to 0.09). From Babout, L., Brechet, Y., Maire, E.,
and Fouge` res, R. 2004a. On the competition between particle fracture and particle decohesion in metal matrix
composites. Acta Mater. 52, 45174525.
15 m
2
5
m
m
Figure 30 Cast duplex stainless steel. a, Initiation of
cavities produced by the formation of cleavage
microcracks in the ferrite phase; b, Vorono cells
illustrating the heterogeneity in the distribution of
cavities initiated from cleavage microcracks.
Source: Devillers-Guerville, L., Besson, J., and
Pineau, A., 1997. Notch fracture toughness of a
cast duplex stainless steel: Modelling of
experimental scatter and size effects. Nucl. Eng.
Des. 168, 211225.
716 Failure of Metals
The interfacial strength between second-
phase particles and the matrix is dependent on
the local chemical composition. The segrega-
tion of impurity elements similar to those
which induce intergranular embrittlement (see
Section 2.06.2.4.4) can reduce the interfacial
resistance. Hydrogen-induced ductility losses
in low-strength steels could also be at least
partly explained in this way (Cialone and
Asaro, 1979); for recent modeling of hydro-
gen-induced decohesion at particle/matrix
interfaces, also see Liang and Sofronis (2003).
Impurity segregation at particles interface due
to irradiation effects has also been mentioned
earlier (see Section 2.06.2.4.4).
Quantifying experimentally the local mechan-
ical condition for void nucleation is an
experimental challenge. Two-dimensional digital
correlation methods can be used to determine
the local strain field corresponding to the onset
of void nucleation but the artifact of a surface
measurement has to be taken into account. In
the method proposed by Beremin (1981),
notched round bars were strained up to different
amount of deformation. After unloading, a sec-
tion on the specimen parallel to the loading axis
and comprising the axis of the specimen was
metallographically prepared in order to deter-
mine the boundaries of the region inside which
void nucleation had taken place. By combining
the experimental determination of the locus of
void nucleation with finite element simulations
of the specimen, it is possible to determine the
local mechanical conditions for nucleation. The
3-D in situ tomography study reported above
(see Figure 29), coupled with finite element
simulations in order to calculate the local stress
and strain fields, allows a more rigorous deter-
mination of the void nucleation condition
(Babout et al., 2004a, 2004b; Maire et al., 2005).
2.06.3.3.3 Computational cell simulations
Computational cell simulations are very use-
ful for capturing quantitatively the local
transfer between a matrix and second phases
as a function of the shape and mechanical prop-
erties of the particles and of the mechanical
properties of the matrix.
Lee and Mear (1992, 1999) have performed a
comprehensive set of calculations on ellipsoidal
inclusions embedded in viscous or elastoplastic
solids. Stress concentration factors were deter-
mined for both interface decohesion and
particle cracking. Selected results from their
studies are given in Figure 31 showing the evo-
lution of the stress concentration factors (K
I
for
the interface stress concentration factor and K
p
for the particle stress intensity factor) as a func-
tion of the applied remote strain E
d
for two
different strain-hardening exponents and mod-
ulus contrasts, under uniaxial tension
conditions. The stress concentration factors sig-
nificantly evolve only during the beginning of
the plastic deformation (during the first 2% of
strains) and only if the particle has a stiffness
similar to that of the matrix. The effect of the
strain-hardening exponent is relatively small.
Figure 32 shows the ratio of the stress concen-
tration factors K
p
/K
I
as a function of remote
axial deviatoric strain for different particle
aspect ratio W
p
. In agreement with the experi-
mental observations of Figure 28, particle
fracture is favored when the particle is
0.01 0.00
1.0
1.5
2.0
2.0
0.02 0.03 0.04 0.05
K
I
E
d
0
N = 5
N = 10
(a)
0.01 0.00
1.0
2.0
1.5
2.5
3.0
0.02 0.03 0.04 0.05
K
p
E
d
0
N = 5
N = 10
(b)
Figure 31 Stress concentration factor: a, at particlematrix interface K
I
and b, within particle K
p
as a function
of remote axial deviatoric strain. The particle aspect ratio is equal to 2 and the modulus contrast E
p
/E is equal
to 1 for the lower pair of curves, equal to 2 for the middle pair, and equal to 4 for the upper pair. The results are
shown for two different strain-hardening exponents N=5 and N=10 (defined through a RambergOsgood
representation). Source: Lee, B. J. and Mear, M. E. 1999. Stress concentration induced by an elastic spheroidal
particle in a plastically deforming solid. J. Mech. Phys. Solids 47, 13011336.
Ductile Fracture in Metals 717
elongated in the direction of loading. Other
important information in the studies by Lee
and Mear (1992, 1999) concern the effect of
the stress state and of the location of the inter-
face decohesion.
Christman et al. (1989), Llorca et al. (1991),
and Tvergaard (1993) conducted a series of unit
cell simulations of the interface separation
between short elastic fibers and a plastically
deforming matrix. The system underlying this
set of studies was metal matrix composites but
most of the results are generic. The interface
behavior is modeled using a traction separation
law (see seminal papers by Needleman, 1987,
1990). The precise location of the beginning of
the decohesion process was found to depend on
many factors such as the stress state and parti-
cle aspect ratio. Partial interface decohesion
was frequently reported. The effect of shear on
void nucleation has been investigated by Fleck
et al. (1989). More elaborated representative
volume elements involving particle clustering
have been addressed recently by Shabrov and
Needleman (2002). The fracture of brittle sec-
ond-phase particles (carbides in steels) has been
investigated by Kroon and Faleskog (2005) in
the framework of a study of cleavage in steels as
indicated earlier in Section 2.06.2.3.2. Strain
gradient plasticity based analysis (e.g., Fleck
and Hutchinson, 1997; Xue et al., 2002;
Niordson and Tvergaard, 2002; Niordson,
2003) or advanced dislocation dynamics cell
calculations (e.g., Cleveringa et al., 1999) per-
mit the analysis of RVE with particle sizes in
the micron or submicron size range, where
classical plasticity theory fails to capture prop-
erly the extra hardening contribution provided
by a large density of geometrically necessary
dislocations.
2.06.3.3.4 Void nucleation models
(i) Nucleation on a single particle
Analytic or closed-form void nucleation cri-
teria constitute the first essential ingredient in a
constitutive models involving damage evolu-
tion. Various void nucleation criteria have
been proposed based either on dislocation the-
ory (for crystalline materials) or on pure
continuum mechanics (e.g., based on Eshelby,
1957) theory to evaluate the load transfer.
Dislocation-based analysis is necessary when
the particle size is smaller than typically 1 mm
to properly account for the large density of
geometrically necessary dislocations that con-
trols the hardening at that scale due to the very
large plastic strain gradients. A good review of
these different criteria has been made by Berdin
(2004); see also Montheillet and Moussy (1986).
We limit the presentation here to general
aspects and to one specific void nucleation
criterion.
If the particle is brittle and deforms elasti-
cally, a simple one-parameter condition can be
motivated from linear-elastic fracture
mechanics arguments. Second-phase particles
always contain tiny submicron defects (e.g.,
Kroon and Faleskog, 2005; see Section
2.06.2.3.2). Considering a given size for the
internal cracks, particle fracture takes place
when the energy release rate becomes larger
than the particle fracture toughness, which can
be translated also into an effective critical stress
condition within the particle (e.g., see Ghosh
et al., 1997; Horstemeyer et al., 2003; Huber
et al., 2005). Now, in very small and clean brit-
tle particles, without defects, a critical stress
condition based on the theoretical strength of
the material is also valid. One thus needs proper
homogenization method to evaluate the stress
in the particle as a function of the overall stress.
In the case of interface fracture, a one-para-
meter linear-elastic fracture mechanics
approach is, in principle, not relevant if the
matrix surrounding the particle is plastically
deforming. Both the separation energy and
interface strength play a role in the problem.
In many instances, the energy condition can
easily be met while enough plastic deformation
must still be accumulated on the interface to
raise the stress above the critical strength level.
This is why both critical-strain-based models
(e.g., see Walsh et al. (1989) for Al alloys and
Joly and Pineau (1991) or Bugat et al. (1999) for
W
p
= 1
W
p
= 2
W
p
= 3
W
p
= 4
3.0
2.5
2.0
1.5
1.0
0.00 0.01 0.02 0.03 0.04 0.05
E
d
0
K
p
/
K
I
Figure 32 Ratio of the stress concentration factors
K
p
/K
I
as a function of remote axial deviatoric strain
for different particle aspect ratio W
p
and a strain-
hardening exponent n =5. Source: Lee, B. J. and
Mear, M. E. 1999. Stress concentration induced by
an elastic spheroidal particle in a plastically deforming
solid. J. Mech. Phys. Solids 47, 13011336.
718 Failure of Metals
cast duplex stainless steel) have been used as
well as a critical stress based models (e.g., see
Kwon and Asaro (1990) for spheroidized steel)
but none of them are totally satisfactory. The
formulation of an adequate condition for void
nucleation by interface fracture is a very diffi-
cult problem that many researchers try to solve
through phenomenological models involving
both strain- and stress-controlled nucleation
conditions with parameters tuned on experi-
mental data or RVE simulations (see Chu and
Needleman, 1980; Tvergaard, 1990). Usually,
the critical stress and critical strains are overall
values and not local values in the particle or
along the interface.
One relatively advanced void nucleation
model has already been introduced in Section
2.06.2.3.2. The fracture of the particle or of the
interface is assumed to occur when the maxi-
mum principal stress in the particle or at the
interface reaches a critical value
s
particle max
1
= s
bulk
d
or s
interf
d
[58[
which is different for each mechanism. Based
on the Eshelby theory (Eshelby, 1957) and the
secant modulus extension to plastically
deforming matrix proposed by Berveiller and
Zaoui (1979), the maximum principal stress
in an elastic inclusion (and at the interface)
o
1
particle max
can be related to the overall stress
state by using
s
particle max
1
= s
max
1
k
s
(s
e
s
0
) [59[
where o
1
max
is the maximum overall principal
stress and k
s
is a parameter of order unity
which is a function of the inclusion aspect
ratio W
p
and of the loading direction (see
Le Roy et al. (1981) for a model in the same
vein). An approximate explicit form for the
function k
s
(W
p
) can be given for long spheroi-
dal particles:
k
s
(W
p
) =
4
9(2L
a
1)
2 3L
a
W
2
p
_ _
[60[
where L
a
=ln(2W
p
1/2W
p
) =lim
l
cosh
1
(W
p
). The higher the aspect ratio, the
larger is k
s
and the sooner void nucleation
takes place. This void nucleation criterion was
initially proposed by the Beremin group
(Beremin, 1981), who also identified k
s
from
experimental data on steels with MnS inclu-
sions. They found that the values of k
s
predicted by the theory were about 2 times
higher than the experimental values because of
an overstiff response of the homogenization
scheme.
Typical mean values for particle fracture
stress or interface fracture stress are gathered in
Table 6. A very interesting set of results was
obtained by Babout et al. (2004a, 2004b) on
ideal composites Al matrixZS spherical
balls using 3-D tomography to detect the nuclea-
tion event and finite element cell calculations to
estimate the local stress in the particle and along
the interface. As explained above, the critical
stress for interface fracture is definitely not
Table 6 Void nucleation stress reported in the literature for particle fracture or particle decohesion
mechanisms
Particles Matrix
Critical stress
(MPa) Reference
Particle fracture
Elongated MnS A508 steel 1100 Beremin (1981)
Cuboidal TiN 4330 steel 2300 Shabrov et al. (2004)
TiN Inconel 718 12801540 Alexandre et al.
(2005)
4% Spherical ZrO
2
SiO
2
(ZS) balls Al2124 (T6) 700 Babout et al. (2004b)
20% Spherical ZrO
2
SiO
2
(ZS)
balls
Al2124 (T4) 700 Babout et al. (2004b)
Interface fracture
MnS A508 steel 800 Beremin (1981)
Si Al (cast) 550 Huber et al. (2005)
4% Spherical ZrO
2
SiO
2
(ZS) balls Al2124 (T6) 1060 Babout et al. (2004b)
4% Spherical ZrO
2
SiO
2
(ZS) balls Pure Al 250 Babout et al. (2004b)
20% Spherical ZrO
2
SiO
2
(ZS)
balls
Pure Al 320 Babout et al. (2004b)
Rounded Fe
3
C Spheroidized 1045
steel
1650 Argon and Im (1975)
CuCr particles Cu alloy 1000 Argon and Im (1975)
TiC Maraging steel 1820 Argon and Im (1975)
C nodules Cast iron 80 Dong et al. (1997)
Ductile Fracture in Metals 719
intrinsic to the particle but depends on the hard-
ening of the matrix. Note that there are many
important information missing in this table, such
as the mean particle size and particle size distri-
bution, the flow properties of each phase. These
information are necessary to allow cross-com-
parison between the various systems.
The model [59] remains rather qualitative
with respect to more accurate finite element
cell calculations presented in Figure 31, for
instance, and which could be used to provide
closer estimate of the load transfer between the
matrix and particle, or with respect to size
effects for small particles. These is definitely
room for improving such models based on
recent developments in homogenization theory
for nonlinear solids (e.g., Doghri and Ouaar,
2003). Nevertheless, the complexity of the frac-
ture process at that scale is such that critical
stress and particle fracture toughness will have
to be identified from experiments.
(ii) Void nucleation rate function
For all the reasons discussed at the beginning
of this section, the heterogeneity of the nuclea-
tion process must be introduced through a
distribution function. Following the spirit of
the model proposed by Needleman, Tvergaard,
and co-workers (see, e.g., review Tvergaard,
1990), a general expression for the void nuclea-
tion rate, based on [59], can be written as
_
f
nucl
= A
n
_ s
max
1
k
s
_ s
e
_ _
[61[
where
A
n
=
f
0
s
2p
_ exp
1
2
s
max
1
k
s
(s
e
s
0
) s
d
)
s
_ _
2
_ _
[62[
for o
1
max
k
s
(s
e
o
0
) =o
1
max
k
s
(s
e
o
0
)[
max
and
_
s
max
1
k
s
_
s
e
>0. In [62], f
0
is the initial void
volume fraction, o
d
) is the average void nuclea-
tion stress, and s is the standard deviation to be
determined experimentally or by inverse identi-
fication. Note that strain-controlled void
nucleation laws are frequently preferred because
they are more simple to implement numerically.
The initial void volume fraction f
0
is equal to
the volume fraction of particles giving rise to
voids if the nucleation mechanism is by com-
plete interface decohesion. For the case of
partial decohesion or particle fracture, the
initial void shape is very flat. The easiest option
is to take the volume fraction of spherical voids
having the same projected area. A more
advanced procedure to identify f
0
for initially
penny-shaped voids has been recently
addressed by Lassance et al. (2006a); see next
section. Different populations of voids with
different associated shapes of the particles, cri-
tical nucleation stress, and standard deviations
can be taken into account by adding their con-
tribution to the void nucleation rate.
2.06.3.4 Void Growth
2.06.3.4.1 Macroscopic evidences
The macroscopic softening effect induced by
the growth of voids in a plastically deforming
matrix is relatively weak. The presence of voids
can be measured macroscopically from density
measurements with accurate scales or from other
indirect methods, such as the change of the elastic
stiffness or change of resistivity (see Montheillet
and Moussy, 1986). Figure 33 shows a typical
evolution of the density observed in samples cut
in the neck of cylindrical bars pulled in tension
and made of a copper containing initially about
0.3% of copper oxide particles.
The most important macroscopic observa-
tion about ductile fracture is that the fracture
strain decays exponentially with increasing
stress triaxiality, as first reported by Hancock
and McKenzie (1976), followed by many other
teams (e.g., Marini et al., 1985; Devaux et al.,
1985; Becker et al., 1988; Bauvineau, 1996;
Decamp et al., 1997; Pardoen et al., 1998;
Jablokov et al., 2001; Huber et al., 2005). This
effect is directly related to a significant increase
of the void growth rate with increasing stress
triaxiality. The fracture strain e
f
is defined from
the reduction of cross-sectional area measured
on broken samples:
e
f
= ln
A
0
A
f
_ _
[63[
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
0.002
0.004
0.006
0.008
0.01
0.012
0.014
f
f
0
e
CuAR
CuA
CuAR
ef
ef
CuA
Figure 33 Variation of the porosity estimated from
density measurements as a function of straining from
samples cut in the minimum cross-sectional area of a
necking region, in as-received (CuAR) and annealed
copper bars (CuA) (see Pardoen and Delannay,
1998a).
720 Failure of Metals
where A
0
is the initial cross-sectional area and
A
f
is the final cross-sectional area. The stress
triaxiality T is defined as
T =
s
ii
3s
e
=
s
h
s
e
[64[
where o
h
is the hydrostatic stress. Figure 34
shows the schematic evolution of the fracture
strain as a function of the stress triaxiality aver-
aged along the deformation path. As
schematically represented in Figure 34, the
stress triaxiality is equal to the following:
v 0 under pure shear.
v 1/3 under single tension. Note that in a single
tension specimen, T is equal to 1/3 up to the
onset of necking and then steadily increases
inside the neck with increasing plastic defor-
mation (see Section 2.06.3.2.1).
v
3
_
,3 under plane strain tension.
v 0.61.8 in the center of the minimum cross
section of a notched round bar, depending
on the radius of curvature of the notch. The
cylindrical notched round bar geometry is
probably the best suited to study ductile frac-
ture. It allows probing a wide range of stress
triaxiality by changing the radius of curva-
ture of the notch. Moreover, the stress
triaxiality remains relatively constant all
over the deformation process in the center
of the minimum cross section. A standard
procedure to test cylindrical notched round
bars and interpret the results has been devel-
oped by the European Structural Integrity
Society (ESIS P6-98, 1998). Finite element
simulations are necessary to accurately eval-
uate the stress triaxiality.
v 2.755 inside the fracture process zone
(FPZ) in front of a crack tip. The value
depends on the strain-hardening exponent.
The specific and very important problem of
void growth at the tip of crack, controlling
the fracture toughness of ductile materials,
is discussed in Section 2.06.3.6.
v under purely hydrostatic stress (o
e
=0).
Usually, ductile fracture in industrial metal-
lic alloys only takes place for stress triaxiality
larger than 0.30.5. This is why most forming
operations are conducted under compressive
loadings, at least in the heavily plastically
deformed regions of materials. The effect of
the stress triaxiality has also been widely
studied by conducting tensile tests with super-
imposed hydrostatic pressure (e.g., Brownrigg
et al., 1983; Vasudevan, et al., 1989; and
review paper by Lowhaphandu and
Lewandowski, 1999). Ductile fracture can
even be suppressed if the pressure is high
enough leading to a full necking process up
to a final material point.
The fracture strain of metallic samples after
plastic forming operations is usually anisotro-
pic due to particle morphology effects and
plastic texture (e.g., Becker et al., 1989b;
Achon, 1994; Bauvineau, 1996; Benzerga
et al., 2004a).
2.06.3.4.2 Microscopic observations
After nucleation, the voids grow by plastic
deformation. The void growth process can be
observed during in situ testing inside a scan-
ning electron microscope. As shown in
Figure 35a, voids nucleated by particle frac-
ture open and become more rounded with
plastic deformation. Voids nucleated by par-
ticle decohesion are initially rounded and tend
to elongate in the principal loading direction
under low stress triaxiality (see Figure 35b). In
Figure 35b, the particle prevents the void to
contract in the direction transverse to the
main loading direction. If the void axis is
not aligned with the principal loading direc-
tion or if it undergoes shear deformation,
void rotation is observed (see Figure 35d).
Recently, in situ 3-D tomography experiments
have provided more complete information
about the complete 3-D morphology of the
voids, while avoiding the artifact of surface
observation (Babout et al., 2001, 2004a,
2004b; Maire et al., 2005; see also X-ray
microtomography experiments by Everett
et al. (2001)). An example of a 3-D recon-
structed image obtained on aluminum
containing ZS particles (see also Table 6) is
shown in Figure 35c.
Goto et al. (1999), Jablokov et al. (2001),
and Chae and Koss (2004) have prepared and
analyzed metallographic sections of samples
deformed up to various levels of straining.
F
r
a
c
t
u
r
e
s
t
r
a
i
n
Stress triaxiality
0.30.5
1
0.1
1.0 1.5 2.5 3.0 3.5
Figure 34 Typical variation of the fracture strain as a
function of the stress triaxiality in metallic materials.
Ductile Fracture in Metals 721
The relevance of these time-consuming experi-
ments very much depends on the quality of
the polishing. Polishing can indeed signifi-
cantly smear out the voids. By measuring a
large number of void sizes, they were able to
generate plots of the evolution of the void
volume fraction as a function of the straining
in different steels. An example is given in
Figure 36 (similar measurements were also
performed by other teams, e.g., Marini et al.
(1985)).
There has been only a very limited number of
experimental studies devoted to the coupling
between void growth and crystal orientation
(a) (b)
(d) (c)
Load
axis
Void
axis
Figure 35 Micrographs of voids growing by plastic yielding of the surrounding matrix: a, voids originating
from particle fracture (void growth in cast Al alloys nucleated by the fracture of Fe-rich particles or Si particles;
b, voids originating from the decohesion of the particle interface (void growth around a copper oxide inclusion
in copper; c, 3-D tomography reconstructed image where dark cavities can be seen around the large gray
particles; d, evidence of void rotation. a, from Huber, G., Brechet, Y., and Pardoen, T. 2005. Void growth and
void nucleation controlled ductility in quasi eutectic cast aluminium alloys. Acta Mater. 53, 27392749.;
b, from Pardoen, T., Doghri I., and Delannay, F. 1998. Experimental and numerical comparison of void
growth models and void coalescence criteria for the prediction of ductile fracture in copper bars. Acta Mater.
46, 541552; c, source: Maire, E., Bordreuil, C., Babout, L., and Boyer, J. C. 2005. Damage initiation and
growth in metals. Comparison between modeling and tomography experiments. J. Mech. Phys. Solids 53,
24112434; d, source: Benzerga, A. A., Besson, J., and Pineau, A. 2004a. Anisotropic ductile fracture. Part I:
Experiments. Acta Mater. 52, 46234638.
722 Failure of Metals
effects (e.g., Cre pin et al., 1996; Gan et al.,
2006) or to the analysis of the effect of the initial
void size on the void growth rate which tends to
decrease when the voids get smaller (e.g.,
Schlueter et al., 1996; Khraishi et al., 2001).
2.06.3.4.3 Void cell simulations
The principle of a void cell simulation is to
idealize the microstructure by considering simple
arrangements of voids and to use the finite ele-
ment method with proper boundary conditions.
The most simple arrangement is to consider per-
iodic distribution of voids. More sophisticated
representative elements can of course be con-
structed based on elementary patterns that
contain several voids. A representative set of
references dealing with void cell calculations is
given by the following list: Needleman (1972a,
1972b), Tvergaard (1981, 1982, 1990), Koplik
and Needleman (1988), Becker et al. (1989a),
Hom and McMeeking (1989a), Needleman and
Kushner (1990), Worswick and Pick (1990),
Needleman et al. (1992), Huang (1993), Becker
and Smelser (1994), Richelsen and Tvergaard
(1994), Brocks et al. (1995a), Kuna and Sun
(1996), Faleskog and Shih (1997), Steglich and
Brocks (1997), Steenbrink et al. (1997), Sovik
and Thaulow (1997), Faleskog et al. (1998),
Thomson et al. (1999), Pardoen and Hutchinson
(2000), Socrate and Boyce (2000), Pijnenburg and
Van der Giessen (2001), and Kimet al. (2004). All
these studies have contributed to improving our
understanding of all the possible factors that gov-
ern the growth of voids in elastoplastic or elasto-
viscoplastic materials.
The key information that can be extracted
from a simple cell made of a cylinder containing
a spheroidal void and loaded axisymmetrically
(see Figure 37) is presented in the following.
Periodic boundary conditions are enforced by
imposing the external faces to remain straight
and parallel to their initial orientation.
Although cylinders leave open space when
packed next to each other, this 2-D representa-
tion is known (Kuna and Sun, 1996; Worswick
and Pick, 1990) to give results very similar to
the 3-D unit cell based on hexagonal-type dis-
tribution of voids.
0.02
0.015
= 1.4
(
m
/
eq
)
ave
(A
f
)
crit
= 1.1 = 0.9 = 0.8
0.005
0.01
0
0
(a)
0.2 0.4 0.6
Equivalent plastic strain
V
o
i
d
a
r
e
a
f
r
a
c
t
i
o
n
0.8 1
Voidvoid
interaction
300 m
(b)
Figure 36 a, Void volume fraction measurements made of multiple micrographs taken from
metallographically polished section of broken notched round bars made of HSLA-100 steel. Finite element
simulations were used to estimate the stress triaxiality and effective plastic strain fields. b, The variation of the
void volume fraction as a function of straining was obtained by relating the local void measurements to the
local mechanical conditions. The method was repeated for different notch radii. Source: Chae, D. and Koss, D. A.
2004. Damage accumulation and failure in HSLA-100 steel. Mater. Sci. Eng. A 366, 299309.
L
0z
L
0x
(a)
(b)
R
0x
R
0z
R
pz
FE unit cell
Figure 37 a, Principle of void cell simulation based
on cylindrical geometry subjected to axisymmetric
loading conditions and involving a spheroidal void
and a particle. b, Finite element mesh.
Ductile Fracture in Metals 723
Motivated by the recognition that the stress
triaxiality is the main parameter controlling the
void growth rate (see Figure 34), the cell calcu-
lations are ideally performed at a constant
imposed stress triaxiality. Figure 38 shows the
overall effective stresseffective strain curves
predicted by void cell calculations performed
at a constant stress triaxiality T=1 for three
initial porosities: 10
2
, 10
3
, and 10
4
. The
voids are initially spherical and the void distri-
bution parameter, noted l
0
and defined as
l
0
=L
0z
/L
0x
, is equal to 1. A J
2
isotropic hard-
ening elasticplastic response is assumed for the
matrix, characterized by the following represen-
tation in uniaxial tension:
s
s
0
=
E
s
0
e for s<s
0
[65[
s
s
0
= 1
E
s
0
e
p
_ _
n
for s>s
0
[66[
where E is the Youngs modulus, o
0
is the initial
yield stress, and n is the strain-hardening expo-
nent. The Poisson ratio of the matrix n is always
taken equal to 0.3. For the sake of comparison,
the stressstrain curve of the reference nonpor-
ous material has also been added. The softening
due to the presence of the porosity only shows
up close to the onset of void coalescence. Until
that point, the effect of the porosity on the
stressstrain curve is small for porosity smaller
than 10
2
.
The various effects of the stress triaxiality are
summarized in Figure 39. Figure 39a shows the
axial stress versus axial strain curves correspond-
ing to different stress triaxiality equal to 1/3, 1,
and 3. In agreement with Figure 34, the fracture
strain significantly drops with increasing stress
triaxiality. As shown in Figure 39b, the drop of
fracture strain is directly related to an increasing
rate of porosity growth with increasing stress
triaxiality.
Figure 39 addresses also the effect of the initial
void aspect ratio of spheroidal void and the
evolution of the void aspect ratio. The aspect
ratio is designated by S or W, with S=ln(W)
and W=R
z
/R
x
(see Figure 37). Three different
initial void aspect ratios are considered: elon-
gated void with an aspect ratio 6; spherical
void with an aspect ratios of 1; and flat void
with an aspect ratio 1/6 (see Pardoen and
Hutchinson, 2000; see also analyses by Becker
et al., 1989b, 1989c). The effect of void shape is
very marked at low stress triaxiality. It is inter-
esting to note the saturation of the porosity at
large strains in uniaxial tension for the voids that
are initially spherical or elongated. As shown in
Figure 39c, the aspect ratio of the voids increases
a lot with plastic straining. As a matter of fact,
the voids both elongate in the loading direction
and contract in the transverse direction while
preserving the volume of the void, leading at
very large strain to a needle-like shape. This
transverse contraction is prevented if a particle
is present in the voids (see Figure 35b). The void
shape effect is still pronounced at T=1 but
decreases with increasing stress triaxiality. The
aspect ratio of voids that are initially spherical or
elongated tends to increase at low or intermedi-
ate stress triaxiality but to decrease at large stress
triaxiality (see Figure 39c; see Budiansky et al.,
1982). Voids that are initially flat always tend to
open first. At the intermediate stress triaxiality,
the porosity growth rate is significantly larger
for flat voids. Hence, the behavior of flat voids
(also called oblate) is different than the beha-
vior of more rounded or elongated voids (also
called prolate).
