Int J Fract (2012) 174:177–202

DOI 10.1007/s10704-012-9689-z

ORIGINAL PAPER

Ductile failure behavior of polycrystalline Al 6061-T6

A. Ghahremaninezhad · K. Ravi-Chandar

Received: 19 August 2011 / Accepted: 14 February 2012 / Published online: 15 March 2012
© Springer Science+Business Media B.V. 2012

Abstract Ductile failure in polycrystalline aluminum 1 Introduction

alloys is explored through uniaxial tension and notched
tension experiments. Specimens obtained through tests Research in the field of ductile failure has been driven
interrupted at various stages of deformation and fail- by the need to define the onset and progression of fail-
ure evolution are examined through microscopy to dis- ure in applications ranging from sheet metal forming to
cern the mechanisms of failure and to evaluate the local structural integrity assessment. Failure in these mate-
strain evolution quantitatively. Fractographic observa- rials is considered to be ductile if final failure is pre-
tions are used to identify the onset and evolution of ceded by significant inelastic deformation; in many of
damage processes during deformation and failure of these materials ductile failure is typically considered to
these aluminum alloys. Local strain levels are esti- occur by the progressive nucleation, growth and coales-
mated from measurements of the change in grain size cence of damage. While such damage must have atom-
with deformation and used to indicate that the local istic origins, in polycrystalline materials, such atomic
values of failure strains are likely to be much larger scale processes are often mediated by the defect struc-
than that estimated from strains averaged over charac- ture in these materials at a much larger length scale:
teristic specimen dimensions such as the gage length typically, microstructural features that are of the size
or the specimen diameter. Lower bound estimates of of grains and/or second phase particles or inclusions
the failure strain at moderate triaxiliaties are obtained (∼10−5 m) have a great influence on the evolution of
from the experiments. damage. Therefore, two approaches are used in mod-
eling ductile failure. First, micromechanical model of
Keywords Damage nucleation · Failure mechanism · damage is considered in the framework of classical con-
Local strain measures · Strain-to-failure tinuum mechanics. Second, a much simpler phenom-
enological model is postulated whereby the existence
of a macroscopic strain-to-failure is postulated (albeit
motivated by micromechanics) and calibrated through
Electronic supplementary material The online version of
this article (doi:10.1007/s10704-012-9689-z) contains
detailed experiments.
supplementary material, which is available to authorized users.

A. Ghahremaninezhad · K. Ravi-Chandar (B) 1.1 Void nucleation, growth and coalescence model
Center for Mechanics of Solids, Structures and Materials,
The University of Texas at Austin, Austin,
TX 78712-0235, USA Nucleation, growth and coalescence are taken to form
e-mail: kravi@mail.utexas.edu the three different stages of ductile failure. First,

123
178 A. Ghahremaninezhad, K. Ravi-Chandar

micovoids nucleate at second phase particles at some of Hancock and Mackenzie (1976) when the stress state
stage during the loading history (Tipper 1949; Puttick in the ligament between two voids is shear dominant.
1959). These second phase particles are a product of In both cases, this localized phenomenon results in a
the manufacturing process in which nucleating agents sharp drop in the stress-strain response of the ductile
and alloying elements are added to metals; the nucle- materials, resulting in final failure. The specific details
ating agents promote nucleation of grains and control of the onset of nucleation, the nature of the growth
grain size distribution while the alloying elements pro- and the mode of coalescence are, of course, material
vide specific microstructural features such as precip- dependent. This class of micromechanical models that
itates, and multi-phase structure, that can be used to incorporate void nucleation, growth and coalescence
develop specific material characteristics such as high are typically called Rousselier or Gurson–Tvergaard–
yield strength, wear or corrosion resistance. However, Needleman (GTN) models and the general framework
in many cases, these second phase particles act as nucle- of these models is now rather well-established. Tver-
ation sites for damage during large inelastic deforma- gaard (1990) and Benzerga and Leblond (2010) provide
tion; this nucleation can be either through particle frac- comprehensive reviews of this approach to modeling of
ture or by decohesion at the particle-matrix interface, ductile failure. Experimental aspects of ductile failure,
depending on the flow properties of the matrix, the particularly in the 6xxx series of aluminum alloys are
particle size and morphology, the stiffness mismatch reviewed by Lassance et al. (2007).
between the matrix and the particle, and the local stress Calibration and validation of these models have
state. These two modes of nucleation are illustrated in been attempted by numerous investigators (see for
the micrographs of two model materials in Figure 1 example, Brocks et al. 1996; Dong et al. 1996;
of Babout et al. (2004). Matrix-particle interface dec- Decamp et al. 1997; Pardoen et al. 1998; Steglich and
ohesion in a material made of soft matrix and hard Brocks 1998; Benseddiq and Imad 2008; Nielsen 2008;
particles results in debonding at the interface while Simar et al. 2010). However, this process is inher-
particle fracture occurs in a material made of higher ently non-unique, since many different combinations
strength matrix and hard particles. Many investigators of parameters appear to capture the softening part of
have performed experiments and created model simu- the material/structural response (Bernauer and Brocks
lations of the nucleation problem in order to provide a 2002). In this paper, attention is focused on quantitative
quantitative measure of the stress at nucleation. Second, assessment of the microscopic deformation and failure
voids that are either present in the material initially or mechanisms in order to evaluate the applicability of
nucleated during the first stage, grow in size with con- micromechanical models of ductile failure in the Al
tinued macroscopic deformation; note that this growth 6061-T6 alloy.
must occur at the submicroscale by slip in the lattice of
polycrystalline metals. However, growth of the voids is
typically evaluated by considering a continuum plas- 1.2 Strain-to-failure models
ticity model of the deformation as in the analysis of
McClintock (1968), and Rice and Tracy (1969). Mac- Due to the complexity of modeling ductile fracture on
roscopic response of the material corresponding to such the micro-scale, simpler macroscopic strain-to-failure
growth of voids is obtained using homogenization the- models have been introduced over the past four decades
ory (for example, Gurson 1977), and exhibits softening to facilitate fracture prediction. These models usually
of the material response with continued deformation. contain several material parameters that need to be
The third and final stage of ductile damage is the coales- calibrated from experiments; strain-to-failure is deter-
cence of voids by plastic strain localization in the liga- mined with respect to some characteristic specimen
ments between the voids in the form of internal necking dimension such as the gage length or cross-sectional
or shear bands; at the onset of coalescence the defor- area. With the advent of numerical tools in recent
mation is localized into a thin layer of highly strained years, the simulations of the experiment have also
material. Internal necking in the intervoid ligament been widely used in calculating the strain at the onset
occurs as shown in Figure 3 of Puttick (1959) when of failure. However, these methodologies of obtain-
the stress state in the ligament is predominantly ten- ing strain-to-failure suffer from strong dependence on
sile, and shear localization occurs as shown in Figure 7 specimen length scale or discretization scheme, and do