Proper understanding of the response of
materials involving flat voids is important as
many materials involve void nucleation by par-
ticle cracking or by partial interface decohesion
leading to initially penny-shaped voids (detailed
analysis can be found in Lassance et al.
(2006b)). Let us consider a constant relative
void spacing while changing only the initial
void aspect ratio and thus the initial porosity.
The initial relative void spacing
0
=R
0x
/L
0x
is
related to the initial porosity and initial void
shape through the following expression:
0
=
R
0x
L
0x
=
R
3
0x
L
3
0x
_ _
1,3
=
R
2
0x
R
0z
R
0x
L
0z
L
2
0x
L
0z
R
0z
L
0x
_ _
1,3
=
f
0
z
l
0
W
0
_ _
1,3
[67[
0
0.5
1
1.5
2
2.5
0
e
/
e
n = 0.1, E/
0
= 500, triaxiality = 1
f = 0
f = 10
2
f = 10
4
f = 10
3
0.5 1 1.5 2
Figure 38 Variation of the overall effective stress as
a function of the overall effective strain predicted by
a 2-D axisymmetric unit cell simulation with an
initially spherical void for three different initial void
volume fractions. The comparison is made with the
response of the nonporous material.
724 Failure of Metals
where is a geometric factor which depends on
the arrangement of voids: =/6=0.523 for a
periodic simple cubic array; z =
3
_
p,9 = 0.605
for a periodic hexagonal distribution; and =2/
3 =0.666 for a void surrounded by a cylindrical
matrix, which is the case under discussion here.
Prescribing the initial void spacing is similar to
considering voids originating from particle
cracking with a fixed particle spacing. Here the
particles are virtual: they have the same proper-
ties than the matrix material. The initial void
aspect ratio W
0
ranges between 0.001 and 2. A
virtual volume fraction of spherical particle
f
p
=1% is considered which is equivalent to pre-
scribing the relative void spacing
0
=0.247. The
initial void volume fraction is given by
f
0
=
W
0
W
p
f
p
[68[
where W
p
is the shape of the particle, equal to 1
here. Figure 40 gathers the results obtained at a
prescribed stress triaxiality equal to 1, in terms of
(a) the overall stressoverall strain curves corre-
sponding to the z direction; (b) the evolution of
the porosity; and (c) the evolution of the void
aspect ratio. The most important result emerging
from Figure 40 is that the evolution with strain-
ing of the overall stress, porosity, and void shape
is independent of the initial void aspect ratio W
0
if W
0
is typically lower than 0.030.1. This shows
that for flat voids, the key parameter controlling
the damage evolution process is the relative void
spacing
0
. The assumption of identifying an
effective initial porosity as equal to the porosity
associated to an equivalent spherical void is
acceptable as long as the particle volume frac-
tion is sufficiently low (f
p
=1%) (see Figure 40)
but not when the volume fraction is large
0
0.5
1
1.5
2
2.5
3
3.5
4
0
W
0
= 6
W
0
= 1
W
0
= 1/6
z
/
z
f
0
= 10
2
,
0
= 1, n = 0.1
T = 3
T = 1
T = 1/3
Onset of coalescence
(a)
0
0.02
0.04
0.06
0.08
0.1
0
W
0
= 6
W
0
= 1
W
0
= 1/6
f
z
T = 3
T = 1
T = 1/3
2
1
0
1
2
3
0
T = 3
T = 1
T = 1/3
S
=
l
n
(
W
)
z
W
0
= 1/6
W
0
= 1
W
0
= 6
(b)
0.2 0.4 0.6 0.8 1 1.2 1.4
(c)
0.2 0.4 0.6 0.8 1 1.2 1.4 0.2 0.4 0.6 0.8 1 1.2 1.4
Figure 39 Void cell results for f
0
=10
2
, l
0
=1, o
0
/E=0.002, n =0.1, and W
0
=1/6, 1, 6, at T=1/3, 1, 3.
a, (True) axial stress vs (true) axial strain; b, porosity vs axial strain; c, void shape vs axial strain.
From Pardoen, T. and Hutchinson, J. 2000. An extended model for void growth and coalescence. J. Mech.
Phys. Solids 48, 24672512.
Ductile Fracture in Metals 725
(f
p
=10%) (see Lassance et al. (2006b): the
response of the material becomes much more
dependent on the initial void shape (when
W
0
>0.1) when f
p
=10% than when f
p
=1%.
Figure 41 shows that the effect of the strain-
hardening exponent on the void growth rate is
not very significant. Note however, as discussed
later in Section 2.06.3.2, that the strain-hard-
ening exponent will have an indirect effect on
the damage accumulation as the prime para-
meter controlling plastic localization, and on
the strain distribution and stress triaxiality
inside the necking regions.
Figure 42 provides interesting information
about the rotation of elongated voids under
combined shear and normal loading. Two dif-
ferent orientations of the voids with respect to
the loading configurations are chosen and
embedded with a 3-D unit cell under fully per-
iodic boundary conditions. In both cases, the
voids rotate toward the direction of the max-
imum principal stress (Scheyvaerts et al., 2005;
see also Bordreuil et al. (2003) for void cell
simulations under shear conditions).
Other features of the void growth process
have been addressed using void cell simulations.
Several authors have shown that heterogeneities
in the void distributions, for instance clustering
effects, do not significantly affect the void
growth rates (Needleman and Kushner, 1990;
Huang, 1993; Thomson et al., 2003). The pre-
sence of small secondary voids in between
primary big voids has also been investigated
using void cell simulations, as discussed in the
next section on void coalescence. Benzerga
(2000) analyzed the effect of anisotropic material
behavior on the void growth rate and the cou-
plings with the morphological anisotropy caused
by nonspherical void shapes. Void cell calcula-
tions have also been performed with a crystal
plasticity constitutive response for the matrix
material (ORegan et al., 1997; Orsini and
Zikry, 2001; Kysar et al., 2005; Potirniche
et al., 2006; Gan et al., 2006) under different
loading conditions. Among others, Potirniche
et al. (2006) have shown that different crystal
orientations can lead up to a factor of two dif-
ference in the void growth rates. The effect of the
strain rate sensitivity on the void growth rate has
been addressed by Klo cker and Tvergaard
(2003). The void growth rate decreases and
void aspect ratio increases faster when increasing
the rate sensitivity. Higher-order constitutive
theories including strain gradient plasticity
effects and internal length scales associated to
the accumulation of geometrically necessary dis-
locations have also been used to study the effect
of the void size on its growth rate (e.g., Fleck and
Hutchinson, 1997; Shu, 1998; Li et al., 2003; Liu
et al., 2003). A decrease of the void growth rate
1
1.5
2
2.5
3
0 0.1 0.2 0.3 0.4 0.5 0.6
f
p
= 1%
T = 1
n = 0.1
W
0
=
z
/
0
Onset of void coalescence
(a)
1 0.5 0.2 0.1 0.010.001
z
0 0.1 0.2 0.3 0.4 0.5 0.6
z
f
p
= 1%
T = 1
n = 0.1
0
0.02
0.04
0.06
0.08
0.1
f
Onset of void coalescence
(b)
W
0
= 1 0.5 0.2 0.1 0.010.001
z
(c)
0
0.5
1
1.5
2
0 0.1 0.2 0.3 0.4 0.5 0.6
W
f
p
= 1%
T = 1
n = 0.1
Onset of void coalescence
W
0
= 1 0.5 0.2 0.1 0.010.001
Figure 40 Results of finite element unit cell
calculations for various initial void aspect ratios W
0
ranging between 0.001 and 1, under a constant stress
triaxiality T=1, with n =0.1 and E/o
0
=500, for
f
p
=1%: a, overall axial stress vs axial strain curves;
b, evolution of the porosity as a function of the axial
strain; c, evolution of the void aspect ratio as a
function of the axial strain. The onset
of coalescence is indicated by a bullet. From
Lassance, D., Scheyvaerts, F., and Pardoen, T.
2006b. Growth and coalescence of penny-shaped
voids in metallic alloys, Eng. Fract. Mech. 73,
10091034.
726 Failure of Metals
with decreasing void size is confirmed by all
these works. Liu et al. (2003) found that the
amplitude of this size effect significantly
increases with increasing stress triaxiality.
2.06.3.4.4 Void growth models
(i) Models for isolated voids
Although the state-of-the-art in void growth
modeling is connected to the advances in the
development of constitutive models for porous
medium, simple analytical models for isolated
voids remain very useful and powerful to per-
form first-order analysis, to guide intuition, and
to sustain pedagogical approaches of ductile
fracture. Models for isolated spherical voids
have been proposed by McClintock (1968) and
Rice and Tracey (1969). The Rice and Tracey
model (RT model) (Rice and Tracey, 1969)
evaluates the growth of a initially spherical
void in an infinite, rigid, perfectly plastic mate-
rial subjected to a uniform remote strain field.
Using well-chosen velocity fields, the varia-
tional analysis of Rice and Tracey (see
Chapter 2.03) leads, for the assumption of sphe-
rical void growth, to the following expression
for the average rate of growth:
dR
R
=
1
3
df
f
= a exp
3
2
T
_ _
de
ep
[69[
where R is the actual radius of the cavity, e
ep
is
the equivalent plastic strain, and a is a constant.
0
0.02
0.04
0.06
0.08
0.1
0 0.1 0.2 0.3 0.4 0.5 0.6
f
e
FE unit cell calculation
cylinder with a single void
2-D axisymmetric T = 1
f
0
= 1%, W
0
= 1,
0
= 1
E/
0
= 500
n = 0.1 n = 0.1
n = 0.1 n = 0.3
Figure 41 Variation of the void volume fraction as a function of the overall strain for strain-hardening
exponents n =0.1 and 0.3 under a constant stress triaxiality T=1, with E/o
0
=500, for f
0
=0.1% and
initially flat voids.
Figure 42 3-D void cell calculations with elongated spheroidal voids under combined plane strain shear/
tension conditions, and constant stress triaxiality. Periodic boundary conditions are enforced. Two different
orientations of the voids are chosen to analyze their rotation. Source: Scheyvaerts, F., Onck, P., Bre chet, Y.,
and Pardoen, T. 2005. Multiscale simulation of the competition between intergranular and transgranular
fracture. In: Proceedings of ICF11 11th International Conference on Fracture (ed. A. Carpinteri), 2025
March 2005, Turin, Italy, CD-ROM 5331.
Ductile Fracture in Metals 727
A value a =0.283 was computed by Rice and
Tracey (1969) and re-evaluated by Huang
(1991) using additional velocity fields. Huang
retrieved the same result but with a=0.427 for
T>1 and a =0.427T
1/4
for T>1. In practice,
the porosity is finite and the stress and strain
fields around the voids interact. This explains
why some authors (Marini, 1984; Marini et al.,
1985) found higher values for a when calibrat-
ing the model toward experimental
measurements. This effect of a finite porosity
is addressed in the next subsection.
The general analysis of Rice and Tracey was
also taking into account the change of void
shape, assuming ellipsoidal voids and plastic
flow conditions in which the directions of the
principal axes of the strain rates remain fixed
throughout the strain path. The rates of change
of the radii of the void, in the principal direc-
tions, write (Rice and Tracey, 1969; Thomason,
1990)
dR
k
R
=
dR
R
(1 E
v
)de
p
k
[70[
where E
v
is a void shape parameter. From void
cell simulations, Worswick and Pick (1990) have
shown that the model would be more accurate if
(1 E
v
) in expression [70] is not taken as a con-
stant equal to 5/3, but involves a dependence on
stress triaxialities, on f
0
, on the strain-hardening
exponent, n, and on e
ep
at low stress triaxialites
(Le Roy et al., 1981; Worswick and Pick, 1990).
For instance, for f
0
=0.01 and n =0.2 and for
low stress triaxialities (<1.2), the results of
Worswick and Pick (1990) lead to (see Pardoen
and Delannay, 1998b)
E
v
=
(1.9 0.72T)exp (3,2)e
ep
_ _
2exp (3,2)e
ep
_ _
1
1 [71[
It is worth mentioning the recent extension of
the Rice and Tracey model to the so-called
Taylor-type dislocation-based strain gradient
plasticity material response (Gao et al., 1999;
Huang et al., 1999; Qiu et al., 2001, 2003) by
Liu et al. (2003) in order to account for size effects
when the void size is in the submicron range.
(ii) Constitutive models for porous media
As shown in Figure 38, the softening induced
by the nucleation and growth of the porosity in
realistic alloys is usually quite weak.
Nevertheless, there are several circumstances
where a full constitutive model involving the
coupling with the porosity on the material
response is needed. Small amount of softening
induced by the porosity is sometimes sufficient
to significantly affect plastic localization (e.g.,
Yamamoto, 1978; Ohno and Hutchinson,
1984), a critical point of investigation for sev-
eral types of forming operations. The drop of
the stress-carrying capacity near fracture initia-
tion and during cracking must be accounted for
when modeling ductile tearing problems.
Constitutive models for porous media also
offer a natural framework for the introduction
of void interaction effects.
Constitutive models for porous elastoplastic
or viscoplastic media (also called dilatant plas-
ticity models) have been proposed based either
on micromechanics or on thermodynamics.
The micromechanical model developed by
Gurson (1977) is probably the model that has
received the largest attention in the literature.
Many efforts have been devoted to extend its
range of validity and of application since the
1970s. A first subsection will thus focus on this
model and its extensions (see also Chapter
2.03). Note that advanced model for porous
elastoplastic or viscoplastic media has also
been formulated based on variational methods
(Zaidman and Ponte Castan eda, 1996; Ponte
Castan eda and Zaidman, 1994, 1996;
Kaisalam and Ponte Castan eda, 1998; Garajeu
et al., 2000). A second shorter subsection will be
devoted to a presentation of the Rousselier
model which has been developed in the frame-
work of a thermodynamical approach. This
model has several merits which are worth
mentioning.
(iii) Gurson model and extensions
The Gurson model is the first micromechanical
model for ductile fracture which introduces a
strong coupling effect between deformation and
damage. The model is representative of the void
growth stage only. It is derived from an analysis
similar to the one performed by Rice and Tracey
for anisolatedvoid(see previous subsection). The
model is based on a simplified representation of
the voided material which consists in an hollow
sphere. The only nondimensional microstructural
feature inthe model is the void volume fraction or
porosity, f. The material of the hollowsphere (i.e.,
the matrix) is assumed to be rigid, isotropic, per-
fectly plastic witha yieldlimit. The matrix obeys a
standard von Mises yield criterion and associated
flow rule. The analysis is based on an upper-
bound method consisting in two steps: (1) find a
family of velocity fields compatible with the
boundary conditions (here homogeneous) and
(2) minimize the plastic dissipation within the
proposed family (see for details Chapter 2.03).
In the work of Gurson, only one field was con-
sidered so that no minimization procedure was
needed. The main result is an estimate of the yield
function for the porous metal which, applying the
normality rule, can be used to derive the plastic
728 Failure of Metals
flow direction. The yield surface is given by the
following equation:
X
s
2
e
s
2
y
2f cosh
1
2
s
kk
s
y
_ _
1 f
2
= 0 [72[
This yield surface is identical to the von Mises
(or J
2
) yield surface when f =0. It is very similar
to the von Mises yield surface for low hydro-
static stresses (i.e., o
h
=o
kk
/3 <<o
y
). The main
difference is that, owing to the presence of a
void, plastic yielding takes place under large
hydrostatic stress whatever the value of the
deviatoric stress (effective stress) while a von
Mises material is incompressible and does not
yield plastically under pure hydrostatic stresses.
As soon as voids have started to nucleate, the
accumulation of plastic deformation causes the
enlargement of the voids and an increase of the
void volume fraction which, by stating volume
conservation, writes
_
f
growth
= (1 f)_ e
p
ii
[73[
where e_
ij
p
are the ij components of the overall
plastic strain rate tensor. The full evolution law
for the void volume fraction is written as
_
f =
_
f
growth
_
f
nucl
[74[
Expressions for the void nucleation rate f
_
nucl
have been discussed in the previous section.
The hardening behavior of the matrix material
is related to the overall stress and plastic strain
rate through the energy balance initially pro-
posed by Gurson (1977):
s
y
_ e
p
y
(1 f) = s
_
e
p
[75[
where the mean yield stress of the matrix mate-
rial o
y
is a function of e
y
p
:o
y
X h(e
y
p
).
In order to calculate the evolution of the two
variables f and o
y
as well as the stress state, use
is made of an associated flow rule:
_
e
p
ij
=
_
g
0
0s
ij
[76[
The stresses are calculated from
_ s
ij
= C
ijkl
_
e
kl
_
e
p
kl
_ _
[77[
where the C
ijkl
are the elastic moduli. The deri-
vation of an analytical expression for the
tangent moduli L
ijkl
relating the stress rates o
ij
to the strain rates e_
kl
_ s
ij
= L
ijkl
_
e
kl
[78[
can be found by imposing the consistency
condition
_
= 0 [79[
which leads to
s
ij
=
1
L
ijkl
_
L
el
ijkl
L
el
ijmn
_
0,0s
mn
_
L
el
klmn
_
0,0s
mn
_
_
_
0,0s
mn
( )L
el
mnrs
_
0,0s
rs
_
_
0,0f
__
1 f
__
0,0s
tt
_
s
mn
_
0,0s
mn
__
0,0s
y
__
0
y
,0S
p
y
_
1
_
_
_
kl
[80[
where the derivatives of the flow potential can
be easily determined analytically and L
ijkl
el
are
the elastic constants. In this tensorial form, the
elastic moduli write for an elastic material
L
el
ijkl
=
E
1 v
1
2
d
ik
d
jl
d
il
d
jk
_ _
v
1 2v
d
ij
d
kl
_ _
[81[
where c is the Kronecker operator.
Many extensions of the Gurson model have
been proposed in the literature. The format of
the model remains basically the same. The
volume conservation of the matrix material
[73] is almost always invoked as well as the
associated flow rule [76]. The extensions
mainly differ by the formulation of more
sophisticated yield surfaces, extra evolution
laws for the new variables introduced in the
model, and/or new hardening laws for the
overall material. A relatively comprehensive
survey of these extensions is provided in
Table 7 (see also Chapter 2.03 for details
about some of the extensions). Note also
that extensions of the Gurson model to poly-
meric materials have been developed
(Steenbrink et al., 1997; Pijnenburg and Van
Der Giessen, 2001; Challier et al., 2006).
(iv) A brief on Rousselier model
Although the Rousselier model (Rousselier,
1987) has been developed from general thermo-
dynamical principles, its formulation remains
close to the Gurson model. The yield surface
of the Rousselier model writes
X
s
e
1 f
D
r
s
1
f exp
[s
h
[
3(1 f)s
1
_ _
s
y
= 0 [92[
where o
1
and D
r
are adjustable parameters.
The normality rule (76) is also assumed. It is
interesting to outline some differences
between the Gurson and Rousselier models.
Under pure shear conditions, that is, o
h
=0,
the Rousselier model predicts a damage evo-
lution (as the normal to the yield surface in a
o
h
o
e
plane does not coincide with the o
e
axis), whereas, in the absence of nucleation,
the Gurson model does not lead to damage
change. Under pure hydrostatic stress states,
that is, o
e
=0, the Rousselier yield surface has
a vertex which implies that at high stress
Ductile Fracture in Metals 729
Table 7 Survey of the extensions of the Gurson model
Purpose of the extension
and main references Brief description
Improve global accuracy Better comparison toward unit cell calculations can be achieved by adding
adjusting parameters q
1
and q
2
into the expression of the yield surface:
Tvergaard (1981)
X
o
2
e
s
2
y
2q
1
fcosh
q
2
2
s
kk
s
y
_ _
1 (q
1
f)
2
= 0 [82[
These q factors can be used to correct for the effect of void shape changes, of large
strain-hardening exponent, and of void interaction. However, the use of this
version of the Gurson model should preferably be restricted to the modeling of
voids that are initially more or less spherical
Perrin and Leblond (1990) The theoretical value of q
1
=1.47 for spherical voids
Koplik and Needleman
(1988)
Calibrations of the q factors toward unit cell calculations have been performed by
several authors. Typically, q
1
~1.5 and q
2
~1. More detailed calibration is given in
the following table, providing the best set (q
1
,q
2
) for different o
0
/E and n
Faleskog et al. (1998) n o
0
/E=0.001 o
0
/E=0.002 o
0
/E=0.004
0.025 (1.88,0.956) (1.84,0.977) (1.74,1.013)
0.05 (1.63,0.95) (1.57,0.974) (1.48,1.013)
0.1 (1.58,0.902) (1.46,0.931) (1.29,0.982)
0.15 (1.78,0.833) (1.68,0.856) (1.49,0.901)
0.2 (1.96,0.781) (1.87,0.8) (1.71,0.836)
Improved modeling of
strain hardening
The validity of the yield surface [72] and of the energy balance [75] deteriorates when
the strain-hardening exponent becomes large (typically when n >0.2, see the
change of the q factors in the table above when n =0.2). An improved yield
function has been worked out, involving two yield strength indices o
y1
and o
y2
:
Leblond et al. (1995)
X
o
2
o
2
y1
2q
1
fcosh
q
2
2
s
kk
s
y2
_ _
1 (q
1
f)
2
= 0 [83[
supplemented by closed-form evolution laws for o
y1
and o
y2
Kinematic hardening Various extensions of the Gurson model to kinematic hardening have been
formulated. Typical yield surface writes Mear and Hutchinson
(1985); Leblond et al.
(1995); Arndt et al.
(1997); Besson and
Guillemer-Neel (2003);
Mu hlich and Brocks
(2003)
X
3
2
(s
ij
a
ij
)(s
ij
a
ij
)
o
2
y
2fcosh
1
2
(s
kk
a
kk
)
s
y
_ _
1 (q
1
f)
2
= 0 [84[
where s
ij
is the stress deviator and a
ij
is the deviator of the symmetric tensor
specifying the center of the yield surface. The model must be supplemented by
an evolution law for a
ij
, see the references
Plastic anisotropy The first approximate way to extend the Gurson model to plastic anisotropy is by
calculating the effective stress based on an anisotropic yield criterion (e.g., Hill,
1948, 1950; or more advanced models by Barlat et al., 1991, 1997; Karafillis
and Boyce, 1993; Li et al., 2003; Bron and Besson, 2004; Van Houte and Van
Baele, 2004) while preserving the expression of the yield surface [72]. For the
Hill model, the effective stress writes
Doege et al. (1995); Liao
et al. (1997); Rivalin et al.
(2001); Grange et al.
(2000); Benzerga et al.
(2002); Wang et al. (2004)
o
e
=
3
2
h
1
s
2
11
h
2
s
2
22
h
3
s
2
33
2h
4
s
2
12
2h
5
s
2
23
2h
6
s
2
31
2
_
[85[
where the h
i
s are the anisotropy coefficients and s is the deviatoric stress tensor
expressed in the orthotropic reference frame
As a matter of fact, the expression of the yield surface also changes when
reworking completely the Gurson model with an anisotropic plastic theory. For
the Hill model, it becomes
(Continued )
730 Failure of Metals
Table 7 (Continued)
Purpose of the extension
and main references Brief description
Benzerga and Besson
(2001); Benzerga et al.
(2004b)
X
o
2
e
s
2
y
2fcosh
1
h
s
kk
s
y
_ _
1 f
2
= 0 [86[
where h is equal to
h =
8
5
h
1
h
2
h
3
h
1
h
2
h
2
h
3
h
3
h
1
4
5
1
h
4
1
h
5
1
h
6
_ _
2
[87[
The error made by keeping the form [72] rather than using [86] could also be
corrected by adjusting the q
2
factor as a function of the anisotropy coefficients
Void shape effect A micromechanical analysis of a spheroidal void embedded in a confocal
spheroidal cell can be performed along the lines of the Gurson analysis leading,
after complex mathematical developments, to the following yield surface
Gologanu et al. (1993,
1994, 1997)
X
C
o
2
y
s9 Zs
g
h
X
2
2q(g 1)(g f)cosh k
s
g
h
s
y
_ _
(g 1)
2
q
2
(g f)
2
= 0
[88[
where o9 is the deviatoric part of the Cauchy stress tensor; o
h
g
is a generalized
hydrostatic stress defined by o
h
g
=o : J; X is a tensor associated to the void axis
and defined by 2/3 e
z
e
z
1/3 e
x
e
x
1/3 e
y
e
y
; J is a tensor associated to
the void axis and defined by (1 2a
2
)e
z
e
z
a
2
e
x
e
x
a
2
e
y
e
y
; | | is the
von Mises norm; C, j, g, k, a
2
are analytical functions of the state variables
S=ln W=ln(R
z
/R
x
) and f, q is a parameter that has been calibrated as a
function of f
0
, W
0
, and n (see references for complete expressions). For
spherical voids (i.e., S=0), the yield surface is identical to the initial Gurson
surface [72].
Pardoen and Hutchinson
(2000)
Benzerga et al. (2004) The evolution law for S writes
_
S =
3
2
(1 h
1
)
_
e
p
_
e
p
kk
3
d
_ _
: P h
2
_
e
p
kk
[89[
Gologanu et al. (1997) where analytical expressions for h
1
, h
2
as a function of S and f can be found in the
references, P is a projector tensor, defined by e
z
e
z
and e
z
is a unit vector
parallel to the main cavity axis; c is the Kronecker tensor
Scheyvaerts et al. (2006a);
Kaisalam and Ponte
Castan eda (1998)
The last new variable which enters the model is the orientation of the main void
axis e
z
. The initial proposal was to rotate the void axis with the material axis.
More correct rotation laws based on homogenization theory have been
formulated and assessed toward unit cell calculations (e.g., Figure 42).
Rate dependency and
viscoplasticity
The most simple extension of the Gurson model to rate-sensitive behaviors of the
type
o
o
0
= 1
E
o
0
p
_ _
n
1
0
_ _
1,m9
[90[
Peirce et al. (1982, 1984);
Tvergaard (1990);
Moran et al. (1991);
Haghi and Anand (1992)
is to retain the expression of the yield surface [72] (or of one of its extension), and use
it as a flow potential. Expression [75] allows to directly calculate the effective
plastic strain rate. This type of extension is frequently used for numerical reasons,
while keeping m9 very small (<0.01). When the rate sensitivity exponent m9
departs significantly from 0, the expression of the flow potential is not relevant
anymore. The flow potential for spherical voids which writes
(Continued )
Ductile Fracture in Metals 731
triaxiality ratios the plastic deformation ten-
sor retains a non-zero shear component.
It has been shown that the use of the
Rousselier model together with a rate-depen-
dent flow law for the matrix does not lead to
realistic results in terms of void growth rates. A
modified Rousselier model has been proposed
by Tanguy and Besson (2002) which keeps the
specific shape of the Rousselier yield surface
(i.e., damage growth under pure shear, vertex
under pure pressure). The yield surface is now
expressed as
X
s
e
(1 f)s
y
2
3
D
r
f exp
q
R
2
[s
h
[
(1 f)s
y
_ _
1 = 0
[93[
where D
R
and q
R
are the damage parameters
used instead of o
1
and D
r
. Another modifica-
tion has been proposed by Sainte Catherine
et al. (2002) where o
1
is now a function of the
effective plastic strain.
When properly calibrated, the Rousselier
model provides predictions as good as the
regular version of the Gurson model with the
advantage of better capturing localization.
2.06.3.5 Void Coalescence
During the voidgrowthphase, plastic deforma-
tion is homogeneous at a scale of a representative
volume element, that is, at the scale of a volume
containing one or a few voids depending on the
heterogeneity of the microstructure. Except for
voids clusters present in highly heterogenous
materials (e.g., Devillers-Guerville et al. (1997)
and Achon (1994)), the interaction during the
void growth phase between voids in industrial
alloys is usually weak (e.g., Needleman and
Kushner (1990) and Huang (1993)) due to the
low initial porosity, and the effect of damage on
the global behavior can be taken into account
with a single porosity parameter as explained in
the previous section about void growth in metals.
Conversely, the coalescence of voids leads to a
transition from a stable phase of diffuse plastic
deformation to a localized mode of plastic defor-
mation within the ligament between the voids
Table 7 (Continued)
Purpose of the extension
and main references Brief description
Leblond et al. (1994)
X o
2
e
f F
s
kk
2
_ _
m9 1
m9 1
1
F
s
kk
2
_ _
_ _
1
m9 1
m9 1
f
2
F(x) = 1
[x[
m91
m9
m9
_ _
m9
[91[
Flandi and Leblond (2005);
Klocker and Tvergaard
(2003)
was developed for arbitrary nonlinear viscous materials (without hardening, i.e.,
n =0) and is identical to the Gurson model when m9 =. Extensions to
spheroidal voids have also been proposed by combining [88] and [91]
Two populations Leblond and co-workers have developed models that account for the presence of
a second population of voids around large primary voids which lead to larger
void growth rates due to the presence of ruined materials around the large
primary voids or to the successive coalescence of these small voids with the
large void
Perrin and Leblond (2000);
Enakoutsa et al. (2005)
Presence of the particle
inside the void
The presence of a hard particle within the void prevents the void to contract in the
direction transverse to the main loading direction under low stress triaxiality
conditions. This effect can only be accounted for with a model which already
incorporates void shape changes
Siruguet and Leblond
(2004a, 2004b); Maire
et al. (2005)
Effect of the void size The effect of the initial void size has been accounted for by reworking the analysis
of Gurson in the case of a so-called Taylor-type dislocation-based strain
gradient plasticity model (Gao et al., 1999; Huang et al. 1999; Qiu et al., 2001,
2003)
Wen et al. (2005)
732 Failure of Metals
located in the most critical region, with material
off the localization plane usually undergoing elas-
tic unloading. Void coalescence leads to the
formation of a macroscopic crack that can then
propagate through the material with limited
amount of additional external work. This local
plastic localization phenomenon called void coa-
lescence is the physical transition to fracture
controlling the resistance of a metallic alloy to
ductile fracture. To establish precise terminology,
void coalescence has been reserved for the part
of the void enlargement evolution after the transi-
tion to the localized mode of plasticity, and void
growth is used to characterize void enlargement
before localization. The voidcoalescence mechan-
ism is a localization mechanism at the scale of the
void size that must thus be distinguished from the
localization in a shear band at the mesoscopic
scale, as discussed in Section 2.06.3.2, whose
width is, in general, not controlled by the void
spacing. The confusion can arise because when
such a mesoscopic-localized band develops, the
void growth rate inside the band increases and
fracture occurs rapidly with small additional
increase of remote displacements.
In ductile materials, void coalescence occurs
after significant amount of void growth, typi-
cally when the diameter of the void has
doubled. In less ductile materials, such as in
high-strength aluminum, coalescence can start
right after the nucleation of the voids
(Worswick et al., 2001) and the mechanism is
said to be nucleation controlled.
2.06.3.5.1 Macroscopic evidences
The schematic of Figure 20 shows a clear
change of slope after the maximum load. This
sharp slope change can be associated to the initia-
tion of fracture. The initiation of fracture consists
in the linking of two or a few voids in the most
damaged region of the material. In an overall
load/displacement plot such as shown in
Figure 20, fracture initiation occurs soon after
the onset of coalescence due tothe highly localized
character of the process. It is thus possible to
detect quite accurately the occurrence of the first
coalescence events on the global stressstrain
curve. This point has been confirmed experimen-
tally by several authors (e.g., Benzerga, 2000). In
ductile alloys, fracture initiation occurs after the
maximum load while in less ductile alloys, frac-
ture initiates during the rising part of the load.
2.06.3.5.2 Microscopic observations
The most interesting experimental observa-
tions of the void coalescence mechanisms are
obtained by interrupting mechanical tests just
after cracking initiation and by preparing
metallographical sections. This is a tedious
work which requires trial and errors and repe-
titive polishing sequences to reveal the region of
the material involving the coalescing cavities.
The ductile fracture process can also be investi-
gated using academic systems in which holes
are drilled into plates (see Dubensky and Koss,
1987; Magnusen et al., 1988, 1990; Becker and
Smelser, 1994; Weck et al., 2006). An example is
provided in Figures 43a43f for different
arrangements of cavities. The cavities are
cylindrical. These two types of experiments
allow the study of the two main modes of coa-
lescence in an ideal situation.
The first mode of coalescence, shown in the
sequence ace, is the internal necking mode
of coalescence, where the ligament between the
two voids shrinks with a shape typical of a
necking process. During the process of coales-
cence, the voids evolve toward a diamond
shape. The fracture profile will be flat oriented
at 90
from
the main loading direction (e.g., Richelsen and
Tvergaard, 1994). The main characteristic of
this tensile mode of coalescence is a sharp tran-
sition into an uniaxial straining mode of the
volume element imposed by the material out-
side the localization band which unloads
elastically and behaves as a rigid block (see
Koplik and Needleman, 1988; Becker et al.,
1989a; Brocks et al., 1995a, and Richelsen and
Tvergaard (1994)).
Figure 45shows the variationof the radial strain
e
r
as a function of the axial strain e
z
predicted by
axisymmetric void cell calculations performed
with three different initial void aspect ratios and
under three different applied stress triaxialities.
The transition to an uniaxial straining mode is
almost instantaneous (see also Koplik and
Needleman (1988) for initially oblate voids, and
Brocks et al. (1995b) andRichelsenandTvergaard
(1994) for 3-Dcomputations). This transitioncon-
stitutes a direct indicator of localization and is
used for quantifying the strain at the onset of
void coalescence. The onset of coalescence can
also be detected by a sharp change in the overall
stressstrain curves (see Figures 38 and 39a).
Void coalescence induces an increase of the
void growth rate and a transition in the void
Figure 44 Fractography of a 6060-T4 Al alloy
showing large dimples originating from the
intermetallic particles and small dimples resulting
from the decohesion of Mn dispersods.
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
W
0
= 6
W
0
= 1
W
0
= 1/6
z
T = 3
T = 1
T = 1/3
Figure 45 Void cell results for f
0
=10
2
, l
0
=1,
o
0
/E=0.002, n =0.1, and W
0
=1/6, 1, 6 at T=1/3, 1,
3 showing the variationof the radial strainas a function
of axial strain (Pardoen and Hutchinson, 2000).
Ductile Fracture in Metals 735
shape evolution. After the onset of void coales-
cence, the radial growth is significantly larger
than the axial growth. The end of the coalescence
process in a real material usually consists in the
failure of the remaining ligament (by microclea-
vage, crystallographic shearing, or with the help
of a second population of smaller voids) rather
than radial void growth until impingement. Thus,
after the onset of void coalescence, the void
expands rapidly in the radial direction until the
final failure of the ligament. During this process,
axial void growth remains small. Consequently,
the mean depth of the void (i.e., R
z
) measured on
the fracture profile is a good approximation of
the void half-height at the onset of void coales-
cence. Hence, if the mean distance between
dimple centers (ideally corrected by the value of
the overall transverse strain) provides informa-
tion about the spacing between particles giving
rise to void nucleation, the depth of the voids is
the most pertinent dimension to measure on a
fracture profile to gain information about the
void coalescence conditions.
Void cell simulations provide important infor-
mation about the state of the damage at the
onset of void coalescence. Figure 46 shows that
the value of the porosity at the onset of void
coalescence varies with the stress triaxiality, the
porosity, and initial void shape (Pardoen and
Hutchinson, 2000). If the initial porosity is
small enough (typically, f
0
<0.001), the critical
porosity can reasonably be considered as inde-
pendent of the stress triaxiality for most
practical purposes. The value of the strain-hard-
ening exponent also affects the critical porosity.
As shown in Figure 47, by elongating the cell
(i.e., by increasing l
0
while keeping the void
spacing constant), the zone undergoing elastic
unloading becomes larger and, in terms of the
overall strain, the unloading slope during coales-
cence becomes more and more steep. It is
remarkable to note the straight unloading slope
during coalescence, which has been verified by
Tvergaard (1997) up to very large strains using
remeshing techniques. The reason for the coales-
cence to take place earlier when the cell length
increases, although the initial porosity decreases
proportionally, is caused by the elevation of the
stress in the ligament which favors the localiza-
tion of the deformation in the ligament (see later
the section on coalescence models).
The effect of the presence of a second popula-
tion of voids has been investigated using cell
calculations by different authors. Tvergaard
(1998) and Faleskog and Shih (1997) treated
the secondary voids explicitly in the finite ele-
ment mesh using 2-D plane strain conditions.
Several authors have modeled the second popu-
lation using a Gurson-type model for the matrix
surrounding the primary voids (e.g., Brocks
et al., 1995a, 1995b; Fabre` gue and Pardoen,
2006). Figure 48 shows the overall stressstrain
curve obtained with a fixed total initial void
volume fraction f
10
f
20
=1.5 10
3
, once
without secondary voids (f
10
=1.5 10
3
,
f
20
=0) and once with secondary voids
(f
10
=1 10
3
, f
20
=0.5 10
3
). The influence
of the presence of a second population
(nucleated here immediately) on the onset of
coalescence is very clear as well as the negligible
effect on the overall stress evolution before coa-
lescence (note that the void growth rate of the
primary void is also not affected by the presence
of the second population).
0
0.02
0.04
0.06
0.08
0.1
0.12
stress triaxiality
f
c
0 1 2 3 4 5 6
W
0
= 1
W
0
= 1/6
W
0
= 6
0
= 1, n = 0.1
f
0
= 10
4
f
0
= 10
2
Figure 46 Stress triaxiality dependence of the critical
porosity at the onset of coalescence, for different
initial porosity and void shapes, with n =0.1 and
l
0
=1. From Pardoen, T. and Hutchinson, J. 2000.
An extended model for void growth and coalescence.
J. Mech. Phys. Solids 48, 24672512.
0
0.5
1
1.5
2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
e
/
e
L
r0
/ R
r0
= 3.22 W
0
= 1 T = 1 n = 0.1
f
0
=
0
f
0
=
1.25 10
3
0
=
16
f
0
=
2.5 10
3
0
=
8
f
0
=
5 10
3
0
=
4
f
0
=
2 10
2
0
=
1
f
0
=
4 10
2
0
=
1/2
f
0
=
10
2
0
=
2
Onset of coalescence
Figure 47 Effective stress vs effective strain curves
for n =0.1, W
0
=1, a constant L
r0
/R
r0
ratio equal to
3.22, and T=1. From Pardoen, T. and Hutchinson, J.
2000. An extended model for void growth and
coalescence. J. Mech. Phys. Solids 48, 24672512.
736 Failure of Metals
Coalescence in shear was first investigated
numerically by Yamamoto (1978) and
Tvergaard (1981, 1982) using finite element
cell calculations involving a cylindrical void
under plane strain conditions in the direction
of the void axis and periodic boundary condi-
tions. The kinematics associated to this loading
configuration favors the development of shear
localization. Coalescence by shear localization
is also observed in 3-D cell calculations invol-
ving spherical voids for low strain-hardening
exponents and low stress triaxiality (Richelsen
and Tvergaard, 1994). Specific packing of voids
where the closest neighbors are located at 45
z
f
10
=
1 10
3
f
20
=
5 10
4
1.5 10
3
0
n2
= 0, T = 1, W
0
= 1, n = 0.1,
0
/E = 500
Figure 48 Overall stressstrain curve obtained with
cylindrical void cell calculations and showing the
influence of the presence of a second population on the
onset of coalescence for a fixed total initial void volume
fraction f
10
f
20
=1.510
3
. The stress triaxiality T is
equal to 1, the initial void aspect ratio W
0
=1, and the
initial void distribution parameter l
0
=1.
p
0.50
0.40
0.30
0.20
0.10
0.00
Figure 49 Void sheet mode of coalescence involving
large primary voids and a population of secondary
small voids for low strain biaxiality and low strain
hardening exponent. From J. Faleskog and C. Shih,
1997, Micromechanics of coalescence I. Synergistic
effects of elasticity, plastic yielding and multi-size-
scale voids. J. Mech. Phys. Solids, 45, 2145.
Ductile Fracture in Metals 737
the introduction of at least some microstruc-
tural information related to the void/ligament
dimensions and geometry.
(ii) Micromechanical model for void
coalescence by internal necking
Coalescence by internal necking is the most
documented mode of coalescence and has, up to
now, received the largest attention. The tensile
void coalescence or internal necking mode is a
diffuse localization at the microscale. Different
criteria have been proposed in the literature to
predict the onset of void coalescence.
The criterion of Brown and Embury (1973) is
based on an explicit micromechanical point of
view: in a perfectly plastic material, internal
necking is considered to start when it is possible
for 45
R
2
x
R
2
z
_
to allow mutual 45
tangents. In
terms of nondimensional parameters, the con-
dition of Brown and Embury rewrites:
2
W
2
1
_
= 1 [94[
The evolution of the relative void spacing in the
case of spheroidal voids, that is, =R
x
/L
x
, can
be written as
_
=
1
3
_ e
z
_ e
x
_
f
f
_
W
W
_ _
[95[
A more realistic void coalescence model can
be obtained by directly treating the mechanism
of tensile plastic localization in the intervoid
ligaments: diffuse plasticity throughout void-
containing cells gives way to localized deforma-
tion within the ligament with the material
outside the ligament unloading elastically.
Consider a thin annular cylindrical disk of elas-
ticperfectly plastic material welded to rigid
plates and constrained against flow at the outer
radius. An approximate analysis for the limit
load of this configuration, with associated aver-
age true stress, can be carried out along the lines
of the Hill (1950) plane strain analysis of a thin
plastic layer welded to and squeezed by two rigid
plates. The analysis assumes that the material in
the disk moves outward flowing in shear and
otherwise supporting only hydrostatic tension
such that the three normal stresses are approxi-
mately equal (see Pardoen and Hutchinson,
2000). More accurate representations have been
developed by Thomason (1990), who extensively
studied the transition to localization for elastic
perfectly plastic solids using slip-line solutions
leading to the following condition for the onset
of void coalescence:
s
n
(1 f
2
)s
y
1
(1
2
)
=
2
3
_
a
1
W
_ _
2
1.24
_
[96[
where the volume fraction of secondary voids f
2
is equal to 0 in the original criterion and the
parameter a has been fitted as a function of the
average value of the strain-hardening exponent
n: a(n) =0.1 0.22n 4.8n
2
(0 _n _0.3); see
Pardoen and Hutchinson (2000). Criterion [96]
states that coalescence occurs when the stress
normal to the localization plane o
n
reaches a
critical value. This critical value decreases as the
voids open (W increases) and get closer to each
other ( increases). The dominant parameter
controlling the transition to the coalescence
mode is the relative void spacing . The poros-
ity affects the coalescence indirectly through the
link with the void spacing and through its
softening effect on the applied stress o
z
. In a
more advanced application of this model
(Scheyvaerts et al., 2006), the localization con-
dition [96] is tested in all possible directions
which requires to generalize the definition of
and W as a function of the orientation of the
localization plane.
The relationship between o
z
/o
0
and the over-
all strain e
z
is sketched qualitatively in Figure 50
for a cell of material containing a finite porosity
made of voids that grow and coalesce through
an internal necking process. At low overall
strain, o
z
/o
0
from [96] is far greater than the
actual value, which is better predicted by a
Gurson-type model. This last model was indeed
constructed with velocity fields capturing plas-
ticity all around the void. However, the actual
z
Exact solution
Approximate solution for
localized flow (e.g., Thomason)
z
/
0
Approximate solution for
diffuse flow (e.g., Gurson)
Figure 50 Schematic description of the competition
between the two modes of plasticity reflected by the
predictions of the two types of theories.
738 Failure of Metals
solution peaks and falls until localization sets in,
and then the actual solution merges with the
artificially constrained localized solution. This
is the transition point, and from this point on,
plasticity is localized within the ligament. The
velocity fields used to build the Gurson model
are not realistic anymore. Internal necking is not
a bifurcation phenomenon. Nevertheless, the
transition to localization occurs sharply, and
the competition depicted in Figure 50 is an aid
to thinking about the transition condition.
Based on experimental and theoretical stu-
dies described previously (e.g., Cox and Low,
1974; Hahn and Rosenfield, 1965;, Marini
et al., 1985; Faleskog and Shih, 1997; Haynes
and Gangloff, 1997; Tvergaard (1998), Perrin
and Leblond, 1990, 2000; Bron et al., 2002;
Asserin-Lebert et al., 2005), the nucleation
and growth of a second population of voids
has been shown to significantly accelerate the
damage process and cut down the ductility.
Motivated by unit cell calculations involving
both large primary voids and small secondary
voids (e.g., Tvergaard, 1998; Fabre` gue and
Pardoen, 2006), the second population of
voids is assumed to affect only the coalescence
between the primary voids and not the growth
of the primary voids. If we consider that the
presence of secondary voids within the ligament
between two large primary voids does not influ-
ence the overall response of the material, the
only possible effect of these small voids is
through a local softening of the matrix in
between the primary voids. Hence, the effect
of secondary voids on the onset of coalescence
by internal necking can be introduced heuristi-
cally into eqn [96] by multiplying the current
mean yield stress of the matrix material by
(1 f
2
). Based on finite element void cell simu-
lations treating the matrix as a Gurson material
(see Figure 48 and Fabre` gue and Pardoen,
2006), it was found that the localization process
is triggered by the local value of f
2
in the most
damaged region of the ligament between two
primary voids. In other words, if the cell calcu-
lations are used to evaluate both f
2
in the most
damaged part of the ligament and o
z
, o
y
, W,
and , then eqn [96] provides relatively accurate
predictions of the onset of void coalescence.
The most damaged region was always found
to be located next to the surface of the primary
voids in the minimum section of the ligament
(see Fabre` gue and Pardoen, 2006).
Consequently, f
2
in eqn [96] should be seen as
the local value of f
2
next to the surface of the
primary voids. The condition [96] was validated
toward the finite element calculations and
turned out to accurately capture the onset of
void coalescence for a wide range of initial
parameters (see Fabre` gue and Pardoen, 2006).
An estimate of the evolution of f
2
in the
region next to the surface of the primary void
is required in order to couple the void coales-
cence condition [96] to a void growth model.
The state of deformation adjacent to the surface
of the primary voids can be described in the
following way. The local circumferential strain
rate on the equatorial section of the void sur-
face (z =0) is directly related to the transverse
rate of the void growth through
_ e
s
yy
=
_
R
x
R
x
=
1
3
_ e
z
2_ e
x
_
f
f
_
W
W
_ _
[97[
The estimation of the local strain rate e
zz
s
is
more complex and requires some approxima-
tion to relate it to the evolution of the cavity:
_ e
s
zz
~ _ e
z
2_ e
x
_
f
f
2
_
W
W
_ _
p
2
1
2
1
W
2
1
_ _
_ _
[98[
The derivation of eqn [98] and its validation is
discussed in Fabre` gue and Pardoen (2006). At
the surface of the voids, we also have that
o
rr
s
=0. By applying these loading conditions to
a constitutive model involving a void nucleation
law (e.g., Beremin criterion) and a void growth
law (e.g., Gurson), one can estimate the evolu-
tion of f
2
and use the extended coalescence
condition [96] to predict the onset of fracture in
the presence of two populations of voids.
(iii) Void coalescence in shear
The first option for predicting coalescence by
shear localization is to analyze the process at
the scale of a row of discrete voids similar to the
Thomason model for internal necking devel-
oped in the previous section. There are only a
very limited number of attempts in the litera-
ture to formulate such a closed-form condition.
McClintock (1968) stated that a shear localiza-
tion band sets in between two ellipsoidal voids
when
1
s
e
ds
e
de
e
=
3
8
_
1 W
2
_
1 W
0
1 W
_ _
2
f
f
0
1 f
0
1 f
_ _
2,3
0
exp(e
I
) [99[
where e
I
is the maximum principal strain. This
coalescence model has been recently used by
Ragab (2004) for analyzing various experimen-
tal data. Note that a simple upper-bound
condition has also been proposed by Richelsen
and Tvergaard (1994) and compared to finite
element cell calculations.
The second option is to analyze the problem
from a slightly different perspective by addressing
the prediction of a shear localization band within
Ductile Fracture in Metals 739
the framework of constitutive model for porous
materials. Such kinds of calculations have been
initially proposed by Yamamoto (1978) and then
by several authors (e.g., Saje et al., 1982; Pan
et al., 1983; Ohno and Hutchinson, 1984;
Mear and Hutchinson, 1985; Tvergaard,
1982, 1987; Tvergaard and Van der Giessen,
1991). Early approaches involved the incor-
poration of heterogenous zones of higher
initial porosity to trigger the shear localiza-
tion. Following the works of Rudnicki and
Rice (1975), Doghri and Billardon (1995)
have proposed to compute the bifurcation
condition to detect cracking initiation and
the direction of the localization band (see
Besson et al., 2003). Theoretically, a shear
band analysis in a continuum described with
a damage model is different than solving a
localization problem with the exact geometry
and constraint resulting from the presence of
discrete voids. Nevertheless, for practical pur-
poses, the two approaches seem to lead to
comparable predictions. Proper modeling of
shear coalescence is still a matter of open
debate. Ideally, a micromechanics-based ana-
lysis of ducticle fracture should rest on
multiple coalescence criteria: one for internal
necking and one for void sheeting.
(iv) Coalescence in columns
Note finally that Gologanu et al. (2001) have
worked out theoretically the third mode of coa-
lescence called coalescence in columns in
which voids concentrate along vertical columns
instead of horizontal layers. Evidences of coa-
lescence in columns can be found in the thesis of
Benzerga (2000). This mode of coalescence can
be of importance in the transverse delamination
of plates having elongated bands of closely
spaced particles.
2.06.3.5.5 Models for the coalescence process
For low-stress triaxiality loadings, the energy
associated to the final enlargement of the void
during the coalescence stage is small with
respect to the energy associated to the void
growth stage. The onset of void coalescence is
an excellent indicator of the fracture strain of
the material. At large stress triaxiality, typical
of the state of loading in the region located in
front of a static or growing crack, coalescence
starts at small strains and then the amount of
energy spent during coalescence can be signifi-
cant. Hence, a full constitutive model for the
response of the material during coalescence is
necessary for simulating crack propagation.
Four types of approaches can be found in the
literature. The first two have been developed in
order to provide simple numerical implementa-
tion, the third one is specific for capturing void
sheeting mechanisms, and the fourth is specific
to the internal necking process.
(i) Phenomenological acceleration
of the void growth rate
Tvergaard and Needleman (1984) (see also
Needleman and Tvergaard, 1984) have pro-
posed to simulate the coalescence process by
artificially accelerating the rate of increase of
the porosity in the following way:
f
*
=
f if f<f
c
f
c
d(f f
c
) if f _ f
c
_
[100[
where f
c
is the porosity at the onset of coales-
cence which is either prescribed as a material
parameter or predicted from a void coales-
cence condition, for instance [94] or [96] (e.g.,
Zhang and Niemi, 1995). The factor c is an
adjusting parameter which varies typically
between 3 and 8 and which can be fitted on
void cell calculations or on experimental data.
The idea is to preserve the format of the Gurson
model after the onset of coalescence which is
numerically very convenient. This model does
not distinguish between different modes of coa-
lescence. Note that Benzerga (2002) proposed a
theoretical analysis for estimating c.
(ii) Numerical reduction of the load-carrying
capacity
Several authors (Xia et al., 1995) proposed,
in the framework of a finite element implemen-
tation, to ramp down the stress in a given
number of time increments after a critical por-
osity f
c
is attained. This method is based on the
assumption that an accurate description of the
coalescence process is unimportant, which is
probably a good approximation for many pro-
blems. There is no control on the energy
dissipated during coalescence. Note that the
seminal simulations of crack propagation
using a noncoupled damage model (e.g., the
Rice and Tracey model) had the same spirit,
i.e., neglecting the coalescence phase by simply
releasing nodes when a critical damage value
was reached (dEscatha and Devaux, 1979).
A more advanced implementation has been
developed by Shih and co-workers (e.g., Xia
and Shih, 1995a, 1995b, 1996; Ruggieri et al.,
1996; Gao et al., 1998a; Gullerud et al.,
2000). Basically, after the critical porosity is
reached, a cohesive zone representation (see
Section 2.06.3.7.1) is used to ramp down the
stress to zero (see Figure 51). In this
740 Failure of Metals
approach, the stress decreases with applied
displacement and there is thus a control of
the energy dissipated during coalescence. The
adjusting parameter is the slope of the
unloading ramp.
(iii) Model for coalescence in shear
Besson et al. (2003) have proposed to specifi-
cally model the void sheet mechanism in the
framework of the finite element technique by
prescribing a critical porosity (condition for the
onset of coalescence) above which secondary
voids are nucleated. This nucleation of a second
population naturally accelerates the rate of
increase of the overall porosity and triggers the
occurrence of a localization band. This
approach, which does not really impose any
extra coalescence condition, allows, for instance,
reproducing quite well the cup-and-cone frac-
ture profile. The localization orientation agrees
with the prediction of a bifurcation analysis
using the Rudnicki and Rice (1975) condition.
(iv) A full constitutive model for the internal
necking process
As a matter of fact, the solution [96] is not
only a condition for the onset of coalescence
but also provides the evolution of the stress
during the full coalescence process as a function
of the evolution of the shape and spacing of the
voids. It constitutes thus the basis for elaborat-
ing a constitutive model for the full coalescence
process. This idea has been followed by
Pardoen and Hutchinson (2003) and Benzerga
et al. (2004b). In order to use solution [96] as a
full constitutive model, it was transformed
heuristically into a yield surface (Pardoen and
Hutchinson, 2003):
coalescence
X
| s9 |
s
y
3
2
[s
h
[
s
y
F(W. ) = 0 [101[
with
F(W. ) X
3
2
(1
2
) a
1
W
_ _
2
b
_ _
[102[
A more physical generalization of [96] in
stress space to occur in a well-defined orienta-
tion and to lead to a large constraint in the
transverse direction; see Scheyvaerts et al.
(2005). The yield surface is supplemented by
the normality rule to evaluate the plastic
strains:
_ e
p
ij
= g
d
ds
ij
[103[
Evolution laws for the geometrical variables,
that is, the void aspect ratio, W, the porosity,
f, and the relative void spacing, , as well as for
the current mean yield stress of the matrix
material have been formulated based on simple
geometrical rules:
_
W =
3W(2
2
1)
4f
_ e
p
e
[104[
_ =
(3 2
2
)
4f
_ e
p
e
[105[
_
f = (1 f)_ e
p
kk
[106[
_ s
y
=
0s
y
0e
p
y
2
f
_ e
p
e
[107[
(see Benzerga et al. (2004b) for more advanced
void shape law during coalescence, accounting
for the evolution toward a diamond shape).
Figure 52 illustrates the competition between
the two modes of plastic deformation (diffuse
plasticity during void growth vs localized plas-
ticity during void coalescence), generalizing the
message of Figure 50 for a general stress state.
The yield surfaces and loading history corre-
spond to a high constraint situation (a
constant stress triaxiality of 3) relevant to
a material element in front of a crack tip.
Figure 52a corresponds to elastic loading: the
stress state lies within the elastic domain of the
two yield surfaces. As shown in Figure 52b,
(b) (a)
(c)
=fraction of extinct cell
forces remaining on
nodes after f f
g
0
0
1
D D
0
D
0
D
0
D
Figure 51 Linear tractionseparation model with
release fraction l. a, Cell at extinction value of the
porosity height is D
0
; b, force release completed at
D=(1 l)D
0
; c, linear tractionseparation model.
Source: Gullerud, A. S., Gao, X., Dodds, R. H., Jr.,
and Haj-Ali, R. 2000. Simulation of ductile crack
growth using computational cells: Numerical
aspects. Eng. Fract. Mech. 66, 6592.
Ductile Fracture in Metals 741
the first yield surface to be reached is F
growth
. In
the beginning of the deformation, the voids are
small and the spacing is relatively large resulting
in a diffuse plastic deformation. With increasing
deformation, F
growth
first tends to expand due to
hardening and then contract due to void growth
softening. Void growth and ligament reduction
also induces a contraction of F
coalescence
.
Figure 52c corresponds to an overall strain just
before the onset of coalescence and Figure 52d to
an overall strain just after the onset of coales-
cence. When the two yield surfaces intersect at
the current loading point, the transition to
coalescence occurs. With increasing deforma-
tion, the coalescence yield surface tends to
contract very rapidly toward the zero stress
state. Figure 52e shows the two yield surfaces
far in the coalescence stage.
This coalescence model has been shown to
capture quite well the drop of the load-carrying
capacity after the onset of coalescence (see
Pardoen and Hutchinson, 2000). However,
numerical problems related to the jump from
one yield surface to another make the implemen-
tation within robust finite element procedures
very complex. Furthermore, it is difficult to
1
0
1
0
1
0
1
0
1
0
coalescence
growth
von Mises
Current
stress
state
Slope
3/2
h
/
f
e
q
/
h
/
f
e
q
/
h
/
f
h
/
f
e
q
/
e
q
/
h
/
f
e
q
/
f
(a) (b)
(d) (c)
(e)
Figure 52 Transition from: (a) elastic behavior, to (b), (c) plasticity/void growth, to (d), (e) plasticity/void
coalescence in terms of the variation of the yield surfaces and loading point as a function of increasing straining.
Calculations performed for a constant stress triaxiality equal to 3.
742 Failure of Metals
formulate rules for the update of the quantities
entering the void growth model during the coa-
lescence stage, in case the coalescence process
stops and the void growth resumes.
2.06.3.6 Fracture Strain of Metals
The understanding and modeling of the dif-
ferent steps of the process of nucleation,
growth, and coalescence of voids have reached
a level of maturity that makes possible the for-
mulation of fairly generic views about ductility.
The ductility is defined here as the fracture
strain of a representative volume element of
material subjected to homogeneous loading
conditions. The fracture strain can be measured
experimentally directly on the fracture surface
based on the reduction of the cross-sectional
area. In the first subsection, a closed-form
model for the fracture strain will be developed
based on the analytical elementary models pro-
posed in the last three sections. In the second
subsection, more complex effects or more
quantitative predictions will be highlighted by
providing results obtained indifferently either
with FE void cell calculations or with an
advanced damage model (the idea is that these
advanced damage models have been success-
fully assessed by comparison with the
numerical unit cell simulations). These generic
predictions will be quantitatively or sometimes
only qualitatively assessed toward selected
experimental data. For the sake of simplicity,
we will always assimilate the void coalescence
strain as the fracture strain which is an approx-
imation because the material keeps deforming,
but only very moderately, after the onset of
coalescence up to final material separation.
2.06.3.6.1 Simple closed-form estimates
of the fracture strain
A first approximate expression for the frac-
ture strain can be formulated by assuming that
voids nucleate at e
e
=e
c
with a spherical shape,
grow spherically, and coalesce following the
Brown and Embury condition [94]. Let us con-
sider an average stress triaxiality T along a
deformation history which is characterized by
a strain e
z
in the main loading direction, and
e
y
=be
x
in the two transverse directions with
[e
x
[ _[e
y
[. The principal axis remains fixed dur-
ing deformation. Elasticity is neglected. The
initial porosity (f
0
) is small, allowing the use of
the isolated voids assumption. Using volume
conservation, the effective strain writes
e
e
=
2
3(1 b)
3(1 b b
2
)
_
e
z
=
2
3
3(1 b b
2
)
_
e
x
[108[
For instance, e
e
=e
z
for axisymmetric condi-
tions and e
e
= 2e
z
,
3
_
for plane strain
conditions. The Brown and Embury void coa-
lescence condition [94] writes for spherical
voids
=
1
2
2
_ [109[
Note that one could also use a coalescence cri-
terion based on a critical void radius which
would lead essentially to the same conclusions.
Coalescence will take place in the x-direction
because [e
x
[ _[e
y
[. Hence, the relevant void spa-
cing measure is given by
=
R
L
x
=
R
0
exp 0.427(e
e
e
c
) exp
3
2
T
_ _ _ _
L
0
exp(e
)
[110[
in which the void radius has been evaluated
using the Rice and Tracey relation integrated
from the nucleation strain e
c
. By combining
[108][110], the effective fracture strain writes
e
f
=
ln 1,2
2
_
0
_ _
0.427exp (3,2)T ( )e
c
0.427exp (3,2)T ( ) (3,2
3(1 b b
2
)
_
)
[111[
The initial relative void spacing can be related
to the initial porosity using eqn [67]. Assuming
immediate void nucleation (e
c
=0) and axisym-
metric loading conditions, eqn [111] rewrites
e
f
=
1.04 (2,3)ln f
0
,z ( )
0.85 exp (3,2)T ( )
[112[
The model can be qualitatively assessed by
comparing the predictions of [112] with a wide
set of results collected for various copper sys-
tems by Edelson and Baldwin (1962), as shown
in Figure 53. The copper bars have been loaded
up to fracture under uniaxial tension condition.
As explained in Section 2.06.3.2.1, the difficulty
with the tensile test is to estimate the stress
triaxiality evolution in the necking region.
A good approximation of the mean value of
the stress triaxiality during a tensile test on a
cylindrical bar can be given as a function of the
fracture strain based on the results of Figure 24:
T =
1
3
if e
f
<n
T =
1
3
1
3
(e
f
n)
2
e
f
if e
f
>n
[113[
The two eqns [112] and [113] were combined
in order to generate the curves drawn in
Figure 53 for different geometrical factors ,
Ductile Fracture in Metals 743
for two strain-hardening exponents n =0.1 and
0.2. The model captures quite well the experi-
mental trends. It underestimates the fracture
strain even for the highest geometrical factor
. In real engineering materials, the effective
geometrical factor should probably be smaller
than the factor resulting from periodic void
distribution in order to account for heteroge-
neous distribution effects. The underestimation
in the predictions mainly comes from that
the change of the void aspect ratio was not
accounted for in the model. Indeed, as
explained earlier, the elongation of the void at
small stress triaxiality leads to larger fracture
strains. Furthermore, this very simple model
neglects possible delayed void nucleation and
is not valid for porosity larger than about 5%.
This first simple model can be improved in
many different ways, for instance by introdu-
cing the Beremin condition for void nucleation
[59], by using the version of the Rice and Tracey
model accounting for void shape changes [70]
(still for initially spherical voids), and/or by
using the more accurate coalescence model by
Thomason [96]. Here, we provide one variant
destinated specifically to materials involving
initially penny-shaped voids. Using the rela-
tionship between the stress triaxiality and the
axial stress in axisymmetric conditions
s
z
s
y
= T
2
3
[114[
the Thomason condition [96] writes
(3,2)T 1 ( )
(1
2
)
= a
1
W
_ _
2
1.24
_ _
[115[
with a(n) =0.10.22n4.8n
2
(0_n_0.3). The
Rice and Tracey evolution law for the radial
evolution of the void radius under constant stress
triaxiality writes (note an error in eqns [23] and
[28] of Lassance et al. (2006b) now corrected in
eqn [116]: the last sign must read )
R
x
R
x0
=
_
1
0.427 exp (3,2)T ( )
_
0.427 exp (3,2)T ( ) 1 ( ) [
exp 0.427 exp (3,2)T ( )(e
e
e
c
) ( ) 1[ [116[
which allows the estimation of the current rela-
tive void spacing :
=
0
R
0
exp
e
z
2
_ _
[117[
The Beremin model for void nucleation [60]
can also be rewritten in terms of the stress
triaxiality, providing the following estimate of
the nucleation strain:
e
c
=
s
0
E
(s
c
,s
0
) 1 T
2
3
k
s
_ _1,n
[118[
In the case of initially flat voids and axisym-
metric loading conditions, one can show that
the following relationship for the void aspect
ratio (note an error in eqn [29] of Lassance et al.
(2006b) now corrected in eqn [119])
0
0.5
1
1.5
0 0.05 0.1 0.15 0.2
= 0.3, n = 0.1
= 0.4, n = 0.1
= 0.5 (simple cubic), n = 0.1
= 0.6 (hexagonal), n = 0.1
= 0.6 (hexagonal), n = 0.2
Experimental mean line
f
Volume fraction of second phases
Present work
Copperironmolybdenum
Copper holes
Copper chromium
Copper alumina
Copper iron
Copper molybdenum
Copper alumina
Copper silica
Zwilsky and Grant
Data collected by
Edelson and Baldwin (1962)
1.5
1.0
0.5
0 0.1 0.2
Volume fraction, f
D
u
c
t
i
l
i
t
y
,
I
n
A
0
/
A
f
0.3
Figure 53 Variation of the fracture strain as a function of the volume fraction of second-phase particles.
Comparison between a large set of experimental results collected on copper by Edelson and Baldwin (1962) and
the predictions of the simple analytical model [112] using different value for the void arrangement factor and
different strain-hardening exponents n.
744 Failure of Metals
W = l
0
exp(e
z
e
c
) 1
exp((e
c
e
z
),2)
[119[
is relatively accurate (see Lassance et al.,
2006b). The coalescence condition can be
solved by combining [114] to [119] in order to
estimate the fracture strain for a given average
stress triaxiality level. The only parameters of
the model are the critical fracture stress of the
particle o
c
and the particle volume fraction and
shape (from which
0
can be calculated).
2.06.3.6.2 More advanced predictions
of the fracture strain
Finite element unit cell calculations involving
one or a few voids under well-chosen boundary
conditions, as described in the three sections
devoted to the nucleation, growth, and coales-
cence of voids, enlighten how and to what
extent changing the loading or microstructural
parameters affects the fracture strain. These
results can be supplemented by results obtained
with an advanced constitutive damage model,
relying on the Beremin model for void nuclea-
tion [60], on the extension of the Gurson model
by Gologanu [88], and on the extension of the
Thomason model for void coalescence [96]. The
predictions of these models will be presented
and compared to experimental data in order
to discuss successively the effects of
1. the initial void volume fraction,
2. the initial void shape,
3. the void distribution,
4. the presence of secondary small voids,
5. the flow properties (yield strength and
strain hardening exponent),
6. the resistance to void nucleation, and
7. the microstructural heterogeneities
on the ductility, considering always the full
practical range of stress triaxiality.
The effect of the initial volume fraction of
voids (or of particles giving rise to voids) as
well as of the stress triaxiality has been dis-
cussed in details above. Once again, the fracture
strain is a strong function of the stress triaxial-
ity and the ductility is thus in no way a material
property.
The effect of the void shape is difficult to
capture with simple models because the void
growth rate depends on the evolution of the
void shape which depends itself on the stress
triaxiality and initial void shape. Figure 54
shows the variation of the ductility as a func-
tion of the stress triaxiality for three different
initial void shapes. At low stress triaxiality, the
effect is such that coalescence might never occur
for elongated voids and might occur for flat
voids due to the important difference in the
relative void spacing =R
/L
(L
is the
same but R
is different).
Specific results related to flat initial voids are
shown in Figure 55 providing the variation of
the ductility as a function of the particle volume
fraction for different initial void shapes, con-
sidering uniaxial tension conditions (T=1/3)
(see Lassance et al., 2006b). The initial void
volume fraction is prescribed by [68]. The
threshold porosity f
p
*
under which coalescence
never occurs is indicated. This figure clearly
shows that, for initially flat voids, a wide
range of particle volume fractions gives rise to
a process of stable void growth followed by
void coalescence, at this low stress triaxiality.
0.01
0.1
1
0 1 2 3 4
W
0
= 1/6
f
0
= 1%, n = 0.1,
0
/E= 0.001
f
Stress triaxiality
W
0
= 1
W
0
= 6
Figure 54 Variation of the ductility as a function of
the stress triaxiality for various initial void shapes,
with an initial porosity equal to 1%.
0.01
0.1
1
0 0.1 0.2 0.3 0.4
n =0.1
n =0.3
f
f
p
h
/
e
=1/3
E/
0
=500
d
/
0
=0
0
= 1
W
0
= 0.01
W
0
= 0.05
W
0
= 0.2
W
0
= 0.5
f
p
*
Necking for n=0.1
Necking for n=0.3
Figure 55 Variation of the ductility as a function of
the particle volume fraction for different initial void
aspect ratios W
0
=0.01, 0.05, 0.2, 0.5, strain-
hardening exponent n =0.1 or 0.3, and a stress
triaxiality equal to 1/3. The necking condition
predicted by the Conside` re criterion is indicated to
show the range of particle volume fraction for which
fracture can be expected before the occurrence of
necking under uniaxial tension conditions (see
Lassance et al., 2006b).
Ductile Fracture in Metals 745
In contrast, more rounded voids involve a
much more abrupt transition: under f
p
*
no coa-
lescence takes place while above f
p
*
coalescence
is almost immediate without any stable void
growth stage. Another interesting outcome of
the calculations presented in Figure 55 is that
void coalescence, and thus ductile fracture, is
possible, if the volume fraction of particles is
sufficiently large, before the onset of necking,
that is, under purely uniaxial tension conditions
(e.g., Miserez et al., 2006). As explained in
Section 2.06.3.2.1, necking starts in the absence
of geometric or material imperfection (and rate
insensitive materials) when e =n. Hence, for
n =0.1 and n =0.3, fracture takes place before
necking if f
p
>23% and f
p
>17%,
respectively.
At larger stress triaxiality, typically larger
than 2, the effect of the shape becomes negligi-
ble (see Figure 39a) which validates afterward
many studies in the literature on crack propa-
gation (high stress triaxiality) performed with
the Gurson model without void shape effects
(e.g., Xia and Shih, 1995a, 1995b; Xia et al.,
1995; Ruggieri et al., 1996).
The analysis of the room-temperature ducti-
lity of 6xxx aluminum alloys provides a
practical illustration of the importance of the
initial void shape in industrial applications (see
Lassance et al., 2006a, for details). The material
consists of industrial direct-chill casts of
AA6060 aluminum alloys. The main micro-
structural feature regarding the damage and
fracture process consists of b-type elongated
intermetallic AlFeSi particles. A heat treat-
ment allows the transformation of the
b-particles into rounded a-intermetallics. The
amount of a-particles depends on the tempera-
ture and duration of the heat treatment. In situ
tensile tests within an SEM have shown that the
b-particles when oriented parallel to the main
loading direction break into several fragments
and when oriented perpendicular to the main
loading direction give rise to decohesion (see
Figure 28). The a-particles always give rise to
decohesion. This system provides a basis for
looking at the effect of the initial void shape
as it allows changing the initial shape of the
primary voids (by increasing the conversion of
b-particles into a-particles). Smooth round bars
with a diameter of 9 mm and a gage length of
40 mm were machined from the homogenized
logs parallel to the casting direction. Notches
were machined in some of the specimens, with
notch radii equal to 2 and 5 mm. These speci-
mens were loaded in tension. The hardening
laws were identified using an inverse procedure
from the tensile tests on smooth bars. Figure 56
shows the variation of the fracture strain (mea-
sured on the fracture surface) as a function of
the volume fraction of a-particles for various
stress triaxialities. As expected, the ductility
increases when increasing the volume fraction
of a-particles (involving the decrease of the
volume fraction of b-particles) due to their
more rounded shape. The predictions of the
constitutive model, assuming immediate
nucleation of all the voids, are also provided
in the figure. The agreement between the
experimental results and predictions is excellent
(see details in Lassance et al., 2006a). A key
point here is that all the parameters of the
model, that is, hardening law, initial porosities,
and initial void shape, were identified experi-
mentally without any adjustment or fitting
procedures.
The effect of the void distribution para-
meter l
0
is relatively difficult to grasp.
Figure 57 shows the variation of the ductility
with stress triaxiality for different l
0
. Here, dif-
ferent l
0
, for the same f
p
, mean different
relative void spacings
0
. This plot is interesting
because it approximately quantifies the maxi-
mum level of anisotropy in the ductility that can
be expected for a material with l
0
=a in one
symmetry direction, which, loaded in the ortho-
gonal direction, would be characterized by
l
0
=1/a. For instance, a metal with 1% of
particles exhibiting an anisotropic void distri-
bution characterized by a =2 will present a
ductility that might change by more than a
factor of 2 when tested in the two orthogonal
directions. Examples of anisotropic particle dis-
tributions can be found after severe plastic
deformation typical of many forming opera-
tions: oxide columns in extruded copper bars,
potassium bubble columns in tungsten wires,
and columns of broken intermetallic particles
in rolled aluminum sheets. Note that the effect
of particle clustering can be approximately cap-
tured by prescribing higher local values of l
0
.
As explained in Sections 2.06.3.5.3 and
2.06.3.5.4, the presence of secondary voids
which nucleate and grow in the ligament
between primary voids can significantly decrease
the fracture strain and lead to void sheet-type
fracture mechanism. Figure 58 shows the varia-
tion of the fracture strain as a function of the
volume fraction of second population, f
20
, for
different nucleation strains (the primary voids
nucleate immediately). Minute fraction of sec-
ondary voids causes a serious drop of the
ductility. In an application on a 6056Al alloy in
either T4 or T78 state, the presence of about 1%
dispersoids was shown to have a major effect on
the ductility (Gallais et al., 2006).
Figure 59 presents the effect of the flow
properties of the materials, that is, yield
strength and strain-hardening exponent on the
fracture strain for initially penny-shape voids,
746 Failure of Metals
considering various initial volume fractions of
particles (the results are similar with initially
rounded voids). All these results consider that
the voids are present from the beginning of the
loading. The strain-hardening exponent has a
moderate effect on the ductility while the yield
strength has no effect. Changing for instance
the strain-hardening exponent n from 0.1 to 0.2
by proper thermal treatment will not markedly
improve the ductility as such. But, of course, an
enhanced strain-hardening capacity will, in
practical structural loading conditions, delay
necking and shear banding which significantly
contributes to improving the ductility by post-
poning the rise of the stress triaxiality
(Figure 24). Those last effects are geometric
and not a result of an intrinsic influence of the
strain hardening on the damage evolution.
Note also that, in practice, strain hardening
can hardly be changed without affecting the
40 60 80 100 20 0
0
0.5
F
r
a
c
t
u
r
e
s
t
r
a
i
n
,
f
Uniaxial with necking: T ~ 0.5
Notch 1: T ~ 0.7
Notch 2: T ~ 0.9
AA6060
= 0.2 s
1
T
homog
= 585 C
-AlfeMnSi content (%)
1
1.5
Model results
Exp. results
.
100% 100%
Figure 56 Evolution of the calculated ductility (fracture strain) as a function of the a-AlFeMnSi content for
an aluminum alloys AA6060 deformed at 20
h
/
e
W
0
= 0.01
E /
0
= 500 n= 0.1
d
/
0
= 0
0
= 2
0
= 1
0
= 0.5
Figure 57 Variation of the ductility as a function of
the stress triaxiality for penny-shaped voids of aspect
ratio W
0
=0.01 and for three volume fractions of
particles (with no effect on the strength of the medium
in the framework of the present model), f
p
=0.1%, 1%,
and 10%, considering three different anisotropy
distribution parameters, l
0
=1/2, 1, and 2, with
E/o
0
=500 and n=0.1 (see Lassance et al., 2006b).
0
0.2
0.4
0.6
0.8
1
0 10
0
4 10
3
8 10
3
1.2 10
2
f
20
0.84
0.56
0.28
0
T = 1, W
0
= 1, n = 0.1,
0
/E = 500
f
10
= 1 10
3
No 2nd population
c
Figure 58 Variation of the fracture strain as a
function of the initial volume fraction of secondary
porosity for different nucleation strains, e
c
. The
volume fraction of the primary porosity
f
10
=1 10
3
, the stress triaxiality T=1, the initial
void aspect ratio W
0
=1, and the initial void
distribution parameter l
0
=1 (see Fabre` gue and
Pardoen, 2006).
Ductile Fracture in Metals 747
strength (one important exception is given by
aluminum alloys which show, at low tempera-
ture, a change of the strain-hardening capacity
without much variation of the yield stress).
The picture about the effect of the flow prop-
erties becomes very different when void
nucleation does not occur immediately but
takes place only when a critical stress o
d
is
reached in the particle or along the interface,
requiring thus moderate to large amount of
plastic deformation accumulation prior nuclea-
tion. Figure 60 shows the variations of the
nucleation strain e
d
and of the fracture strain
e
f
as a function of the stress triaxiality for a
material involving 1% of spherical hard parti-
cles giving rise to penny-shaped voids
(W
0
=0.01). Let us define the void growth
strain e
f
as the strain increment required to
bring freshly nucleated void to coalescence,
that is, e
g
=e
f
e
c
. Different void nucleation
stresses o
d
/o
0
=0, 4, 5, and 6 are analyzed,
with n =0.1. The increase of the ductility e
f
with increasing nucleation stress is smaller
than the increase of the void nucleation strain
e
c
. In other words, the void growth strain
decreases with increasing resistance to void
nucleation. The main reason for this effect is
that the mean spacing between particles in the
plane normal to the principal loading direction
decreases before void nucleation. The main
effect of the strain-hardening capacity will be
to accelerate the attainment of the nucleation
condition by raising the stress in the particle.
Hence, improving the strain-hardening capa-
city is beneficial for the ductility only when
the ductility is not controlled by the void
nucleation step.
The effect of the flowproperties on the nuclea-
tion of voids and on the resulting ductility has
been investigated in an application on quasi-
eutectic cast aluminum alloy by varying the
heat treatment without affecting the second-
phase particles (15% of Si spherical particles)
(see Huber et al., 2005, for more details). The
annealed state noted A has a low yield stress
(o
0
=87MPa) and the T6 state has a high yield
stress (o
0
=234 MPa) while the strain-hardening
exponents are similar. The sequence of events in
the damage accumulation process in both A and
T6 materials observed during in situ tensile test-
ing are gathered in Figure 61. The two materials,
A and T6, present a similar behavior: the first
step is the cracking of the silicon particles, giving
birth to penny-shaped voids which then grow
and coalesce. The only obvious difference
between the two materials is that in the T6
0.1
1
0 1 2 3
f
p
= 10%
f
p
= 1%
f
p
= 0.1%
h
/
e
W
0
=0.01
0
=1
E/
0
=500
d
/
0
= 0
n =0.3
n =0.2
n=0.1
n = 0.01
0 1 2 3
f
h
/
e
W
0
=0.01
0
=1
d
/
0
= 0 n =0.1
0.1
1
f
p
=10%
f
p
=1%
f
p
=0.1%
E/
0
=100
E/
0
=500
E/
0
=2500
(a)
(b)
Figure 59 Variation of the ductility as a function of
the stress triaxiality for penny-shaped voids of aspect
ratio W
0
=0.01 and for three volume fractions of
particles (which have no effect on the strength of the
medium in the framework of the present model),
f
p
=0.1%, 1%, and 10%, considering: a, different
strain-hardening exponents n=0.01, 0.1, 0.2, 0.3 with
E/o
0
=500 and l
0
=1; b, different ratios E/o
0
=100,
500, 2500 with n=0.1 and l
0
=1 (see Lassance et al.,
2006b).
0.01
0.1
1
0 1 2 3
f
o
r
c
h
/
e
W
0
= 0.01 f
p
=1%
0
=1
E /
0
= 500 n =0.1 E
p
/E =10
d
/
0
= 0
d
/
0
= 4
d
/
0
=5
d
/
0
=6
Figure 60 Variation of the ductility and void
nucleation strain as a function of the stress triaxiality
for a material involving 1% of spherical hard particles
giving rise to penny-shaped voids (W
0
=0.01) when a
nucleation stress o
d
is attained (Beremin criterion [59])
for various nucleation stresses o
d
/o
0
=0, 4, 5, 6, and
n =0.1 (see Lassance et al., 2006b).
748 Failure of Metals
sample, particle cracking occurs right from the
beginning of plastic yielding, whereas it occurs
much later in the softer material A. Tensile tests
on smooth and notched round bars were per-
formed on both A and T6 material samples.
The results are given in Figure 62 in terms of
the variation of the ductility as a function of the
stress triaxiality.
The damage evolution during the tensile tests
on the notched and smooth bars was modeled
using the void nucleation condition [60], void
growth law [88], and void coalescence criterion
[96]. The only free parameter to be calibrated
is the critical stress o
d
for void nucleation which
has been adjusted based on the fracture strain
of material A measured on broken smooth spe-
cimens loaded in uniaxial tension. The initial
porosity f
0
was related to the particle content
using f
0
=W
0
f
p
(see eqn [68] and W
0
was
imposed to be equal to 0.01. The calibration
provides a value of o
d
equal to 550 MPa. The
response of material T6 has been simulated
using exactly the same values, that is,
o
d
=550 MPa and W
0
=0.01, and the proper
flow properties. The comparison between the
experimental results and the model is given in
Figure 62. For the case of the T6 treatment,
o
d
=550 MPa leads to very early nucleation as
also observed experimentally. On the other
hand, modeling the A alloy without accounting
for delayed void nucleation, that is, using
o
d
=0, underestimates the ductility by more
than a factor of 2. In other words, depending
on the state of hardening in the matrix, the
ductility of these materials can be controlled
by both the nucleation and growth stages.
None of these two stages can be neglected.
In some materials, heterogeneities in the
microstructure leading to heterogeneties in
local mechanical properties are responsible for
the coexistence of different ductile failure modes.
The effect of the presence of both soft and hard
regions in a material can have a marked influ-
ence on the fracture strain due to the accelerated
damage rates induced inside the soft regions due
to the constraint imposed by the surrounding
harder zones. Many examples of alloys consist-
ing of two ductile phases are provided by Ankem
et al. (2006). In particular, in some aluminum
alloys, the microstructure consists of precipitate-
free zone (PFZ) along the grain boundaries
covered with large inclusions and a precipita-
tion-hardened state within the grain involving
also coarse intermetallic particles. As shown in
Figure 63, the failure mode of the material can be
either intragranular or intergranular ductile frac-
ture, or a combination of the two (e.g., Dumont,
2001; Pardoen et al., 2003).
A schematic of the microstructure is shown
in Figure 64a. The grain interior, after heat
treatment, has a high yield stress o
0g
and a low
strain-hardening rate n
g
. On the other hand, the
L
pz
L
px
2R
px
2R
pz
5 m
Figure 61 The schematics outline the damage events sequence. The micrographs were taken during in situ
tensile test at the final stage of deformation (see Huber et al., 2005).
0
0.2
0.4
0.6
0.8
0.4 0.8 1.2
Experimental
Model
f
T
T6
A
Figure 62 Variation of the ductility as a function of
the mean stress triaxiality, comparison of
experiments and modeling for (1) the A heat
treatment using f
p
=15%, W
0
=0.01, o
d
=6.15; (2)
the T6 heat treatment using f
p
=14%, W
0
=0.01,
o
d
/o
0
=6.15 (see Huber et al., 2005).
Ductile Fracture in Metals 749
PFZ has a low yield stress o
0p
and a high strain-
hardening rate n
p
. The idealized microstructure
is shown in Figure 64b with the various micro-
structural parameters entering the problem and
the relevant dimensionless quantities.
The competition between intergranular and
transgranular failure can be qualitatively under-
stood in the following way (see Figure 65). The
PFZ is soft and is thus the first to deform plas-
tically. The elastic grain imposes a strong
constraint on the PFZ involving a large stress
triaxiality. The large void growth rate in the PFZ
leads to rapid coalescence of the voids. However,
in some circumstances, the stress in the grain
reaches the yield stress before the onset of coa-
lescence in the PFZ. The stress triaxiality then
drops in the PFZ which, due to its higher strain-
hardening capacity, now imposes a higher con-
straint inside the grain. Voids then tend to grow
more rapidly within the grain. Due to the low
strain-hardening capacity of the grain, a state of
damage-induced softening is rapidly attained
until voids finally coalesce within the grain.
Experimentally, by increasing the time of
heat treatment, the yield stress first increases,
favoring the propensity toward intergranular
fracture, and then decreases, favoring again a
transgranular fracture mode. In parallel, the
strain-hardening rate decreases when precipita-
tion occurs. A more quantitative analysis of this
highly nonlinear problem of failure mode tran-
sition requires a detailed model for void growth
and coalescence to be incorporated in each
layer. In Pardoen et al. (2003), the PFZ and
the grain interior have been modeled using the
same void growth and coalescence constitutive
laws (Gologanu model combined by Thomason
coalescence criterion). The material was mod-
eled as a bilayer (see Figures 64b and 64c),
neglecting thus the effect of inclined grain
boundaries (see Scheyvaerts et al. (2006) for
more advanced representation of the micro-
structure). The initial relative spacing between
particle along the grain boundary, L
p0
/D
p0
, is
the most relevant parameter to interpret the
results. Figure 66 presents failure maps for
10 m
100 m
(a) (b)
Figure 63 Fractography of (a) intergranular ductile fracture and (b) intragranular ductile fracture in a 7xxx
alloy. From Dumont D. 2001. Relations Microstructures/Te nacite dans les alliages ae ronautiques de la se rie
7000. Ph.D. thesis, Institut National Polytechnique Grenoble. Pardoen, T., Dumont, D., Deschamps,
A., Brechet, Y. 2003. Grain boundary versus transgranular ductile failure. J. Mech. Phys. Solids 51, 637665.
Grain particles
Grain-boundary
particles
Grain damage
Grain-boundary
damage
Grain boundary
Precipitate-free zone
d
h
L
gx
L
gz
D
px
D
pz
L
px
D
gx
D
gz
n
g
E
f
g0
W
g0
n
p
E
0g
0p
g0
f
p0
W
p0
p0
(a) (b) (c)
Figure 64 Description of: a, the real microstructure and failure mechanisms; b, the idealized microstructure;
and c, the continuum micromechanical model. The parameters appearing in (b) and (c) are defined in Table 1
(see Pardoen et al., 2003).
750 Failure of Metals
three different ratios of o
0g
/o
0p
. As expected, an
increase of the grain yield stress promotes grain
boundary failure. An increase of the PFZ
strain-hardening capacity has an effect similar
to a decrease of o
0g
/o
0p
. The most important
parameter, as exhibited in Figure 67, is the spa-
cing relative to the PFZ thickness, o
p0
.
These failure maps show that, whatever the
flow properties and microstructure, lower par-
ticle spacing L
p0
/D
p0
and high stress triaxiality
always tend to promote intergranular fracture
as expected from the qualitative description of
Figure 65. The stress triaxiality failure mode
dependence has been qualitatively observed by
Dumont (2001). Realistic values for L
p0
/D
p0
are
between 2 and 5. In that range, the stress triaxi-
ality corresponding with the failure mode
transition is very much dependent on particle
spacing which also complicates the optimiza-
tion of the material properties.
2.06.3.7 Fracture Toughness of Thick
Ductile Metallic Components
Two sections are now devoted to the fracture
resistance of cracked structures. The first sec-
tion is limited to thick components or thick
test specimens which is the classical topic of
elastoplastic fracture mechanics. By thick, it
is meant that the thickness is larger than the PZ
size and that plane strain or near plane strain
conditions prevail, on average, along the crack
front line. We limit this presentation to mode I
fracture, which is the most important for a
majority of practical applications. The second
section is devoted to the fracture of thin
sheets, a subject which has recently received a
recrudescence of interest, motivated by an
increasing use in structural application of thin-
ner and thinner sheets made of more and more
ductile alloys. In thin sheets, the thickness
plays a key role in controlling the dissipation
of energy at the crack tip through its direct
effect on the energy spent in localized necking
and an indirect effect on the rate of damage
evolution in the FPZ. While in thick plane
strain samples, loss of constraint results from
in-plane effects, in thin plate one must also
consider out-of-plane effects. In thin plates,
mixed mode I and III fracture is also an impor-
tant issue.
In this section devoted to plane strain frac-
ture (thick components), we start by reviewing
some basics of elastoplastic fracture mechanics
necessary to address the cracking mechanisms,
also useful for the next section on thin sheet
fracture. Then, a second subsection addresses
the resistance to fracture initiation, which is
S
t
r
e
s
s
t
r
i
a
x
i
a
l
i
t
y
,
T
=
i
i
/
3
e
Relative particle spacing in PFZ, L
p0
/D
p0
Intergranular
fracture
Transgranular
fracture
Grain
PFZ
e
Grain
PFZ
e
Figure 65 Failure map providing a qualitative
understanding of the competition between
intergranular and transgranular ductile failure (see
Pardoen et al., 2003).
0
1
2
3
4
4 8 12 16 20
T
L
p0
/D
p0
n
p
=
0.4, W
p0
=
1
p0
=
1, R
0
=0.2
0g
/
0p
=4
0g
/
0p
=5
0g
/
0p
=6
Intergranular
fracture
Transgranular
fracture
Figure 66 Effect of the yield stress ratio on the
failure mode in a stress triaxiality vs relative particle
spacing map (see Pardoen et al., 2003).
0.5
1
1.5
2
2.5
3
4 8 12 16 20
p0
=
1
p0
=
3
p0
=
1/3
T
L
p0
/D
p0
Intergranular
fracture
Transgranular
fracture
0g
/
0p
=5, n
p
=0.4
W
p0
=
1, R
0
=
0.01
Figure 67 Effect of the PFZ thickness/void spacing
ratio on the failure mode in a stress triaxiality vs
relative particle spacing map (see Pardoen et al.,
2003).
Ductile Fracture in Metals 751
usually (but not always) the definition of the
fracture toughness of a material, while the third
subsection focuses on the resistance to ductile
tearing (crack propagation).
2.06.3.7.1 Basics
(i) Essentials of elastoplastic fracture
mechanics
The theoretical foundations of these concepts
have been presented in details in Chapter 2.03
(see Broberg, 1999; Anderson, 1995). In thick-
walled components or relatively thick labora-
tory samples, the zone in front of the crack tip
in which the fracture phenomena take place
undergoes plane strain conditions. The crack-
tip loading is quantified through the value of
the J-integral defined by (Rice, 1968)
J =
_
W
V
n
x
n
i
s
ij
0u
j
0x
_ _
ds [120[
where G is a contour surrounding the crack tip
in the anticlockwise direction, W
V
(e
ij
) is the
strain energy density, and u
i
are the displace-
ments. Rice (1968) demonstrated that the value
of the integral [120] is independent of the path
of integration G. Assuming that a nonlinear
elastic behavior provides an adequate descrip-
tion of the mechanical response (which is true
under radial or approximately radial loadings),
J provides a unique intensity measure for char-
acterizing the crack-tip loading under yielding
conditions. The analysis of the crack-tip stress
fields within the context of small strain defor-
mation theory of plasticity has been performed
by Hutchinson (1968) and Rice and Rosengren
(1968). It is known as the HRR solution. A
power law relation of the form
e
e
0
= a
s
s
0
_ _
N
[121[
is assumed to represent the uniaxial flow proper-
ties of the material (a is a constant and N=1/n is
the inverse of the strain-hardening exponent n).
The asymptotic solution for the stress and strain
fields is given by
s
ij
= s
0
J
as
0
e
0
I
N
r
_ _
1,(N1)
s
ij
(y. N) [122[
s
e
= s
0
J
as
0
e
0
I
N
r
_ _
1,(N1)
s
e
(y. N) [123[
e
ij
= ae
0
J
as
0
e
0
I
N
r
_ _
N,(N1)
e
ij
(y. N) [124[
u
i
= ae
0
r
J
as
0
e
0
I
N
r
_ _
N,(N1)
u
i
(y. N) [125[
where I
N
is a dimensionless constant which
depends on N, and o~
ij
, o~
e
, e ~
ij
, and u~
i
are functions
of 0 and N. The constant I
N
and functions o~
ij
, o~
e
,
e~
ij
, and u
i
have been computed and tabulated by
Shih (1983) for both plane stress and plane strain
conditions. In the elastic case, that is, N=1, eqn
[122] reduces to the inverse square root singular-
ity of linear-elastic fracture mechanics. The level
of J determines the crack-tip stress field in the
nonlinear elastic cases.
According to the displacement fields [125],
the opening separation of the two crack faces
varies like r
1/(N1)
as r 0. The opening of the
crack at r 0 is thus zero. Rice (1968) has
suggested to define the opening separation of
the two crack faces by taking the opening
c =2u
y
at the intercepts of the two 45
lines
drawn back from the tip of the deformed pro-
file, that is where r u
x
=c/2 (Figure 68). This
local opening separation, c, called the crack-tip
opening displacement (CTOD), can be mea-
sured experimentally on a section transverse to
the crack front. The use of the HRR solution
[125] implies that
d = d(ae
0
. N)
J
s
0
[126[
where d has also been tabulated by Shih (1983)
as a function of ae
0
and N. For typical metal
alloys, d ranges between 0.3 and 0.6 in plane
strain and between 0.5 and 0.9 in plane stress.
The detailed analysis of the crack-tip blunt-
ing process in the framework of finite strain J
2
plasticity has been made first by McMeeking
(1977) using the finite element method
(extending the slip line analysis of Rice and
Johnson (1970) limited to perfectly plastic
materials). The evolution of the key parameter
controlling the ductile damage process, that is,
the stress triaxiality, is shown in Figure 69a as a
function of the distance to the crack tip normal-
ized by J/o
0
. As shown in Figure 69b, the
maximum stress triaxiality is equal to 2.75 for
perfectly plastic materials (n =0) and increases
with increasing strain-hardening exponent.
These values of the stress triaxiality are
correctly predicted by Hutchinson and Rice
and Rosengren (the so-called HRR theory).
The stress triaxiality drops down near the
y
x
Initial crack
r
u
x
u
y
45
Figure 68 Definition of the CTOD.
752 Failure of Metals
blunted crack tip due to the presence of the
free surface and reaches the plane strain tension
value T =
3
_
,3 ~ 0.58. The plastic strains
become larger than a few percent at a distance
equal to about 13 CTOD c. The FPZ will
thus extend in a zone equal to roughly 1 to
3c. The plastic strains up to a value equal to
0.10.2 are correctly predicted by the HRR
solution.
The value of J corresponding to the initiation
of cracking is noted in mode I J
I
and the corre-
sponding critical CTOD is noted c
c
. Although
its validity is not guaranteed anymore, the
J-integral is still used to quantify the loading
during the propagation of the crack, at least for
propagation over small distances (typically a
few millimeters). The tearing resistance of the
material is thus quantified through the evolu-
tion of J with crack advance, Da. A typical J
R
curve is shown in Figure 70. An engineering
fracture toughness is defined by the value of J
after 0.2 mm of crack advance, noted J
Ic
. In
some materials, J
0.2
and J
Ic
can be very differ-
ent. In this chapter, we do not distinguish
between these two definitions. The fact that J
keeps increasing with the loading is extrinsic to
the fracture process taking place in the near-
crack-tip region. It is thus not a true increase of
the fracture resistance (see Cotterell and
Atkins, 1996). J increases due to the plastic
dissipation taking place in the crack wake due
to the progressive elastic unloading, due to the
nonradial loadings in the active PZ of a propa-
gating crack, and due to changes in the crack-
tip geometry (see further). In very large sam-
ples, a steady-state regime should in principle
be reached but it is almost never observed in
metallic samples because of specimen size
limitations.
The analysis of the stress and strain conditions
in front of a growing crack in a plastically
deforming solid has been made by Rice and
Sorensen (1978) (see also Hutchinson (1974) for
the case of steady-state conditions). Under SSY
conditions and for an elasticperfectly plastic
material, the main difference in the predicted
stressstrain field between a stationary and a
growing crack lies in the strain singularity and
not the stress profile at the crack tip. This is an
important point for the analysis of the brittle-to-
ductile transition which is discussed in more
details in Section 2.06.4. Many observations
have shown a major difference in crack-tip geo-
metry between initiation of crack growth and
crack propagation. This is illustrated in
Figure 71. At cracking initiation, the crack tip
is blunted while during propagation the unzip-
ping process from one inclusion to another one
gives rise to a crack tip which is much sharper.
This modification in crack profile leads to an
elevation of the normal stress ahead of the
crack tip when the crack propagates (this effect
is not accounted for in the theoretical analysis
by Rice and Sorensen, 1978): the higher this
increase of the normal stress, the steeper the
slope of the J resistance curve. This phenom-
enon will also be a key in the analysis of the
brittle-to-ductile transition (see Section 2.06.4).
(a)
(b)
3 4 5 1 6 2
0.050.1
p
0.577
Stress triaxiality
Loss of
constraint
Increasing
n
Constraint
changes
r /
Fracture process zone
0
1
2
3
4
5
6
0 0.05 0.1 0.15 0.2 0.25 0.3
S
t
r
e
s
s
t
r
i
a
x
i
a
l
i
t
y
n
Plane stress
Plane strain
Figure 69 Variation of the stress triaxiality as a
function of (a), the distance to the crack tip
(supplemented by the effective plastic strain
evolution); (b), the strain-hardening exponent (for
plane strain and plane stress conditions).
a 0.2 mm
J
i
J
Loss of constraint
J
Ic
Steady-state regime
Active plastic
zone
Crack wake
SS
B
l
u
n
t
i
n
g
l
i
n
e
Figure 70 Schematic J
R
curve and PZ extension
during crack propagation.
Ductile Fracture in Metals 753
The J
R
curve is very sensitive to slight changes
of constraint which affect a lot the extrinsic
plastic dissipation (see, e.g., Sumpter, 1993;
Brocks and Schmitt, 1994; Anderson, 1995; Xia
et al., 1995; Xia and Shih, 1995a, 1995b). The
constraint effect can be quantified in the frame-
work of fracture mechanics through either the
so-called T-stress (only as long as SSY condi-
tions prevail) or the so-called Q-stress, which
quantifies the departure from the HR solution
(ODowd and Shih, 1991, 1992). A good review
about the fracture mechanics approach of con-
straint is given in the book by Anderson (1995)
and in Chapter 2.03.
To sum up, the work of fracture can be writ-
ten for plane strain conditions as
=
0
p
[127[
where G
0
=J
1c
and G
p
is highly dependent on
the level of constraint (the question whether
G
0
is dependent on the constraint will be dis-
cussed in the next subsection as it requires to
address the fracture mechanisms inside the
FPZ).
(ii) Computational strategies to simulate crack
propagation in ductile materials
There are four main types of modeling stra-
tegies which have been proposed in the
literature to address the prediction of the frac-
ture resistance, that is, in order to couple the
damage mechanisms and the mechanical fields
described just before. In all these approaches,
the key is that a characteristic length must be
introduced. The energy dissipation scales with
the height of the zone in which the fracture
mechanisms take place. This size is directly
related to the void spacing or to the spacing
between void clusters.
1. The first strategy consists in using the
void nucleation, growth, and coalescence laws
described in Sections 2.06.3.3, 2.06.3.4, and
2.06.3.5, and integrate them using the solutions
of the elastoplastic crack problems (see
Figure 69a), thus neglecting the coupling
between the mechanical fields and the damage
evolution. This approach allows relating the
cracking initiation toughness to the microstruc-
ture and also allows the simulation of crack
propagation using finite element methods with
proper node release technique (see pioneering
work by dEscatha and Devaux (1979), who
used a critical void radius condition to be
attained at a distance X
0
from the crack tip).
2. The second strategy is shown in
Figure 72b: finite element calculations can be
performed with the voids explicitly modeled
using a refined finite element mesh (e.g.,
Aravas and McMeeking, 1985a, 1985b;
Needleman and Tvergaard, 1987, 1991; Hom
and McMeeking, 1989a, 1989b; McMeeking,
1992; Tvergaard and Needleman, 1992;
Ghosal and Narasimhan, 1996; Gao et al.,
1996; Yan and Mai, 1997; Tvergaard and
Hutchinson, 2002; Kim et al., 2003; Gao et al.,
2005; Petti and Dodds, 2005b). These analyses
accurately model the growth and coalescence
process while properly accounting for the
length scale introduced by the void spacing. In
order to simulate the propagation of the crack,
this approach still requires a criterion for the
final failure of the intervoid ligament to simu-
late crack propagation, for example, by
modeling shear localization of the ligament
due to a second population of smaller voids as
in Needleman and Tvergaard (1987), or by pre-
scribing a critical void spacing as in Tvergaard
and Hutchinson (2002). This approach is com-
putationally intensive. It is only able to account
for a few voids ahead of the crack tip and thus
probably not attractive for simulating full
structures or test specimens, especially in 3-D.
Nevertheless, such simulations provide very
useful results for assessing the validity of the
assumptions and the predictions of the other
approaches.
3. The third strategy pursued mainly by
groups in France, Germany, the UK, and the
US employs a constitutive model, such as the
a
bl a
tear
a
tot
a
0
2
5
0
m
(b)
(a)
200 m
Figure 71 Ductile crack initiation and crack growth
from a fatigue precrack: a, in copper (see Pardoen
and Delannay, 2000); b, A 508 RPV steel. The crack
propagates from one inclusion to another one
leading to a zigzag pattern (Lautridou, 1980).
754 Failure of Metals
Gurson or the Rousselier model, which accounts
for the damage-induced softening, see Figure 72c
(e.g., Mudry et al., 1989; Rousselier et al., 1989;
Bilby et al., 1993; Xia et al., 1995; Xia and Shih,
1995a, 1995b; Brocks et al., 1995a; Ruggieri
et al., 1996; Gao et al., 1998a; Koppenhoefer
and Dodds, 1998; Zhang et al., 2000;
Roychowdhury and Narasimhan, 2000; Rivalin
et al., 2001a, 2001b; Pardoen and Hutchinson,
2003; Chabanet et al., 2003; Ne` gre et al., 2003,
2004, 2005). The constitutive model is implemen-
ted in a finite element code to simulate the
initiation and growth of the crack. A microme-
chanics-based damage model, such as the
Gurson model (and its extensions), is derived in
such a way that it should adequately reproduce
the behavior of a material cell involving a single
void subjected to homogeneous conditions at the
boundaries. Here, however, near a crack tip,
strong strain and stress gradients develop at the
scale of the void cell size. These gradients are
averaged in a relatively crude way by only using
a single element to represent one void cell. The
error coming fromthis approximation is difficult
to evaluate a priori.
As explained above, this approach requires
the introduction of a length scale in the model
related to the spacing between voids. This is
usually accomplished by tying the element size
to the void spacing, calibrated on experimental
crack growth data. This simple approach has the
disadvantage to artificially tie a physical length
to a numerical parameter. Hence, teams have
been working, motivated by seminal contribu-
tions in the field of fracture in concrete (e.g.,
Pijaudier-Cabot and Bazant, 1987), on formu-
lating nonlocal constitutive models for ductile
fracture (e.g., Leblond et al., 1994; Tvergaard
and Needleman, 1995; Engelen et al., 2003;
Geers et al., 2003; Reusch et al., 2003a, 2003b;
Zhenhuan et al. 2003; Geers, 2004; Yuan and
Chen, 2004; Mediavilla et al., 2006a, 2006b,
2006c). There are several methods for introdu-
cing internal lengths into the model and no
consensus has emerged yet about the best
approach.
4. A fourth strategy, schematically illu-
strated in Figure 72d, initiated by Tvergaard
and Hutchinson (1992), makes use of cohesive
zone surfaces to simulate the fracture process in
ductile metals (e.g., Tvergaard and Hutchinson,
1996; Keller et al., 1999; Siegmund and Brocks,
1999, 2000; Li and Siegmund, 2002; Roy and
Dodds, 2001; Roychowdhury et al., 2002;
Brocks et al., 2003; Chen et al., 2003; Cornec
et al., 2003; Scheider and Brocks, 2003; Chen
(b)
Large inclusions
Carbides
Active layer
(ductile tearing)
(a)
Voids
Figure 72 (Continued)
Ductile Fracture in Metals 755
et al., 2005; Chen and Kolednik, 2005; Ne` gre
et al., 2005; Scheider et al., 2006). The response
of the FPZ is approximately modeled by a trac-
tionseparation curve (see Figure 72d). The
main characteristics of the tractionseparation
curve are the work of separation and the max-
imum stress, also called cohesive stress or peak
stress. The main advantage of this method is
that the characteristic length is introduced in a
natural way into the model since the cohesive
properties involve the work of separation and
the cohesive stress. However, problems are
encountered when introducing dependencies of
the cohesive zone parameters on the mechanical
fields next to the cohesive surfaces (Tvergaard
and Hutchinson, 1996; Keller et al., 1999) in
order to artificially account for constraint
effects on the damage evolution. Also, in
regular finite element implementations, the
crack path must be prescribed in advance pre-
venting thus the modeling of complex crack
paths.
Even though these various modeling
approaches have all attained a relatively high
level of maturity, comparisons with experimen-
tal data are still extremely scarce. Nevertheless,
encouraging results have been demonstrated
mostly using strategies 3 and 4 (see Chapter
7.05).
c
1
1
1
c
2
c
1
1
0
Trilinear
function
c
1
1
0
Polynomial
function
Exponential
function
f
E
f
0
Cell element with void
Layer of void-containing cell elements
Crack
D
D
D/2
(c)
(d)
Figure 72 The three coupled modeling strategies referred in the text as methods 2, 3, and 4 described schematically
in (b)(d) in order to simulate cracking in ductile materials represented in (a); b, discrete voids modeling; c,
computational cell model; d, cohesive zone model approach with the three typical traction separation laws used in
the literature. a, From Gullerud, A. S., Gao, X., Dodds, R. H., Jr. Haj-Ali, R. 2000. Simulation of ductile crack
growth using computational cells: Numerical aspects Eng. Fract. Mech. 66, 6592. b, From Tvergaard, V. and
Hutchinson, J. W. 1992. The relationship between crack growth resistance and fracture process parameters in elastic
plastic solids. J. Mech. Phys. Solids 40, 13771397. c, From Gao, X., Faleskog, J., and Shih, C. F. 1998a. Cell model
for nonlinear fracture analysis. II: Fracture-process calibration and verification. Int. J. Fract. 89, 374386.
756 Failure of Metals
2.06.3.7.2 Fracture initiation toughness
The resistance to crack initiation of pre-
existing sharp crack provides a measure of the
so-called fracture toughness. It provides, for
material scientist, a very useful way to index
the quality and compare materials with respect
to their ability to resist cracking. The resistance
to cracking initiation constitutes also, for many
applications, the failure conditions on which
structural integrity assessment methods are
based. In this section, we are mainly interested
in understanding and predicting the fracture
toughness at cracking initiation in ductile
metals through its relationship with the micro-
structure and flow properties.
Figure 73 depicts the model envisioned for
simulating cracking in an idealized ductile
metal. The initial geometry is a precrack of
opening c
0
in an idealized material having reg-
ularly distributed voids with initial spacing X
0
.
The crack is long, and SSY is assumed to apply.
The matrix is characterized by the following
mechanical properties E, v, o
0
, and n. The
voids have an initial shape W
0
, volume fraction
f
0
, and distribution parameter l
0
.
The predicted evolutions of the fracture
toughness reported hereafter have been
obtained by using the third type of modeling
strategy (computational unit cell, see
Figure 72c) using the combination of the
Gologanu and Thomason models (see Sections
2.06.3.4.4 and 2.06.3.5.4) and under SSY con-
ditions in order to avoid any complications
related to constraint effects (see Pardoen
and Hutchinson, 2003, for details). The size of
the elements is directly related to void spacing
in the portion of the mesh that experiences
void growth and coalescence, that is, in the
FPZ. In all cases here, the elements in the FPZ
are taken to be square and of dimension X
0
.
Thus, normalization of the toughness by the
only length scale X
0
obviously leads to results
that are independent of the degree of mesh
refinement. Dimensional analysis shows indeed
that
J
Ic
s
0
X
0
= F
s
0
E
. n. f
0
. W
0
. l
0
. s
d
_ _
[128[
The validity of this relationship has been
confirmed experimentally, for instance by
Lautridou and Pineau (1981) (see also the
recent work by Miserez et al., 2006). The results
of the simulations have been successfully com-
pared to the results of Tvergaard and Hutchinson
(2002) and Gao et al. (2005), obtained with the
second modeling strategy. A few selected results
are presented hereafter. Note that void nucleation
is not considered (i.e., o
d
=0).
As shown in Figure 73, two limiting situations
can be found (Tvergaard and Hutchinson, 2002;
Pardoen and Hutchinson, 2003):
1. At sufficiently high porosity, the void near
the tip is influenced by its nearest neighbor,
which experiences almost the same rate of
growth. The interaction among the voids,
including voids even farther from the tip,
results in significantly higher rate of void
growth for all of the voids. Coalescence
between several voids and with the crack starts
early, almost simultaneously. This is the mul-
tiple void interaction mechanism.
2. For sufficiently small void volume frac-
tion, a single void process prevails, which is
essentially the process envisioned by the Rice
Johnson (1970) model. The void nearest to the
tip grows with little influence from its nearest
neighbor further from the tip. This is the void-
by-void growth mechanism. Experimental data
X
0
Z
0
2R
0
2R
z0
0
I. Multiple void interaction
Finite
strain
zone
II. Void by void growth
Plastic localization
= coalescence
= finite strain zone
Plastic localization
= coalescence
Figure 73 The initial geometry of a precrack in a ideal material with regularly distributed spheroidal inclusions;
sketch of the two ideal modes of crack initiation, that is, multiple void process and single void/crack process.
Ductile Fracture in Metals 757
on HSLA steels (Luo et al., 1989) show, from
local strain measurements, strains of about
0.50.75 in front of the crack tip at cracking
initiation, a value that would never be attained
with the multiple void interaction mechanism
which involves typical fracture strains of about
0.1. Most metallic alloys have initial void
volume fraction smaller than 10
2
and will
thus fail from void by void growth mechanism.
Whatever the mechanism, initially spherical
voids get first oblate (see Aravas and
McMeeking, 1985a). In the single void interac-
tion problem, if the initial porosity is very low,
the voids enter the low stress triaxiality zone
(see Figure 69) before coalescing with the
crack. In that case, the voids tend to elongate
before cracking initiates.
Figure 74 shows the variation of the fracture
toughness (normalized by o
0
X
0
) as a function of
the initial porosity, for various (a) ratios o
0
/E,
(b) strain-hardening exponents n; and (c) initial
void aspect ratios W
0
. The ratio o
0
/E has no
effect on the fracture toughness as long as void
nucleation is not considered. Figure 74b exhi-
bits the effect of the strain-hardening index, n,
on the fracture toughness. Strain hardening has
a major influence on fracture toughness. The
fact that the fracture toughness is linearly pro-
portional to the yield stress seems to contradict
much experimental evidences showing that the
toughness (J
Ic
) of a family of alloys usually
decreases with increasing yield stress (o
0
).
Several points are relevant to this apparent con-
tradiction. First, in many alloys, an increase of
o
0
by metallurgical intervention is usually
accompanied by a decrease of the strain-hard-
ening index n that has the opposite effect on the
toughness. For instance, these countervailing
trends occur in precipitation hardening of alu-
minum alloys, where the precipitates do not, in
general, take part in the failure process. Second,
the present model does not incorporate void or
microcrack nucleation criterion. In many
instances, an increasing yield strength will
affect the nucleation stage by raising the stress
on the second-phase particles or grain bound-
aries. A larger yield stress may also favor
nucleation on smaller particles or on a second
population of particles at an earlier stage of the
deformation. However, if all other parameters,
including strain hardening, can be kept con-
stant, a higher yield stress directly implies a
higher fracture toughness. A good example is
given by the decrease with increasing tempera-
ture of the fracture toughness of ferritic steel in
the upper shelf region (see Section 2.06.4). For
the typical temperature range covered when
measuring a ductilebrittle transition curve,
no modification of microstructure and
hardening mechanisms is expected, except for
the decrease of the yield stress with increasing
temperature.
Figure 74c shows that the effect of the initial
void shape follows the intuition: at a given
porosity, prolate voids have a smaller area frac-
tion projected onto the fracture plane than
spheres and conversely for oblate shapes.
Thus, prolate shapes increase J
Ic
/o
0
X
0
while
oblate shapes reduce it relative to spherical
voids at the same volume fraction. The results
0
2
4
6
8
(a)
10
6
10
5
10
4
10
3
10
2
10
1
0
/E = 0.01
0
/E = 0.003
0
/E = 0.001
f
0
J
I
c
/
0
X
0
n = 0.1, W
0
= 1,
0
= 1
(b)
10
5
10
4
10
3
10
2
f
0
J
I
c
/
0
X
0
0
2
4
6
8
10
0.2
n =
0.1
0.01
0
/E = 0.003, W
0
= 1,
0
= 1
(c)
10
5
10
4
10
3
10
2
f
0
J
I
c
/
0
X
0
0
2
4
6
8
1/3
1
W
0
=
1/10
3
10
0
/E = 0.003, n = 0.1,
0
= 1
Figure 74 Variation of the fracture toughness
(normalized by o
0
X
0
) as a function of the initial
porosity, for various: a, ratios o
0
/E; b, strain-
hardening exponents n; and c, initial void aspect
ratios W
0
(Pardoen and Hutchinson, 2003).
758 Failure of Metals
of Figure 74c can be used to guide understand-
ing of, as well as to predict the variation of, the
fracture toughness as a function of the loading
direction for rolled plates with preferential
orientation of the second phase. Clearly, void
shape has a significant effect on fracture tough-
ness. The effect of anisotropic void
distributions (l
0
not equal to 1) has also been
investigated by Pardoen and Hutchinson (2003)
and turns out be also very significant.
Finally, it is important to recognize that the
previous analysis is valid only for tensile locali-
zation mode leading to internal necking
between voids. However, localization in shear
is sometimes observed at the crack tip of high-
strength/low-hardening materials. Examples of
shear coalescence in an FPZ under plane strain
condition are provided by observations of
zigzag cracking in low-hardening steels
(Clayton and Knott, 1976; Needleman and
Tvergaard, 1987; Xia and Shih, 1995b). The
early shear localization process between the
blunted crack tip and the nearest void is detri-
mental to the fracture toughness as it involves
much less plastic work than a full void growth/
coalescence to final impingement.
The most simple method to estimate the frac-
ture toughness at crack initiation is the
aforementioned method 1, which requires to
integrate the constitutive model with the
mechanical fields evaluated for a crack in a
nondamaging elastoplastic material. An even
simpler approach is to consider only the stress
triaxiality as the dominant feature controlling
the damage process. Figure 75 shows the
variation with the imposed stress triaxiality of
the work per unit area spent in deforming a
material element involving an initial volume
fraction f
0
=0.01 of spherical voids up to frac-
ture. For that specific f
0
, the high stress
triaxiality work of fracture G
0
calculated this
way is about a factor of 2 smaller than the
fracture toughness calculated with the full com-
putational cell model (see Figure 74). The
important message to extract from Figure 75 is
that the work of fracture is relatively indepen-
dent of the stress triaxiality as long as the stress
triaxiality is larger than typically 22.5. This
point has also been raised by Siegmund and
Brocks (2000). For thick component, typical
loss of constraint encountered with small speci-
men sizes or short crack lengths does not
usually induce a drop of the stress triaxiality
under a value equal to 22.5. This explains
why, again for plane strain conditions, con-
straint effects mainly affect the tearing
modulus and not much the initiation of crack-
ing. Nevertheless, under large-scale yielding
conditions, the magnitude of plastic strains
near a crack tip can sometimes depend on the
specimen geometry, resulting in differences in
the crack initiation toughness (see Pardoen
et al., 2000).
2.06.3.7.3 Ductile tearing resistance
For several applications, the presence of
cracks is tolerated, for example, in some aircraft
structural components or pipelines, and the
integrity of structure is assessed toward
unstable crack propagation. It is thus essential
to develop materials with the highest possible
tearing resistance as well as to develop models
that allow transferring resistance curves
obtained on laboratory specimens to complex
structures undergoing realistic complex loading
conditions.
The first comprehensive effort to predict
crack resistance curves based on an embedded
FPZ model reproducing the response of a row
of voids during deformation, has been pro-
posed by Tvergaard and Hutchinson (1992)
using the cohesive zone methodology (see
above, method 4), SSY, and plane strain
conditions. These authors have shown that the
key factor controlling the dissipation of energy
during crack propagation is the cohesive
strength. In their work, the tearing resistance
was quantified by the ratio G
SS
/G
0
where G
SS
is
the steady-state work of fracture (see
Figure 70). Figure 76 shows the variation of
G
SS
/G
0
as a function of the cohesive strength
for different strain-hardening exponents. If
the peak strength is lower than about 3,
G
SS
/G
0
~1, that is, the J
R
curve is flat and
there is no other dissipation than the work of
fracture spent in the FPZ, G
0
. When the
cohesive strength increases, plastic strains
must accumulate next to the FPZ in order to
raise the stress up to the cohesive strength.
This stress increase is made possible owing to
0
0.5
1
1.5
2
0 1 2 3 4 5
0
/
0
X
0
Stress triaxiality, T
f
0
= 0.01; n = 0.1;
0
/E = 500
Typical stress triaxiality for
thick cracked specimens with
possible constraint effects
Figure 75 Variation of the work of fracture per unit
area (normalized by o
0
X
0
) as a function of the stress
triaxiality of a material element involving an initial
volume fraction f
0
=0.01 of spherical voids.
Ductile Fracture in Metals 759
the strain-hardening capacity of the material.
Hence, the magnitude of the plastic defor-
mation as well as the size of the PZ increase
with increasing load, leading to an increasing
contribution of the plastic dissipation
terms G
p
(see eqn [127]). The value of the
cohesive strength is representative of the values
of the parameters controlling the damage
process.
The first comprehensive effort to predict
crack resistance curves based on the more rea-
listic computational cell methodology
(referred as method 3 above) was proposed by
Xia and Shih (1995a, 1995b, 1996) and Xia
et al. (1995) for 2-D plane strain conditions. In
their application of the computational cell
method, void growth remains confined to a
single layer of material symmetrically located
about the crack plane and having a thickness D,
where the parameter D is identical to the
parameter X
0
used before. This layer consists
of cubical cell elements with dimension D on
each side; each cell contains a centered spherical
cavity of initial volume fraction, f
0
. Progressive
void growth and subsequent macroscopic mate-
rial softening in each cell are usually described
with the version of the Gurson model extended
by Tvergaard (see Section 2.06.3.4). When the
calculated void volume fraction, f, in the cell
adjacent to the crack tip reaches a critical value,
f
c
, the load is ramped down linearly with the
increase of the normal displacement. This pro-
duces the growth of the crack tip in discrete
increments of the cell size. Figure 77 is a selec-
tion of predicted crack resistance curves for (a)
different initial porosities f
0
, (b) different strain
hardening exponents n, and (c) different ratios
o
0
/E.
0
0
1 2
2
3 4
4
5 6
6
8
10
n = 0 0.1 0.2
/
y
s
s
/
0
Figure 76 The extrinsic plasticity contribution to
plane strain steady-state toughness as predicted by
the cohesive zone model for mode I crack growth in a
ductile solid with tensile yield stress o
y
and strain-
hardening exponent n. The curves give the ratio of
the steady-state macroscopic work of fracture to the
intrinsic work of separation as a function of the ratio
of peak separation stress to yield stress. From
Tvergaard, V. and Hutchinson, J. W. 1992. The
relationship between crack growth resistance and
fracture process parameters in elastic plastic solids.
J. Mech. Phys. Solids 40, 13771397.
0
0
2
4
6
8
10
12
14
16
18
20
5
(b)
(a)
10
a /D
a /D
15 20
/
(
D
0
)
f
0
= 0.005
n = 0.3
n = 0.2
n = 0.1
n = 0.05
/
(
D
0
)
E/
0
= 500
E/
0
= 400
E/
0
= 300
E/
0
= 200
0
0
2
4
5
(c)
10 15 20
f
0
=
0.005
n = 0.1
0
0
2
4
6
8
10
12
n = 0.1
5 10 15 20
a /D
0
= 0.02
0
= 0.01
0
= 0.005
0
= 0.0025
0
= 0.001
/
(
D
0
)
Figure 77 Crack resistance curves for: a, different
initial porosities f
0
; b, different strain-hardening
exponents n; and c, different ratios o
0
/E. From
Xia, L. and Shih, C. F. 1995a. Ductile crack growth. I:
A numerical study using computational cells with
microstructurally based length scales. J. Mech. Phys.
Solids 43, 233259.
760 Failure of Metals
As already shown in Figure 38, the maximum
stress that a voided ductile solid can attain
under a given stress triaxiality increases with
decreasing initial porosity. This maximum
stress is the peak stress used in a cohesive zone
type model. The result of Figure 77a can thus be
directly related to the predictions of Figure 76
by Tvergaard and Hutchinson (1992). A lower
initial porosity has a strong effect on the tearing
resistance by affecting the cohesive strength of
the FPZ. Increasing the strain-hardening capa-
city or decreasing the yield strength tends to
increase the PZ size, hence the plastic dissipa-
tion. Note again in Figure 77c, that the fracture
initiation is not affected by the ratio o
0
/E, in
agreement with Figure 74a.
Xia and Shih (1995a, 1995b) have also used
the computational cell simulations to address
constraint effects. Figure 78 shows crack resis-
tance curves corresponding to different level
of constraints imposed through varying the
T-stress values (see Chapter 2.03 for the defini-
tion of the T-stress), still using SSY conditions.
Loss of constraints are associated to negative
T-stress. The constraint effect on the tearing
resistance is very clear, while, in agreement
with the conclusions drawn from Figure 75,
cracking initiation is not much affected.
Figure 79 shows crack resistance curves belong-
ing now to three-point bending laboratory
specimens. The three different specimens are
homothetic with three different width W.
When decreasing the specimen size, the con-
straint moves from SSY to large-scale yielding
conditions. In the small specimens, the PZ
interacts with the specimen boundaries leading
to significant loss of constraint.
Although the number of comparisons with
experimental data remains limited (see
Chapter 7.05), these crack resistance curve pre-
dictions have granted a considerable success to
the so-called local approach (Chapter 7.05) or
top-down approach to ductile fracture
(Hutchinson and Evans, 2000) in the commu-
nity dealing with the fracture integrity of
structural components under large-scale yield-
ing conditions. The success of this approach is
underpinned by the requirement that the
microstructural parameters (the void volume
fraction, void spacing, etc.) must be set such
that the model reproduces experimental crack
data for specific specimens (see Chapter 7.05).
Once calibrated, these approaches have accu-
rately accounted for a wide range of constraint
effects. Three-dimensional aspects of crack
initiation and growth have also been simulated,
as explained in Chapter 7.05. In that case, the
comparisons with experiments are even more
scarce (see Gao et al., 1998a; Rivalin et al.,
2001b; Chabanet et al., 2003; Ne` gre et al.,
2005).
2.06.3.8 Fracture Resistance of Thin Metallic
Sheets
2.06.3.8.1 Introduction to the fracture
mechanics of thin metallic sheets
This section extends the presentation given in
Section 2.06.3.7.1 of the elastoplastic fracture
mechanics basics to specific aspects related to
cracked thin sheets. A plate is considered to be
thin if the PZ size during cracking is larger
than the thickness, thus preventing plane strain
conditions to build up next to the crack tip.
0
0
1
2
3
4
5
6
7
8
9
f
0
= 0.005
5
a /D
/
(
D
0
)
T/
0
=0.0
T/
0
= 0.25
T/
0
= 0.5
T/
0
= 0.5
10 15 20
Figure 78 Crack resistance curves simulated by the
computational cell method under SSY conditions for
various T-stress levels. From Xia, L. and Shih C. F.
1995a. Ductile crack growth. I: A numerical study
using computational cells with microstructurally
based length scales. J. Mech. Phys. Solids 43,
233259.
a (mm)
J
(
k
J
m
2
)
a/W = 0.6
W = 50 mm
W = 100 mm
W = 1000 mm
D = 200 m
n = 0.1
f
0
= 0.005
1000
800
600
400
200
0
1 2 3 4 5 0
Figure 79 Crack resistance curves simulated by the
computational cell method for three-point bending
specimens of different sizes (different widths W).
From Xia, L. and Shih, C. F. 1995a. Ductile crack
growth. I: A numerical study using computational
cells with microstructurally based length scales.
J. Mech. Phys. Solids 43, 233259.
Ductile Fracture in Metals 761
Detailed 3-D finite element simulations of
cracked sheets made of a J
2
elastoplastic solid
have been performed by Hom and McMeeking
(1989a, 1989b) and Nakamura and Parks
(1990) within a finite strain setup in order to
investigate the stress and strain fields at the tip
of a static crack (see also Pardoen et al., 1999).
The first key result of these simulations is, in
agreement with many experimental observa-
tions, the development of localized necking
zone in the near-crack-tip region, as schemati-
cally shown in Figure 80. This localized necking
region will be considered to be the FPZ. As
shown in Figure 80 for the specific case of a
DENT (double edge notched tension) specimen
geometry, the necking region is surrounded by
a large zone of gross plasticity, such as in thick
plates. Due to the development of this neck,
moderate out-of-plane stress builds up in the
near-crack-tip region.
Figure 81 shows the variation of the stress
triaxiality as a function of the distance to the
crack tip in sheets of various thicknesses, at the
midplane, and along the surface. Along the sur-
face and away from the crack tip, the plane
stress HRR solution is recovered (see
Figure 69). But, locally, stress triaxiality rises
in the FPZ. Hence, the plane stress condition
is an idealization which is only attained
approximately for the limit of extremely thin
sheets. Note that the maximum stress triaxiality
prevailing inside the FPZ typically varies
between 0.6 and 2.0 depending on the thickness,
a range in which the work of damage and mate-
rial separation G
0
is expected to change a lot
(see Figure 75). Finally, as first predicted by Hill
(1952), plane strain or near plane strain condi-
tions prevail along the neck in the ligament
direction.
In thin sheets, the total work of fracture G is
thus made of three contributions:
(Da. t) =
0
(Da. t)
n
(Da. t)
p
(Da. t) [129[
where G
0
is the more intrinsic fracture tough-
ness accounting for damage and material
separation, G
n
is the work per unit crack
advance required for localized necking, and G
p
is the extrinsic contribution resulting from
gross plastic dissipation during crack propaga-
tion. As is discussed in details in this section, the
presence of the necking work in [129] will be
L
0
t
0
X
0
Diffuse plastic
zone
Localized necking
zone
Final microzone of
damage-induced localization
Figure 80 The DENT geometry with the diffuse and localized PZs (macroscale), the localized necking zone
(mesoscale), and the true fracture zone with the void spacing definition (microscale) (see Pardoen et al., 2004).
3 4 5 6 1 2
HRR ~ 0.6
Stress triaxiality
Increasing
sheet thickness
r/
Fracture process zone
= necking zone
0.6
1.0
Midplane
Surface
Figure 81 Variation of the stress triaxiality in thin
ductile sheets as a function of the distance to the
crack tip (normalized by the CTOD) for various
sheet thicknesses.
762 Failure of Metals
responsible for the thickness dependence of the
fracture toughness in thin sheets.
At cracking initiation (Da =0), G
0
X G
0
init
,
G
n
X G
n
init
, and G
p
=0. The fracture toughness,
quantified by the value of the J-integral at
cracking initiation, J
c
, involves both damage
and necking works:
J
c
=
init
0
init
n
[130[
The so-called essential work of fracture
(EWF) method introduced by Cotterell and
Reddel (1977) provides an alternative to the
fracture mechanics approach for characterizing
the tearing resistance of thin sheets. The EWF
method is an experimental method allowing the
measurement of an index w
e
which should, in
principle, be equal to the steady-state work
spent in the FPZ:
w
e
=
SS
0
SS
n
[131[
This important and revealing method for
thin sheet fracture analysis will be discussed in
a first specific section.
After cracking has been initiated, two possi-
ble fracture modes, shown in Figure 82, are
usually observed experimentally, flat fracture
and slant fracture. The first mode of fracture,
called here flat fracture, consists in a crack
running in mode I, usually accompanied with
significant amount of crack-tip necking. Two
sections will be devoted to that (probably not
enough studied) fracture mode: Section
2.06.3.8.3 provides a specific discussion about
the necking contribution which is one of the
main peculiarities of thin sheet fracture while
Section 2.06.3.8.4 gives other information
about this fracture mode and focuses on the
importance of the necking contribution with
respect to the damage contribution leading to
the thickness dependency of the fracture tough-
ness in many thin sheet materials. In the slant
fracture mode, rapidly after initiation the crack
plane tilts at 45
)
0 0.5 1 1.5 2 2.5 3 3.5
y = 2.2123e + 05 + 2.0406e + 05x R = 0.99723
Sheet thickness
0
ss
n
ss
Stainless steel AISI 316L
(b)
0
1.4 10
6
1.2 10
5
1 10
6
8 10
5
6 10
5
4 10
5
2 10
5
0 1 2 3 4
t
0
= 0.65 mm t
0
= 1.49 mm t
0
= 3.04 mm
y = 3.4025e + 05 + 38787x R = 0.99272
y = 5.4627e + 05 + 31076x R = 0.96309
y = 8.3418e + 05 + 24556x R = 0.90604
w
f
(
J
m
)
l
0
(mm)
w
e
w
p
(a)
Figure 83 Application of the essential work of
fracture method to AISI 316L stainless steel of
different thicknesses, allowing first to: a, to separate
the EWF w
e
from the gross plasticity contribution w
p
by linear interpolation as a function of the ligament
length, and b, separate the damage work G
0
from the
necking contribution G
n
(results by Marchioni,
2002).
764 Failure of Metals
usually much larger than 1. Dimensional ana-
lysis shows that, for a material with a flow
behavior represented by eqn [135] and for a
given geometry and stress state, the average
work per unit volume w
n
spent in the neck can
be expressed as
w
n
s
0
= F n. k.
s
0
E
. i. e
f
_ _
[136[
where e
f
is the strain at fracture. As shown by
Hill (1952), a plane strain tension stress state
can be assumed.
The model (not described here) is closed form
except for the evaluation of the shape para-
meter of the necking region (height over
thickness of the active PZ) whose adjustment
required conventional 2-D finite element simu-
lations already presented in Figure 23. Figure 84
shows the variation of the ratio w
n
/o
0
k
n
X G
n
/
o
0
k
n
t
0
as a function of the equivalent strain at
fracture e
f
minus the equivalent necking strain
e
u
(corresponding to plane strain tension con-
ditions, see Section 2.06.3.2.2) for different n.
Indeed, the important factor for the necking
contribution is not the fracture strain but the
difference between the strain at necking and the
fracture strain. The work of necking per unit
volume levels out at high fracture strains as the
active necking zone becomes increasingly small,
involving less and less additional plastic work.
Note that plastic anisotropy can significantly
affect the work necking. For instance, in alumi-
num sheets, the Lankford coefficient is lower
than 1, favoring the through thickness reduction
and leading thus to a smaller work of necking
than the predictions given in Figure 84.
2.06.3.8.4 Flat mode I fracture in thin plates
The perceived wisdom about thin sheet frac-
ture is that (1) the crack propagates under
mixed mode I and III giving rise to a slant
through-thickness fracture profile and (2) the
fracture toughness remains constant at low
thickness and eventually decreases with increas-
ing thickness. In a study by Pardoen et al.
(2004), fracture tests performed on thin
DENT plates of various thicknesses made of
stainless steel, 6082-O and NS4 aluminum
alloy, brass, bronze, lead, and zinc (see
Table 8) systematically exhibit mode I
bathtub, that is, cup-and-cup, fracture pro-
files with limited shear lips and significant
localized necking. Furthermore, the fracture
resistance systematically increases with sheet
thickness, in a linear way, as anticipated in
Section 2.06.3.8.2. The discussion about flat
fracture in thin sheets of the present section is
essentially based on this work.
For the sake of illustration, Figure 85 shows
that the two matching fracture surfaces of an
aluminum alloy (NS4) are similar (many other
micrographs can be found in Pardoen et al.
(2004) or Rivalin et al. (2001a)). A regular frac-
ture surface with dimples is observed along the
sides of the specimen without any evidence of
the shear distortion typical of fracture surfaces
resulting from shear failure. As represented
schematically in Figure 86, the mode I bathtub
fracture profile originates from the difference of
stress triaxiality between the center and the sur-
faces of the specimens. The surface is in a pure
plane stress state involving thus a fracture
strain larger than in the center where the stress
triaxiality is larger due to necking induced
stress concentration. This significant difference
between the stress state in the center and along
the surface leads to the tunneling effect with the
crack length in the center plane longer than the
crack length along the surface by about one
thickness when the steady-state regime is
attained. The steady-state regime, associated
to constant thickness reduction on the crack
surface, is attained after the crack has propa-
gated on a distance equal to about 1 or 2
thicknesses.
The following properties of the materials
listed in Table 8 have been systematically
measured:
v Flow properties using the hardening law
[135]. It is worth noticing the moderate or
high strain-hardening capacity of all these
metallic materials.
v Thickness reduction factor, r
f
, defined as
r
f
=
t
0
t
f
t
0
[137[
where t
f
is the final plate thickness along the
fracture plane. As no significant or systematic
dependence of r
f
on thickness is observed, the
r
f
values are averaged over the different
0
0.1
0.2
0.3
0 0.5 1 1.5 2 2.5
w
n
/
0
k
n
n = 0.1
n = 0.25
n = 0.4
n = 0.5
f
u
Figure 84 Variation of the average work of
necking (per unit volume) as a function of the
fracture strain minus the necking strain (see
Pardoen et al., 2004).
Ductile Fracture in Metals 765
thicknesses. It is interesting also to point out
the behavior of lead which fails by full neck-
ing, that is, without any apparent damage
mechanism. For the other materials, the
reduction of thickness is always larger than
about 50%.
v The fracture surfaces have been observed by
SEM in order to estimate the dimple sizes
and spacing. The spacing has been quantified
in the direction of crack advance which,
owing to the near plane strain conditions,
gives a direct image of the initial defect spa-
cing, X
0
(see Table 8).
v The two contributions to the fracture resis-
tance, G
0
and w
n
, were evaluated using the
extended EWF procedure presented in
Section 2.06.3.8.2 from tests performed on
multiple DENT specimen thicknesses and
ligament lengths (Figure 83 shows the results
for the steel A316L). The most striking fea-
ture in the measured values reported in
Table 8 is that the work of necking per unit
area G
n
(i.e., w
n
multiplied by the initial
thickness of the plate) is at least similar and
usually larger than G
0
for thicknesses in the
millimeter range. This implies that the frac-
ture resistance quantified by G
0
G
n
significantly increases (linearly) with thick-
ness, in the investigated range of thicknesses.
The thickness dependence of the fracture
resistance in thin sheets, although it has not
received much attention in the literature (e.g.,
Bluhm, 1961; Swedlow, 1965), is a very impor-
tant effect that must be properly understood
and controlled in order to design fail-safe struc-
tures made of thin sheets and in order to allow
comparing the fracture resistance of different
materials processed with different thicknesses.
The thickness dependence directly results from
the necking work. A simple model for the neck-
ing work has been proposed in the previous
subsection. It remains now to develop a model
Table 8 Properties of ductile metallic sheets with thicknesses in the range 0.55 mm
Materials
E
(GPa)
o
0
(MPa) n k r
f
X
0
(mm)
w
n
exp
(MJ m
3
)
w
n
model
(MJ m
3
)
G
0
exp
(kJ m
2
)
G
0
model
(kJ m
2
)
Steel A316L 210 310 0.48 25 0.48 2550 204 165 221 172345
Al 6082-O 70 50 0.26 265 0.6 1020 33 32 28 816
Brass A 110 100 0.6 33 0.78 510 161 195 87 108217
Al NS4 // RD 70 140 0.17 159 0.8 815 42 39 51 2648
Zinc // RD 61 100 0.15 118 0.59 25
*
64 23 34 16.5
Lead 16 7 0.25 290 1 7.3 5.3 0 0
Bronze A 100 120 0.51 38 0.7 45 218 166 70 5163
E is the Youngs modulus, o
0
is the yield stress, k and n are the parameters in relation [135], obtained by a power law fit on the
uniaxial stress strain curve, r
f
is the thickness reduction factor defined by eqn [137], X
0
is the mean initial void spacing (* mean
grain size for Zn), G
0
is the true work of fracture, and w
n
is the work of necking per unit volume.
(a) (b)
200 m
10m
Figure 85 Transverse sections (along the thickness) of fracture surfaces of thin A1 NS4 sheets: a, the bathtub
profile; b, higher magnification showing a regular mode 1 fracture surface with dimples (the voids are not
distorted by shear type localization associated to a slant fracture mode).
Triax=T
T
center T
sides
=0.6
f
Figure 86 Bathtub fracture profile (or cup and
cup) resulting from a higher stress triaxiality and
thus lower fracture strain in the center of the plate.
766 Failure of Metals
for G
0
in order to capture the relative influence
of the necking work and the importance of the
thickness effect as a function of the flow prop-
erties and microstructural features.
The model presented aims at calculating the
energy G
0
spent for the growth and coalescence
of voids in front of a crack tip following the
reasoning developed in Section 2.06.3.7.2 for
thick plates, but now in the case of thin sheets.
Based on the schematic drawing of Figure 75,
we expect now much larger values of G
0
because
the stress triaxiality is much lower (see
Figure 81). We again assume, as shown in
Figure 73, a material made of regularly distrib-
uted voids (the voids are supposed to be present
from the beginning of the loading) with initial
spacing X
0
and initial volume fraction f
0
. Only
voids that are initially spherical will be consid-
ered. Dimensional analysis for the hardening
law [135] gives
0
s
0
X
0
= F n.
s
0
E
. k. i. f
0
_ _
[138[
In order to simplify the analysis, the complex
3-D stress state existing at the crack tip will be
approximated by plane strain tension allowing
for necking development (see Hill, 1952). The
2-D finite strain simulations have been per-
formed using the same constitutive material
response as for the thick plate case (i.e.,
Gologanu and Thomason models as presented
in Sections 2.06.3.4.4 and 2.06.3.5.4). The impor-
tance of using a damage model which properly
incorporates void shape changes is more impor-
tant than for thick plates. Indeed, the low stress
triaxiality in thin sheets leads to significant void
elongation. G
0
is evaluated in the most loaded
element, that is, the element located in the center
of the minimum section, from the calculated
stress displacement response.
The model has been validated in the follow-
ing way. The initial void volume fraction f
0
was
identified by simulating the uniaxial tensile tests
and finding the value that allows reproducing
the experimental fracture strains. The void spa-
cings X
0
measured experimentally (see Table 8)
have been used to estimate G
0
. The experimen-
tal and predicted G
0
are compared in Table 8.
These results remain semiquantitative due to
the numerous approximations in the model
and the experimental uncertainty but the main
trends are captured.
For a given geometry, loading configuration,
and a constant product o
0
X
0
, the two most
important parameters affecting ductile fracture,
that is, affecting the nondimensional function F
in [138], are the initial void volume fraction f
0
and the strain-hardening exponent n. The other
parameters that have been kept constant are:
o
0
/E=1/k =10
3
, i =0.3. Figure 87 presents
the variation of G
0
/X
0
o
0
as a function of f
0
for
n equal to 0.1, 0.3, and 0.5. The effect of both n
and f
0
is obvious: large n and low f
0
significantly
increase G
0
.
The plane stress fracture energy G
0
is also
compared to the plane strain fracture toughness
J
Ic
in Figure 87 for n =0.1. To a first approx-
imation the ratio G
0
/J
Ic
is not significantly
affected by the initial porosity, nor by the
value of the strain-hardening exponent (result
not shown). It ranges between 2.5 and 3.5 which
agrees very well with full 3-D finite element
simulation of crack propagation in thin and
thick HSLA steel plates reported by Rivalin
et al. (2001b). The plane stress fracture energy
is significantly larger than the plane strain value
because of the much smaller stress triaxiality,
which involves smaller void growth rate. To our
knowledge, there exist no noncontroversial
experimental data in the literature to assess
this prediction. The difficulty is to find a mate-
rial which will be (1) ductile enough to show the
necking mechanism at small thickness but (2)
not too ductile, otherwise it is not possible to
measure plane strain fracture toughness
values because the required thickness for
valid measurements would be too large. We
will come back to the general analysis of thin
versus thick plate fracture resistance in
Section 2.06.3.8.6.
At this point, it is possible to come back to
the problem of the coupling between crack-tip
necking and crack-tip damage. Figure 88 gath-
ers the results of Figures 84 and 87 in terms of
the variation of the ratio G
0
X
0
k
n
/G
n
t
0
as a func-
tion of f
0
. The proportion of damage and
necking contributions in the work of fracture
0
/J
Ic
0
/
0
X
0
n = 0.1
n = 0.3
n = 0.5
n = 0.1
f
0
0.1
1
10
100
1000
10
5
10
4
10
3
10
2
10
1
10
0
Figure 87 Variation of the predicted fracture energy
as a function of the initial porosity for different
strain-hardening exponents; variation of the ratio of
the plane stress fracture energy over the plane strain
fracture toughness as a function of the initial
porosity (see Pardoen et al., 2004).
Ductile Fracture in Metals 767
depends very much on n and f
0
and linearly
scales with X
0
/t
0
. Figure 88 exhibits two limits:
one when f
0
0.10.2 and one when f
0
0.
The first limit corresponds to highly porous
materials where the fracture strain becomes
smaller than the necking strain and thus G
n
is
equal to 0. This limit is attained for very large
initial porosity (>0.1) that is not encountered
in typical industrial alloys. The other limit when
f
0
0 also leads to fracture toughness that is
mainly controlled by the damage mechanisms,
because G
n
then saturates at large fracture
strains (see Figure 87). However, this limit is
not really meaningful as the model is no longer
valid when f
0
tends to zero and thus X
0
goes to
infinity in finite specimens. Indeed, the thick-
ness of the plate sets a second length scale. Thus
the real limit for f
0
0, which probably corre-
sponds to the experimental results obtained for
lead (see Table 8), is that G
0
tends to 0 and that
fracture is then only controlled by plastic neck-
ing. In between these two limits, a local
minimum appears over the range of porosity.
This minimum corresponds to the maximum
possible amount of necking dissipation with
respect to damage. Figure 88 also shows the
marked effect of the strain-hardening capacity
on the energy partitioning. Increasing the strain-
hardening capacity affects much more G
0
than
the necking contribution for a given ratio X
0
/t
0
.
The wide variety of behaviors that can be
deduced from Figure 88 probably explains the
wide variations of apparent properties encoun-
tered in thin plates and the difficulty of
rationalizing experimental measurements from
only a macroscopic point of view (e.g., Broek,
1978).
The summary of this section is that the work
of necking (per unit area) G
n
(1) scales linearly
with thickness, (2) depends on the strain-hard-
ening exponent, and (3) increases with the
fracture strain to reach a constant value at
large fracture strains. The work of fracture
(per unit area) G
0
scales linearly with the yield
stress and void spacing, and strongly depends
on the initial porosity and strain-hardening
exponent.
2.06.3.8.5 Competition between flat
and slant fracture
The most important structural metallic
sheets, that is, high-strength aluminum alloys
in an age-hardened state, high-strength steels,
and the classical TA6V titanium alloy, exhibit a
slant fracture mode (e.g., Irwin et al., 1958;
Krafft et al., 1961; Zinkham, 1968; Allen,
1971; Knott, 1973; Krambour and Miller,
1977; Broek, 1978; Atkins and Mai, 1985;
Sutton et al., 1995; Taira and Tanaka, 1979;
Rivalin et al., 2001a; Mahmoud and Lease,
2003; James and Newman, 2003; Chabanet
et al., 2003; Asserin Lebert et al., 2005). As
depicted in Figure 89, the crack starts propagat-
ing in a flat mode I with significant tunneling.
After moderate amount of crack growth, shear
localization sets in and the crack plane tilts at
an angle which depends on the plastic aniso-
tropy of the material (45
in plastically isotropic
materials). As a result, the fracture surface
shows a first triangle of flat fracture before
the profile becomes slanted.
The following conditions have been shown to
favor slant fracture:
(i) Flow properties
v Low strain-hardening capacity. As explained
above, most thin sheets made of high-
strength alloys exhibit a slant fracture
mode. It is a common rule in materials
10
10
5
10
4
10
3
10
2
10
1
100
1000
10 000
0
t
0
k
n
/
n
X
0
f
0
n =0.3
n =0.5
n =0.1
Figure 88 Variation of the ratio of the fracture and
the necking work as a function of the initial porosity
for different strain-hardening exponents.
Slant fracture
Tensile
mode
Tunneling
a
s
a
Crack
front
Fatigue surface
B
W
2
Figure 89 Typical slant fracture surface. From
James, M. A. and Newman, J. C., Jr. 2003. The effect
of crack tunneling on crack growth: Experiments and
CTOA analyses. Eng. Fract. Mech. 70, 457468.
768 Failure of Metals
science that high strength, attained by cold
rolling and/or proper aging treatment lead-
ing to optimal precipitation hardening, is
usually associated to low strain-hardening
capacity. For instance, as shown by
Asserin-Lebert et al. (2005), the same
Al6056 alloy exhibits a slant profile in the
age-hardened heat-treated condition T751
(o
0
=300 MPa, n ~0.06) but a flat fracture
mode in the annealed state (o
0
=70 MPa,
n ~0.2).
v Plastic anisotropy. Sutton et al. (1995)
reported that the fracture of a 2024-T3 Al
sheet was slant when loaded in a LT config-
uration but flat when loaded in a TL
configuration. The strain hardening capacity
was slightly larger in the TL orientation. It is
not clear whether it is the only reason to
explain the change of fracture mode.
(ii) Microstructure
v Second population of voids. Bron et al. (2004)
have shown for a 2024 T4 Al alloy that the
slant regions were covered by both primary
and secondary dimples whereas the initial flat
triangular region had only primary dimples.
The presence of secondary voids favors the
transition from internal necking coalescence
to the void sheet shear-type coalescence
mechanisms (under specific loading condi-
tions), which is naturally related to the
transition to a slant fracture mode (see also
Section 2.06.3.5 on void coalescence).
v One grain along the thickness. Slant fracture
is favored when only one grain is present
along the thickness.
(iii) Geometry
v Small thickness. Increasing the thickness
changes the stress state at the crack tip by
increasing the stress triaxiality, leading to a
decrease of the propensity toward slant frac-
ture (see Asserin-Lebert et al., 2005).
v Flat surfaces. Side grooves (rather than flat
surface) are machined in order to force the
crack to remain flat.
(iv) Loading conditions
v Low load biaxiality. Although no evidence
have been found in the literature, adding a
load transverse to the crack plane would
probably decrease the propensity toward
slant fracture by increasing the constraint in
the near-crack-tip region.
v High loading rates. As shown by Rivalin
et al. (2001a) in an application on a high-
strength X70 ferritic pearlitic steels, high
loading rates can lead to adiabatic shear
bands and slant fracture, whereas at low
loading rates fracture remains flat.
As a matter of fact, all these conditions tend
to favor plastic shear localization, with the
most important factors being a low strain-hard-
ening capacity and a low constraint. Although
the problem can be addressed theoretically (see
Hahn and Rosenfield, 1965), formulating a
general predictive criterion for shear localiza-
tion in cracked sheets is a formidable task that
is most probably outside the scope of any ana-
lytical theoretical developments. The
complexity is related to the finite strain 3-D
character of the boundary value problem and
to the complexity of realistic elastoviscoplastic
flow laws of metallic materials. Hence, only full
3-D numerical simulations can provide quanti-
tative insights into the competition between
slant and flat fracture in thin sheets. There
have been only a few number of works devoted
to the simulation of the slant fracture able to
capture the complex initial crack tilting; see
Mathur et al. (1996) for 3-D dynamic simula-
tions involving two populations of voids as
well as adiabatic heating and associated ther-
mal softening effects; see also Rivalin et al.
(2001b).
2.06.3.8.6 General views about thickness
dependence of fracture resistance
This section, which goes beyond the analysis
by Pardoen et al. (2004), aims now at synthesiz-
ing the messages conveyed in Sections 2.06.3.7
and 2.06.3.8 about the ductile tearing resistance
of ductile metals. A generic crack resistance
curve of a ductile metallic plate (sufficiently
thin to exhibit some amount of crack-tip neck-
ing) is shown in Figure 90 with the variation of
the different contributions to the overall work
of fracture. As explained earlier, crack-tip neck-
ing increases with increasing crack advance
before reaching a steady value after the crack
has propagated for a distance equal to one or
two thicknesses (see Cotterell and Reddel, 1977;
Pardoen et al., 1999, 2004). The work of neck-
ing thus increases with increasing crack
advance to reach a constant value noted G
n
SS
when Da =12 t. The fact that the fracture strain
increases with increasing crack advance means
that the damage work also changes, as a result of
a change of stress state. Hence, G
0
also moder-
ately evolves and reaches a value G
0
SS
when
Da =12 t. A change of G
0
with crack advance
is also expected in thick plates due to the change
of constraint during crack propagation when
compared to cracking initiation. This change is
probably small and could be a decrease rather
than an increase of G
0
. The gross plastic
Ductile Fracture in Metals 769
dissipation G
p
keeps increasing (and could theo-
retically reach a steady-state value too if large
enough samples are used). The EWF
w
e
=G
n
SS
G
0
SS
is also added on Figure 90.
Now, the generic evolution of the cracking
resistance characterized by the sumG
0
G
n
=G
c
,
either at cracking initiation or during propaga-
tion as a function of the plate thickness, is shown
in Figure 91. The thickness is normalized by the
theoretical SSY PZ size. The fracture resistance
G
c
is normalized by the product o
0
X
0
. These
curves are anticipated from the discussions of
the last two sections (note that generic curves
showing the toughness increasing and then
decreasing with increasing thickness are shown
in several classical fracture mechanics textbooks
(e.g., Broek, 1978; Atkins and Mai, 1985;
Barsom and Rolfe, 1987). Nevertheless, to our
knowledge, fully valid fracture resistance curves
covering the whole range of behavior have not
yet been obtained experimentally nor simulated
for the full range of thicknesses. The reasoning
was the following.
First, the transition between the quasi-plane
stress and plane strain regime occurs when the
thickness is roughly equal to PZ size r
y
esti-
mated by
r
SSY
y
~
1
10
K
Ic
s
0
_ _
2
=
1
10
E
s
0
J
Ic
s
0
[139[
This transition corresponds to the significant
rise of the stress triaxiality in the FPZ
owing to the increase of the through-thickness
stress.
Then, we distinguish between slant fracture
(curve A) and flat fracture (curves B and C).
Slant fracture usually involves lower fracture
strain in the FPZ due to the appearance of
shear localization bands with high internal con-
straint. Necking can thus not significantly
develop and the increase of the fracture resis-
tance is expected to be limited before it
decreases down to the plane strain limit. This
is confirmed qualitatively by the results of
Asserin-Lebert et al. (2005) on as- received
high-strength low hardening 6056 Al alloys
showing slant fracture with minor increase of
the fracture resistance at small thickness
(<3 mm) and a significant drop for larger thick-
nesses (>3 mm). Note also that Zinkham (1968)
reported an increase of the fracture resistance
with increasing thickness in a 7075-T6 Al alloy
showing a slant fracture mode, while the
decrease at larger thickness has been measured
by several authors (Bluhm, 1961; Taira and
Tanaka, 1979; Lai and Ferguson, 1986; Guo
et al., 2002). Now, the level of the slant fracture
curve cannot be compared to the level of the flat
fracture curve, even expressed in normalized
terms. Among others, even for the same mate-
rial differing only by the yield strength and
strain-hardening exponent (e.g., heat-treated
12
a/t
0
w
e
p
ss
0
ss
n
ss
n
init
0
init
J
c
0
Figure 90 Generic crack resistance curve of a
ductile metallic plate.
1
plane
stress
plane
strain
A
B
C
C
0
A Slant fracture (tentative)
B
C
Flat fracture with large
0
small n
Flat fracture with small
0
large n
w
n E
30
k
n
1
C
n
=
0
x
0
0
x
0
0
x
0
0
k
n
t /r
y
ssy
=J
Ic
2
Figure 91 Tentative generic variations of the fracture toughness (either at initiation or during propagation) as
a function of the plate thickness for slant fracture (curve A) of for flat fracture (curve B and C).
770 Failure of Metals
Al alloy), the representative X
0
might be differ-
ent due to different damage mechanisms.
More detailed considerations can be elabo-
rated for the flat fracture mode. We start by
expressing the normalized fracture resistance as
a function of the normalized thickness in the
small thickness range (where G
0
is almost con-
stant and equal to the plane stress low
triaxiality G
0
Po
limit):
c
s
0
X
0
=
Ps
0
s
0
X
0
n
s
0
X
0
=
Ps
0
s
0
X
0
1
s
0
Ps
0
w
n
s
0
t
0
_ _
=
Ps
0
s
0
X
0
1
1
30
k
n
E
s
0
w
n
k
n
s
0
t
0
r
y
_ _
[140[
where relationship [139] has been used as well as
G
0
Po
~3J
Ic
(see Section 2.06.3.8.4). The slope of
the initial linear increase of the fracture resis-
tance is thus given by the product
slope X a ~
1
30
k
n
E
s
0
w
n
k
n
s
0
[141[
This expression reveals several limiting cases.
First, if the metal presents a low fracture strain,
e
f
is not much larger than e
u
(a good example is
provided by some multiphase TRIP steels
which show excellent ultimate tensile strain
but limited postnecking ductility; see Jacques
et al., 2001). In that case, the amount of necking
remains limited and the behavior will ressemble
curve B.
Secondly, if the metal is sufficiently ductile,
the necking term w
n
/k
n
o
0
does not depend on
the fracture strain (see Figure 84), that is, on the
details of the microstructure. It only depends
moderately on the strain hardening. The domi-
nant term in [141] is thus k
n
E/o
0
. For instance,
between a moderately strain-hardening mate-
rial such as Al NS4 and a high strain
hardening metal alloy such as brass (see
Table 8), this last term can increase by almost
a factor of 10 (high strain hardening is usually
associated to low-yield-strength alloys), while
w
n
/k
n
o
0
only increases by a factor of 2. These
two extreme situations correspond to curves B
and C in Figure 91. Curve B belongs to high-
strength low-hardening alloys (such as in most
nonannealed cold-formed metals). In that case,
the necking contribution will be limited. In fact,
a continuous decrease of the toughness with
increasing thickness will probably be measured
in such materials. Indeed, G
0
is affected by the
stress state. With increasing thickness, stress
triaxiality will tend to increase at the crack tip,
to accelerate the void growth rate with respect
to the plane strain tension situation, to decrease
the fracture strain, and thus to lead to a
decrease of G
0
. This can justify results (e.g.,
Broek, 1978), where the toughness is observed
to always decrease with increasing thickness
even at small thicknesses. Curves C belong to
low-strength high-strain-hardening materials
(such as in most FCC annealed pure metals
and alloys), like most materials of Table 8. In
those materials, the fracture toughness mark-
edly increases with thickness, while remaining
in the quasi-plane stress regime (see Table 8 or
Figure 83b). It is thus not surprising that the G
0
contribution has not been detected in some
materials (e.g., Mai and Powell, 1991). In such
materials, the fracture resistance is essentially
controlled by the necking contribution.
Again, we have to be cautious when inter-
preting Figure 91. First of all, as explained
above, these curves have never been measured
experimentally in a fully conclusive manner.
Second, several metals can show a transition
of fracture mode (from slant to flat or vice
versa) when changing the thickness (see
Asserin-Lebert et al., 2005) or when using or
not side-grooves. Finally, only full 3-D calcula-
tions are capable of describing either the
transient or to encompass both thin and thick
sheets, as well as to better capture the competi-
tion with the slant fracture mode.
2.06.4 DBT IN FERRITIC STEELS
2.06.4.1 Introduction
This part is devoted to the analysis of the
macroscopic transition in fracture mode which
is observed in ferritic steels. This transition is
also observed in other BCC metals and in some
HCP metals which are susceptible to cleave at
low temperature. The situation corresponding
to ferritic steels is selected because of its theore-
tical and practical importance. This section does
not deal with the microscopic transition between
blunted cleavage and pure cleavage since this
topic has already been briefly discussed when
describing cleavage theories (see Section 2.06.2).
The competition between ductile tearing and
cleavage fracture controls the macroscopic frac-
ture toughness of ferritic steels in the upper part
of the DBTregime. Ductile crack growth (DCG)
can occur under increasing load, and the struc-
ture can withstand a significant amount of stable
ductile tearing without substantial loss of load-
bearing capacity. Cleavage fracture, on the other
hand, leads to catastrophic failure of structural
components, and the onset of cleavage is the
critical mechanism limiting the load-bearing
capacity of the structure. Here it is worth noting
that, in this part, the DBT transition is only
analyzed when cleavage fracture is preceded by
some DCG. Cleavage fracture occurring during
blunting, that is, for a stationary crack, is not
considered, since this has been largely discussed
elsewhere (see Chapter 7.05).
DBT in Ferritic Steels 771
It is well known that the DBTTin ferritic steels
is strongly affected by the strain rate in smooth
and notched specimens or the loading rate in
cracked geometries. The DBTT decreases when
the strain rate or the loading rate are increased.
The transition at relatively low loading rate is
usually investigated using fracture toughness spe-
cimens. The transition in these specimens
corresponds to relatively simple loading condi-
tions which are quasi-static and isothermal. The
transition temperature is defined for a given
value of the stress intensity factor (for instance,
K=100MPam
1/2
). The DBT at high loading
rates is usually investigated by using impact
Charpy V-notch specimens. This includes the
measurement of the total energy absorbed to
fracture and the determination of the DBTT.
The transition temperature is defined for a
given level of the Charpy (CVN), for instance
CVN=57 J. Charpy impact testing has remained
for a long time essentially a technological tool
which has proved to be extremely useful for
ranking the fracture properties of materials.
However the correlations between the Charpy
energy and the fracture toughness have remained
largely empirical. It is only recently that the local
approach to fracture has been used for a better
understanding of the Charpy test. A recent con-
ference for the centenary anniversary of the
Charpy test has been devoted to this topic
(Franc ois and Pineau, 2002).
In this section, the DBT behavior observed in
fracture toughness tests is described first.
Simplified and more advanced models based
on the study and the numerical simulation of
the fracture micromechanisms are presented.
Then the results obtained from recent studies
devoted to the analysis of Charpy V impact
tests are presented.
2.06.4.2 DBT in fracture toughness tests
2.06.4.2.1 Introduction
A simple explanation for the existence of a
transition is that the stressstrain curve
decreases with increasing temperature, which
results in decreasing normal stress ahead of a
crack tip with increasing temperature.
Therefore, in the transition regime, there exists
a temperature at which the maximum normal
stress ahead of the crack tip reaches the
required cleavage fracture stress. Ductile tear-
ing can then be initiated which can transform
into cleavage fracture due to the increase of the
normal stress associated with crack growth.
However, no single explanation can be given
for this change in failure mechanism from duc-
tile rupture to cleavage fracture. A combination
of different causes seems to be more realistic.
Possible causes for this change are:
1. The stress level ahead of a growing ductile
crack is higher than the stress ahead of a static
blunted crack.
2. The volume of material, with a high max-
imum normal stress, which is sampled during
the fracture process, increases as the crack
extends. This results, according to the weakest
link theory for cleavage fracture, in an increas-
ing probability of cleavage fracture. However,
due to the competition between ductile damage
and cleavage fracture initiated from second-
phase particles, the number of cleavage initia-
tion sites may decrease due to ductile void
formation. This competition between ductile
and cleavage fracture implies that some modi-
fications to the original cleavage theories which
were presented previously (Section 2.06.2.2)
must be made.
The two possible causes for the change in
fracture mode in the transition region are ana-
lyzed and discussed successively. A simple (and
nave) approach to the prediction of the DBT
behavior is introduced first. Then, more sophis-
ticated models accounting for the selection of
potential cleavage initiation sites are presented.
The application of these models to a number of
ferritic steels is illustrated.
2.06.4.2.2 A simplified approach
One simple, but likely too nave, approach to
the prediction of the DBT behavior is to con-
sider the existence of a unique ductile tearing
resistance curve giving the value of J-integral
(or better J/o
0
to account for a slight tempera-
ture dependence) versus crack length, and to
calculate the probability to fracture, using, for
instance, the Beremin model for various station-
ary crack lengths, as schematically shown in
Figure 92a. In this figure, it is schematically
shown that the value of J, at which the DBT
occurs (for a given probability), increases with
temperature, essentially because of the decrease
of the yield strength due to the increase in tem-
perature. This approach is easy to implement
since the probability to fracture is calculated
using a postprocessing procedure. The method
requires to determine experimentally or to simu-
late numerically the JDa curve at a temperature
at which cleavage fracture does not occur and
then to normalize the value of the loading para-
meter J by the corresponding yield strength.
This simplified approach which does not
account for the details associated with a propa-
gating crack was applied to A508 RPV steel
(Amar and Pineau, 1987) using cracked round
772 Failure of Metals
bars which were tested at four temperatures:
100, 80, 50, 20
C
(Koers et al., 1995). These authors tested four-
point single edge notched bend specimens to
determine the fracture properties of the mate-
rial, and applied the Beremin theory to predict
cleavage fracture occurring after some ductile
tearing. They showed that, in their material,
the calculated fracture probabilities were
significantly larger than those determined
experimentally, at least for small ductile crack
extensions (Da _1.5 mm). However, no expla-
nation was given about how crack advance due
to crack blunting effect was taken into account.
The difference in the conclusions drawn by
these two groups of researchers might be due
to the definition in crack extension and crack
blunting effect or, more likely, to the difference
in materials and specimen geometries. In parti-
cular, as shown by Scibetta et al. (2000), the
circumferentially cracked round geometry
used in the study on A508 steel maintains a
high constraint effect, which is not the situation
met with bend specimens used in the study on
Fe 510 Nb steel. This insight partly explains the
differences in both studies.
As already stated, this simplified approach
involves strong limitations since it does not
account for the modifications in the stress
field ahead of a propagating crack and does
not rely on a detailed study of the microme-
chanisms accompanying crack growth. In
particular, this approach cannot provide
good results when cleavage fracture is con-
trolled by the nucleation of microcracks
initiated from particles, which was likely the
mechanism in Fe 510 Nb steel tested at very
low temperature. On the other hand, the
assumptions behind this approach are prob-
ably better verified when cleavage fracture is
controlled by the propagation of microcracks
arrested at grain (or packet) boundaries
which is more likely when the test tempera-
ture is increased.
2.06.4.2.3 Advanced models
(i) Stress profiles ahead of a growing crack
The starting point for any analysis of the
brittle-to-ductile transition is to rely on a
sound model for ductile tearing. These models
have been presented in details in Section
2.06.3.7. Only the points specific to the under-
standing of the conditions leading to the
transition into cleavage fracture are elaborated
here. As explained in Section 2.06.3.7.1, the
theoretical work by Rice and Sorensen (1978)
has shown that, under SSY conditions and for
an elasticperfectly plastic nondamaging mate-
rial, the main difference in the stressstrain field
between a stationary and a growing crack lies in
the strain singularity and not the stress profile
at the crack tip. The Prandtl slip-line field is
thought to apply also to a propagating crack
a (mm)
a (mm)
J
/
y
(
m
m
)
J
/
y
(
m
m
)
T
1
< T
2
< T
3
T
1
T
2
T
3
P
R
= 0.10
P
R
= 0.90
P
R
= 0.10
P
R
= 0.10
P
R
= 0.90
P
R
= 0.90
T = 50 C
T = 50 C
T = 20 C
T = 20 C
T = 80 C
T = 80 C
T = 100 C
0
0
1
2
3
A508
0.25 0.5 0.75 1 1.25 1.5 1.75
0.90
0.90
0.10
0.10
0.10
0.90
u
= 3000 MPa
(a) (b)
Figure 92 J-resistance curve. a, Sketch showing the probabilities to cleavage fracture of 10% and 90% after
some ductile crack extension at three increasing temperatures, T
1
, T
2
, T
3
; b, experimental results on a pressure
vessel steel (A508) (Amar and Pineau, 1987).
DBT in Ferritic Steels 773
while the strain singularity is much lower, that
is, varying as ln(r), than that corresponding to
the HRR field for a stationary crack. The nor-
mal stress ahead of the crack tip remains close
to 3 times the yield strength of the elasticper-
fectly plastic material with a yield strength o
0
.
These theoretical results apply to a crack which
does not give rise to blunting effect. This
assumption is far from reality. As already
shown in Figure 71, the crack tip is blunted at
crack initiation while during propagation the
unzipping process from one inclusion to
another one gives rise to a crack-tip profile
which is much sharper.
The detailed simulations by Xia and Shih
(1995a, 1995b, 1996), already discussed in
Section 2.06.3.7.3, have shown that, during
DCG, the maximum tensile stress, o
22
, ahead
of a simulated propagating crack, increases
with crack extension, as shown in Figure 93.
These results were obtained from numerical
simulations of three-point bend specimens
(W=50 mm) in a given material; for example,
the ratio between Youngs modulus and yield
strength is equal to 500 and the work-hardening
exponent is equal to 0.1. The mesh size, D, was
equal to 300 mm, and ductile damage was simu-
lated using the version of the Gurson potential
enhanced by Tvergaard and Needleman (see
Section 2.06.3.4). Three crack depths corre-
sponding to a/W=0.10, 0.25, and 0.60 were
simulated. In this bend specimen geometry,
the constraint effect is largely dependent on
the crack depth. Figure 93 shows that at the
early stage of crack growth, Da =D, the max-
imum tensile stress for a/W=0.25 is lower than
that for a/W=0.60. When the stress level for a/
W=0.25 is followed for increasing crack
lengths, it is observed that o
22
increases quickly
with crack growth, and at Da =20D=6 mm,
the peak stress has reached the level for
a/W=0.60. A similar effect is observed for
a/W=0.1 although the effect is less pro-
nounced. This steady elevation of crack-tip
constraint with ductile crack extension can
then increase the risk of cleavage fracture.
This effect of the stress elevation during
crack growth is likely less pronounced in speci-
men geometries in which the constraint is less
dependent on the crack length, such as tensile
specimens with one single edge crack, as shown
by Xia and Cheng (1997). For further discus-
sion on the effect of a growing crack on stress
profiles at crack tip, see also Dodds et al.
(1997), Tanguy (2001), and Tanguy et al.
(2002a, 2002b). It appears therefore that the
DBT behavior will be strongly dependent on
the specimen geometry. Cleavage fracture will
be favored in bend specimens in which the
initial crack size is relatively small. This is
the situation which will be illustrated later in
the analyses of Charpy V-notch specimens.
The introduction of damage ahead of the
crack tip produces a reduction of the local
stresses, when using for instance the GTN
model (see Section 2.06.4.3.2) to simulate
DCG. This softening effect and its consequence
on the calculation of the probability to cleavage
fracture has been quantified by Busso et al.
(1998). However this reduction of the local
stress is macroscopic. At metallurgical scales
much smaller than the cell size, D, formation
and growth of the macroscopic cell voids driv-
ing ductile crack extension likely alter and
amplify the local stress fields acting on the
smaller particles that can trigger and control
1500
1000
500
0
0
(a)
(b)
0
0
1
2
3
4
5
10
2
2
/
0
W = 50 mm
a/W = 0.60
a/W = 0.25
a/W = 0.10
a = D a = 10D a = 20D
20
X
1
/D
30 40
1 2 3
a (mm)
W = 50 mm (TPB)
a/W = 0.1
a/W = 0.25
a/W = 0.6
J
(
K
J
m
2
)
4 5
Figure 93 a, Distribution of tensile stress o
22
ahead
of a propagation crack for Da =D, Da =10D, and
Da =20D and for three different ratios, a/W, of
initial crack length to specimen width; b, DCG
resistance curves for three different a/W ratios in
three-point bend specimen, with W=50 mm;
E/o
0
=500, n =0.10, v =0.3; D=200 mm. Source:
Xia, L. and Shih, C. F. 1995a. Ductile crack
growth. I: A numerical study using computational
cells with microstructurally based length scales.
J. Mech. Phys. Solids 43, 233259. Xia, L. and
Shih, C. F. 1995b. Ductile crack growth II. Void
nucleation and geometry effects on macroscopic
fracture behavior. J. Mech. Phys. Solids 43, 1953
1981. Xia, L. and Shih, C. F. 1996. Ductile crack
growth. III: Transition to cleavage fracture
incorporating statistics. J. Mech. Phys. Solids 44,
603639.
774 Failure of Metals
cleavage fracture. This local stress amplifica-
tion has been studied recently by Petti and
Dodds (2005b). These authors used a very sim-
plified model consisting in cylindrical inclusions
parallel to the crack front and extending over
all the specimen thickness. The effect of local
stress intensification due to the presence of duc-
tile cavities has been evidenced experimentally
by a number of authors; see, for example,
Carassou (1999) and Carassou et al. (1998).
These authors showed that, under given cir-
cumstances, cleavage cracks in A508 RPV
steel were initiated from small carbide particles
located around cavities initiated from larger
inclusions. This effect of stress intensification
on cleavage fracture will be more pronounced
at low temperature and in materials containing
a significant amount of large inclusions when
cleavage is controlled by the nucleation of
microcracks from particles. This influence of
ductile damage on cleavage fracture appears
therefore opposite to the softening effect
described earlier. Only detailed metallographi-
cal observations on the position and the nature
of the cleavage initiating sites can be used to
differentiate these two conflicting effects of
local ductile damage on the DBT behavior.
(ii) Sampling effect due to crack growth
The potential FPZ for an extending crack
now comprises two distinct zones depending
on the material stress history. Figure 94
provides a schematic illustration of these
zones, denoted as A and C. Material in the
unloaded zone A behind the current physical
position of the tip experienced severe stress and
strain fields without triggering cleavage frac-
ture. Due to the reduced blunting at the
location of the physical crack tip, the peak
value of opening mode stress develops at a
small distance (roughly the blunted opening)
ahead of the tip. Material located outside the
blunting region (zone C) experiences increased
stresses involving the generation of new micro-
voids and cracked carbides due to progressive
deformation. Catastrophic cleavage fracture is
eventually initiated in this zone not only
because the stresses due to crack growth are
increased but also because of an increasing
probability to contain larger microdefects
which were not sampled during the earlier
crack extension.
The probability to cleavage fracture is an
increasing function of the volume of material
experiencing large stresses (see Section
2.06.2.4). The volume of material hatched in
Figure 94 must be taken into account. This
introduces a correction of DCG (see Bru ckner
and Munz, 1984; Wallin, 1989, 1991a, 1991b,
1993). The basic assumptions for the DCG cor-
rection have been presented by Wallin (1989).
The volume increment due to both increase in
loading parameter K
I
=f(Da) as well as crack
growth, Da, leads to an increase of the Weibull
stress (see eqn [27]). The probability to fracture,
P
R
, can thus be written as
Ln
1
1 P
R
_ _
=
(f(Da))
4
K
4
0
2s
2
f
K
4
0
B
_
Da
0
(f(Da))
2
dDa
[142[
where K
0
is a normalizing value for the stress
intensity factor which can easily be calculated
under SSY conditions (Beremin, 1983), and o
f
is the fracture (cleavage) stress. B is the thick-
ness of the specimen.
The above expression of the DCG correction
proposed by Wallin (1993) is not unique.
Another DCG correction has been proposed
by Bru ckner and Munz (1984). Their expres-
sion can be written as
Ln
1
1 P
R
_ _
=
K
i
K
0
_ _
4
1
K
4
0
W
i
_
Da
0
(f(Da))
4
dDa
[143[
where K
i
is the value of the stress intensity
factor corresponding to the initiation of DCG,
and W
i
is a constant describing the size of the
active volume.
Equations [142] and [143] are quite similar.
The difference is that Wallin assumes that the
effective volume continues to grow as a func-
tion of (K
I
2
)
2
even after the onset of ductile
crack extension, whereas Bru ckner and Munz
assume that the size of the active volume is
constant during crack growth. In both cases,
eqns [142] and [143] require the evaluation of
the crack growth integrals. This means that the
A C
y
x
a
a +a
Figure 94 Sketch showing the PZs ahead of an
initial crack (a) and after crack growth (a Da).
The plastic wake left behind the propagating crack
is shown.
DBT in Ferritic Steels 775
R-curve must be known. In order to overcome
this difficulty, a simplified expression of the
DCG correction has been proposed by Wallin
(1993a, 1993b). If the DCG is independent of
K
I
, the DCG correction can be simplified and
written as
Ln
1
1 P
R
_ _ _ _
1,4
=
K
i
K
0
1
2Das
2
f
K
2
i
b
_ _
[144[
where b ~ x/(K
i
/o
f
)
2
defines the cleavage FPZ
size. When the crack growth is small or the R-
curve is relatively flat or both, eqn [144] can
easily be used to approximate eqn [142] by
replacing K
I
by K
i
.
(iii) Elimination of eligible particles
As already stated in Section 2.06.2.3,
enhancements of the Beremin model have
been proposed to account for continuous
nucleation of cleavage microcracks from parti-
cles such as carbides. The idea emerged from a
number of observations, in particular those
made by Chen et al. (1990), who showed that
the fracture of carbide particles is a continuous
process; see also Chen et al. (1996, 2003), Chen
and Wang (1998), and Wang and Chen (2001).
The microcrack blunts and generates a micro-
void if the conditions to nucleate a cleavage
microcrack from these particles are not met
(see eqn [16]). The nucleation of microvoids is
also statistically distributed. If the probability
to nucleate a microvoid is called P
void
i
, the
Weibull stress in eqn [27] must therefore be
modified and written as
s
w
=
n
i=1
1 P
i
void
_ _
s
i
I
_ _
mV
i
V
0
_ _1
m
[145[
where V
i
is the volume of the ith element and o
I
i
is the maximum principal stress in V
i
. Equation
[145] means that once a void has been nucleated
from a given particle, this particle cannot con-
tribute to cleavage fracture. This idea was
implemented by Koers et al. (1995).
Unfortunately, there are very few experimental
results in the literature dealing with quantita-
tive measurements of void nucleation rate (see
Section 2.06.3.4.2). This explains why, in most
cases, this elimination of eligible particles by
void formation is neglected.
(iv) Applications
In the early work by DEscatha and Devaux
(1979), the simulation of DCG was simply per-
formed with an uncoupled model using the Rice
and Tracey expression [69]. It was assumed that
fracture occurred over a critical distance D
(equal to the mesh size) ahead of the crack tip
when the calculated value of the void growth
reached a critical value, (R/R
0
)
c
. Crack growth
was simulated using the node release technique.
The value of (R/R
0
)
c
was determined from tests
on notched bars. It was shown later that this
technique gave consistent results when applied
to round cracked bars in A508 RPV steel
(Devaux et al., 1989). This pioneering work
was followed by many authors who used the
computational cell methodology initially pro-
posed by Xia and Shih (1995a, 1995b, 1996) to
model ductile crack extension, as described in
Section 2.06.3.7.3. Materials properties for this
computational cell methodology are limited.
They include, for the base material, Youngs
modulus (E), Poissons ratio (i), yield stress
(o
0
), and hardening exponent (n), or the actual
measured stressstrain curve, and for the com-
putational cells, D, f
0
, and f
c
. The strategy
which is generally used to calibrate these prop-
erties and these parameters, in particular D, f
0
,
and f
c
, is based on a series of finite element
analyses of conventional specimens. Very few
studies have attempted to use these parameters
identified from volume element specimens, such
as notched specimens, although there are a
number of exceptions (see, e.g., Tanguy, 2001;
Tanguy et al., 2002a; Bauvineau, 1996; Decamp
et al., 1997; Devillers-Guerville, et al., 1997).
The Weibull stress model is used to simulate
the initiation of cleavage fracture. The calculation
of the Weibull stress is based on the Beremin
model (see Section 2.06.2.3). The application of
this model to the situation corresponding to sig-
nificant DCG requires the definition of an
effective Weibull stress. Each part of the material
is subjected to a loading history, o(t), p(t) where t
is time. The probability of survival of each point
at time, t, is determined by the maximumvalue of
the loading parameter experienced by this point
during the time interval [0, t] (Tanguy, 2001;
Lefe` vre et al., 2002). The effective Weibull stress
is thus defined by
s
eff
(t) = max s
I
(t9) t9 (0. t). _ p(t9)>0 [146[
where o
0
(t)9 is the maximum principal stress.
The condition that the cumulative plastic strain
rate is positive, p(t9) >0, means that plastic
deformation must be active to trigger cleavage
fracture. In eqn [146], the correction due to
plastic strain which appears in the original
Beremin model (eqn [37]) is sometimes intro-
duced. The Weibull stress is thus defined as
s
w
=
_
PZ
s
m
eff
dV
V
0
_ _
1,m
[147[
Three applications of these advanced models
to predict the DBTbehavior are presented below.
776 Failure of Metals
In the first two applications, the emphasis is laid
on the comparison between experiments and cal-
culations. The third application which deals with
welds has been selected in order to illustrate how
the local approach to fracture can give an insight
into complex, but representative, situations.
1. HSLA steel. This application based on com-
putational cell methodology was made by
Ruggieri and Dodds (1996). These authors used
the results obtained on an HSLA steel
(o
0
=663 MPa) with a relatively low strain hard-
ening (UTS/o
0
=1.08). The material was tested
at 120
2
)
3PB specimens
a/W = 0.10
a (mm)
f
0
= 0.00025
Figure 95 Crack growth resistance curve in an
HSLA steel. Experimental results (Toyoda et al.,
1991); numerical simulation (Ruggieri and Dodds,
1996). Reproduced from Ruggieri, C. and Dodds,
R. H., Jr. 1996. A transferability model for brittle
fracture including constraint and ductile tearing
effects: A probabilistic approach. Int. J. Fract. 79,
309340, Copyright 1996, with kind permission of
Springer Science and Business Media.
DBT in Ferritic Steels 777
Figure 97 compares the experimental and the
theoretical results. The model gives a good
representation of the JDa curve, except for
low values of ductile crack extension where the
model tends to overestimate the ductile tearing
resistance of the material. This might be related
to the definition of crack growth since it is not
clear how in the calculation the extension corre-
sponding to crack blunting effect was taken into
account. Gao et al. (1999) have also calculated
the probability to cleavage fracture during crack
extension. Their results are also included in
Figure 97, where it is observed that the calcu-
lated values for P
R
are in reasonable agreement
with test results, although the calculated values
of J
c
at cleavage initiation tend to underestimate
the experimental results. This effect might be
partly reduced by using the strain correction,
which has to be applied in particular for large
crack extensions, as originally proposed in the
Beremin model (eqn [37]). However, due to the
simplicity of the model used by the authors, it
can be considered that these results validate the
use of this methodology to predict with a reason-
able accuracy the DBT behavior even after large
crack extensions.
3. Welds. In this example, it is assumed that a
crack is located in the weld metal and propagates
in this material parallel to the fusion line, but far
fromthe transition between the weld metal (WM)
and the base material (BM). Both undermatched
(o
0
WM
<o
0
BS
) and overmatched (o
0
WM
>o
0
BS
)
conditions are considered. The application of
the local approach to welds has already been
5 5.5
90% Conf. limits
Predicted
a/ W = 0.1
3.5
2.5
1.5
0.5
0.5
1.5
6 6.5
ln J
c
(kJ m
2
)
l
n
[
l
n
(
1
/
(
1
P
R
)
)
]
7 7.5
1
2
m = 15.6 ;
th
= 0
P
R
(
%
)
8
5
20
45
63
80
95
Figure 96 Calculated probability to fracture vs J-integral. Numerical simulation of crack growth resistance in
an HSLA steel. Reproduced from Ruggieri, C. and Dodds, R. H., Jr. 1996. A transferability model for brittle
fracture including constraint and ductile tearing effects: A probabilistic approach. Int. J. Fract. 79, 309340,
Copyright 1996, with kind permission of Springer Science and Business Media.
1200
1000
800
600
400
200
0
0 1 2 3 4 5 6
a (mm)
J
(
k
J
m
2
)
2Cr1Mo
CT (a /W = 0.6)
Plane strain model
D = 300 m, f
0
= 0.004 5
P
R
= 0.90
P
R
= 0.50
P
R
= 0.10
No cleavage
Experiments (Wallin, 1993)
Model prediction (m = 11.86;
= 2490 MPa;
th
= 0)
Figure 97 Crack growth resistance curve in a 2Cr1Mo steel. Experiments (Wallin, 1993) and numerical
simulations (Gao et al., 1999).
778 Failure of Metals
illustrated (Chapter 7.05). The focus here is on the
fracture resistance due to ductile tearing and the
onset of cleavage fracture. Moran and Shih
(1998) have used the computational cell model
to investigate the theoretical effect of mismatch
between the base and the weld metal.
Overmatched welds have the advantage that
the weld metal is stronger than the surrounding
base material, inducing thus a state of low con-
straint. Consequently, crack growth within the
weld is accompanied by extensive plastic defor-
mation both in the weld and in the surrounding
base metal. Therefore overmatched welds are
very resistant to ductile tearing. However,
because the stress required to cause ductile tear-
ing is relatively high, it can reach levels
sufficient to initiate unstable cleavage fracture.
Therefore, cleavage fracture can occur after
very little DCG. The resulting fracture tough-
ness of the assembly can thus be low even
though it is resistant to ductile tearing.
Lowering the strength of the weld to produce
undermatching confines the plastic deforma-
tion into the weld. Both the tearing resistance
and the stress ahead of the crack are relatively
low. As a result, cleavage fracture is less likely
to occur. Therefore, a crack can grow stably
over considerable distances before catastrophic
cleavage occurs. Hence, the critical fracture
toughness of undermatched weld specimens
can be higher than that of overmatched compo-
nents although the JDa resistance curve is
lower, as schematically shown in Figure 98.
This qualitative analysis of mismatching effect
was supported by the numerical simulations per-
formed by Moran and Shih (1998). These authors
also used the computational cell methodology.
As their analysis is purely parametric, it is not
essential to give here the values of the parameters
used in their simulations. Moran and Shih
investigated the importance of the relative thick-
ness of the weld metal. Here we refer only to
situations where this thickness is such that the
contour maps of the maximum principal stress
ahead of the growing crack located in the middle
of the weld can hit the interface between the weld
and the base metal (i.e., typically a few milli-
meters for J values of the order of 500kJ m
2
).
The results of the numerical calculations are
schematically shown in Figure 98 where the
JDa resistance curves corresponding to under-
matching and overmatching situations are
drawn. The effect of increasing the initial
value of the void volume fraction, f
0
, is dis-
played. In Figure 98, we have also reported the
value of the cleavage fracture toughness corre-
sponding to a probability of 50%. From this
figure, we see that the overmatched cases lead
to critical crack length, which is larger when the
initial volume fraction of defects increases.
Both of the undermatched cases are much
more resistant to cleavage fracture. The cleaner
weld has a critical crack length which is smaller
than in the dirty weld. As stated previously, this
can be anticipated since the crack in a material
containing large defects can grow by ductile
tearing under a low stress, and this produces a
relatively low Weibull stress. However, the
obvious disadvantage is that the JDa curve is
also lower, as shown in Figure 98.
These simulations cannot be easily compared
to experiments since there are no detailed
results in the literature providing a complete
set of data for ductile tearing and the subse-
quent failure by cleavage in welds. This is why
the conclusions drawn from this third applica-
tion of advanced models to predict the DBT
behavior remain speculative.
2.06.4.3 DBT under Charpy V impact testing
2.06.4.3.1 Introduction
Modeling the Charpy impact test is a challen-
ging issue, since several aspects of this test
require a detailed analysis. They include: (1) the
inertial effects, (2) the complexity of the loading
at high impact rate (5 ms
1
) and the boundary
conditions, (3) the effect of high strain rates on
constitutive equations, (4) the nonisothermal
character of the test, (5) the 3-D aspect of the
fracture behavior, in particular the tunneling
effect associated with DCG preceding cleavage
fracture above the lower-shelf temperature, and
(6) the competition between ductile and brittle
fracture. However, recent developments in the
instrumentation of the Charpy test largely facil-
itates the task. Moreover, recent developments
in the local approach to fracture have also
evolved the Charpy V impact test from a purely
Overmatched
Overmatched
P
R
= 0.50
a (mm)
J
(
k
J
m
2
)
Undermatched
Undermatched
f
0.2
f
0.1
f
0.1
<
f
0.2
Figure 98 Schematic variation of J-integral as a
function of crack length for two weld configurations
(under- and overmatched) and two values for the
initial value of the inclusion volume fraction.
DBT in Ferritic Steels 779
quality control test to an evaluation tool for
structural integrity assessment of materials. As
already stated in the introduction of this part, a
recent conference has been devoted to this test
(Franc ois and Pineau, 2001).
It is out of the scope of this chapter to review
in detail the models used to simulate the Charpy
test and to calculate the Charpy energy (CVN).
This is already presented in Chapter 7.05. This
review was largely based on the work by
Tanguy (2001). Further details can also be
found elsewhere (Tanguy et al., 2002a, 2002b,
2002c, 2005a, 2005b). Here the focus is laid on
salient features in modeling Charpy impact
tests. Besides the work by Tanguy (2001),
other studies should also be mentioned, in par-
ticular those by Rossoll (Rossoll, 1998; Rossol
et al., 1999, 2002a, 2002b) and those published
by the Freiburg group (Bo hme et al., 1992,
1996; Schmitt et al., 1994, 1999, 1998; Sun
et al., 1995). Other theoretical studies but with-
out detailed comparisons with experiments
should also be indicated (Mathur et al., 1993;
Tvergaard and Needleman, 1988, 2000;
Needleman and Tvergaard, 2000). After the
analysis of the salient features arising from
those studies, an attempt is made to underline
how modeling Charpy V test can be used to
investigate the fracture properties of materials
under specific conditions, in particular those
found with irradiated materials.
2.06.4.3.2 Modeling Charpy V-notched
impact test salient features
Inertial effects have been shown to affect fail-
ure only at very low temperatures in the lower
shelf regime and not in the transition region
where plastic deformation is sufficiently large
to damp the oscillations on the loaddisplace-
ment curve recorded in an instrumented
Charpy test (Tvergaard and Needleman, 1988;
Tahar, 1998; Rossoll et al., 1999). The impact
Charpy test can thus be simulated under quasi-
static conditions when dealing with results
obtained in the upper part of the DBT curve
and at the upper shelf.
The Charpy test specimen is essentially 3-D.
Finite element modeling must therefore
account for this effect. Contact between the
striker and the support must also be taken
into account using a friction coefficient.
Local strain rates as large as 10
3
s
1
are cal-
culated during the deformation of the notch
under impact and during crack propagation.
Modeling the mechanical response of the speci-
men requires to properly capture the strain rate
effect in the constitutive equations of the mate-
rial. Many authors have used the Cowper
Symonds law to represent the strain rate effect.
However, this representation assumes that the
strain rate effect is the same over all the rate
range encountered in these tests. This is the
reason why in Tanguy (2001) the flow strength
of the material was expressed as a function of
temperature, T, and plastic strain, p, with two
isotropic components as:
s
y
(p. T) = R
0
Q
1
[1 exp(b
1
p)[
Q
2
[1 exp(b
2
p)[ [148[
where the parameters R
0
, Q
1
, and b
1
are tem-
perature dependent, while Q
2
and b
2
are
constant. The equivalent plastic strain rate, p,
is given by a viscoplastic flow function
_ p =F(o
e
o
0
) expressed as
1
_ p
=
1
F9
=
1
_ e
1
1
_ e
2
[149[
with
_ e
i
=
s
e
s
0
K
i
_ _
n
i
i = 1. 2 [150[
where o
e
is the von Mises equivalent stress. The
strain rate p, e
1
, and e
2
are representative of one
of the following deformation micromechan-
isms: (1) Peierls friction forces acting mainly at
moderate strain rates, and (2) phonon drag
effect prevailing at higher strain rates
(,10
3
s
1
).
The comparison between the experimental
and the calculated loaddisplacement curves is
generally used to test the quality of the simula-
tions. This is however a global method which
does not guarantee that the local stressstrain
fields are correctly calculated. In Tanguy
(2001), a special effort was made to measure
the local strains around the notch of the
Charpy specimen by using a recrystallization
technique.
In most studies, ductile damage is simulated
using the Gurson model as extended by
Tvergaard and Needleman (see Section
2.06.3.4), referred hereafter as the GTN
model. However, Tanguy (2001) used also the
Rousselier model (see Section 2.06.3.4.4) to
simulate ductile fracture in impact Charpy
tests. The original Rousselier model was mod-
ified to account for strain rate effect (Tanguy
and Besson, 2002). The damage parameters
appearing either in the GTN model or in the
Rousselier model were calibrated to simulate
the ductility of notched tensile bars. The initial
volume fraction of inclusions initiating cavities
was determined from detailed metallographical
observations and chemical analysis (Tanguy,
2001).
Cleavage fracture was simulated using the
Beremin model (eqn [27]). A special care must
780 Failure of Metals
be taken when computing the probability to
fracture occurring after some crack extension
since, for a growing crack, propagation pro-
duces the unloading of the material left behind
the advancing crack tip. The equivalent stress
must be defined as in eqns [146] and [147] to
calculate the probability to fracture. Here it
should be added that interrupted tests which
are difficult to perform have proved to be extre-
mely useful to compare the observed ductile
crack extension and the measured crack growth
(Tanguy, 2001; Tanguy et al., 2005a).
This methodology was applied to an A508
RPV steel to predict the Charpy V transition
curve. The Charpy energy, CVN, correspond-
ing to a failure probability of 10%, 50%, and
90% is plotted as a function of the test tempera-
ture (Figure 99). The normalizing stress o
u
of
the Beremin model was assumed to remain con-
stant. The predictions are satisfactory up to
T=80
C. The
energy spent in crack propagation largely
increases when the temperature is increased.
2.06.4.3.3 Other applications
Similar models can be applied to multiple-
material components, such as those found in
welds. The notch of the Charpy specimens can
be machined at different positions from the
fusion line to test the various regions of the
material, that is, the weld metal, the HAZ
which is often more brittle than the weld
metal, or the base material. The simulations
must therefore incorporate the constitutive
equations for at least three materials. The lim-
ited dimensions of the weld metal and the HAZ
makes difficult to extract specimens from these
zones in order to determine the corresponding
mechanical properties. Thermal welding cycles
are generally applied to bulk specimens by
using Gleeble-like thermomechanical simula-
tors (see, e.g., Bilat et al., 2006) before
mechanical testing. The constitutive equations
and fracture criteria are introduced in the finite
element modeling of multiple-material Charpy
200
200
0 0
100 100
50
(a) (b)
50
150 150
200 200
250 250
150
150
100
100 50
50
0
0
50
50
10%
10%
USE
Temperature (C) Temperature (C)
u
= Cte
u
(T )
C
V
N
(
J
)
C
V
N
(
J
)
50%
90%
90%
Figure 99 A508 RPV steel. Prediction of the Charpy V transition curve assuming a constant value for the
parameter o
u
or a temperature-dependent o
u
(USE: upper shelf energy) (Tanguy et al., 2005a, 2005b).
DBT in Ferritic Steels 781
specimen to simulate the mismatch effects pre-
viously discussed.
Charpy V test results are also widely used in
the nuclear industry for surveillance program of
RPV embrittlement by neutron irradiation.
Service life extension of nuclear power plants
and more stringent safety requirements increase
the request for smaller test specimens than the
typical 10 10 mm
2
Charpy test pieces, such as
subsized Charpy specimens. This rises the pro-
blem of the transferability of fracture criteria.
In a recent study, it was shown that reasonable
predictions of the Charpy energy measured on
subsized Charpy specimens could be obtained
using the Beremin cleavage model at low tem-
perature and the Rousselier model at the upper
shelf (Poussard et al., 2002). See also the study
by Schmitt et al. (1998).
Charpy V impact testing is also largely used to
detect and to monitor the effect of irradiation
embrittlement in RPV steels. A shift of the
DBTT, DT, is observed, as indicated earlier, and
as schematically shown in Figure 19. Very
recently, an approach based on the simulation of
Charpy V-notch specimens has been developed to
predict the temperature shift in A508 RPV steel
(Tanguy et al., 2006). These authors showed that
the increase in the DBTTwith the neutron fluence
can be well predicted using the Beremin theory.
The results have already been reported in Table 5.
All these examples show that it appears now
possible to adapt the Charpy impact test from a
purely technological test to a more quantitative
tool for the evaluation of the fracture properties
of materials and components.
2.06.5 CONCLUSIONS
A unifying approach of fracture must start
with a physical model, not just a phenomeno-
logical model. In this chapter, an attempt has
been made to illustrate the benefits of the
micromechanical modeling of fracture. In par-
ticular, it has been shown that sophisticated
models have now been developed which capture
the influence of a large number of physical
parameters describing the detailed microstruc-
ture of materials, under complex loading
conditions.
The micromechanical approach to fracture is
far more complex than the global approach
which assumes that fracture can be described
by a single (eventually two) loading parameter,
such as K or J. In particular, it requires for
metallic alloys detailed metallographical mea-
surements (grain size, grain or packet
orientation, second-phase volume fraction, par-
ticle shape, etc.), and also advanced FEM
calculations. Contrary to the global approach
in which the material is considered as a black
box to which macroscopic, statistically based
failure criteria apply, the local approach to
fracture has largely contributed to the issues
related to the transferability of laboratory test
results to components in case of size or con-
straint effects. It has also allowed to model
complex macroscopic phenomena, such as the
nonexistence of a unique crack growth resis-
tance curve for ductile rupture or the
beneficial warm-prestressing effect observed
when a material is prestressed above the DBT
curve and then loaded below this transition
curve.
The micromechanical approach to fracture is
not really new. At the turn of the millennium,
one can wonder how much progress has been
made over the past 10 or 20 years. It is clear that
the modeling of cavity formation, cavity
growth, and coalescence in the frame of conti-
nuum solid mechanics has been tremendously
improved. Similarly, it is now well accepted that
cleavage fracture toughness is largely scattered
and is specimen size dependent. These pro-
gresses have been possible thanks to the
prodigious development of numerical methods.
The introduction of new experimental techni-
ques, such as in situ mechanical tests, in situ
observations, and tomography has also largely
contributed to these improvements.
Many results derived from the micromecha-
nical approach to fracture still remain
speculative in the absence of a sufficiently
large basis of experimental verifications. It is
well to remember that 20 years ago the unique-
ness of the ductile crack growth resistance curve
was considered as acquired in spite of all the
1
Exp. tests 60 C
Calculated P
R
a
max
(mm)
C
V
N
(
J
)
(60 C)
(40 C)
0.50
0.50
0.10
0.10
0.90
0.90
A508 C1.3
0
0
20
40
60
80
100
120
140
160
180
200
220
2 3 4 5
Figure 100 A508 RPV steel. Variation of the
Charpy energy (CVN) with crack extension Da
max
from the notch. Experimental results at 60
C.
Calculated probabilities to cleavage fracture
(P
R
=0.10, 0.50, 0.90) for two temperatures (60
and 40