123
Ductile failure behavior of polycrystalline Al 6061-T6 179

Table 1 Material constants calibrated from strain measurement in Sect. 3; the macroscopic strains are measured with
based on specimen dimensions for Al 6061-T6 (reproduced from the digital image correlation method and used to char-
Lesuer et al. 2001)
acterize the constitutive response of the material. The
Material constant D1 D2 D3 D4 D5 failure behavior is characterized by performing inter-
Al 6061-T6 −0.77 1.45 −0.47 0.0 1.60
rupted tests on notched tension tests, and exploring the
development of strain at the level of grains and attempt-
ing to characterize the damage evolution quantitatively.
Quantitative optical microscopy is used to extract local
not represent the intrinsic failure characteristics of the strain estimates at the level of the grains. These meth-
material. In this section we discuss the widely used ods as well as the results on Al 6061-T6 are described
Johnson–Cook fracture model as an example of this in Sect. 4. Based on the experimental identification of
class of models. the development of failure, we perform numerical sim-
The Johnson–Cook model (Johnson and Cook 1985) ulations of the uniaxial and notched tension tests. The
is an empirical model that is written in the following primary aim of these simulations is to be able to iden-
form: tify the stress-state in the regions where failure was
    
σm ε̇ observed. This enables the formulation of a failure cri-
ε f = D1 + D2 exp D3 1 + D4 ln
σeq ε̇0 terion as discussed in Sect. 5.
 ∗

× 1 + D5 T (1)
where ε f is the strain-to-failure, σm /σeq is the ratio of 2 Material
the mean stress to the von Mises effective stress and
is termed stress triaxiality, ε̇/ε̇0 is the dimensionless Attention is focused in this paper on specimens of Al
strain rate with ε̇0 = 1.0 s−1 , T ∗ is the homologous 6061-T6. The nominal chemical composition of the
temperature and D1 − D5 are material constants that are aluminum alloy is listed in Table 2. In the T6 condition,
calibrated from experiments at the macroscale. We will the material is peak-aged with a coherent distribution
examine the applicability of the Johnson–Cook model of fine precipitates of Mg2 Si within the grains; this
to Al 6061-T6 through measurements at different gage structure results in a material with a very high yield
lengths, all the way down to the size of grains. The strength, but very low strain-hardening. In order to
material constants, D1 − D5 , conventionally calibrated reveal the grains, sections of the specimen were pre-
through the strain measurements based on specimen pared using standard metallographic polishing tech-
dimensions for Al 6061-T6 are listed in Table 1. niques. The specimens were mounted in epoxy and wet-
In this paper, the results of an investigation into polished starting from 320-grit size and going down
the elastic-plastic response of Al 6061-T6 and the to 1200-grit size sand paper. Special care was taken
development of damage leading up to the point of not to apply excessive force on the specimens so as to
final failure are reported. Macroscopic as well as prevent the polishing compound from embedding into
microscopic responses are investigated in order to soft aluminum matrix. In the next stage, an abrasive
facilitate model building. Uniaxial tests and notched diamond paste with particle diameter 3 µm and 1 µm
tension tests were used in order to vary the degree of was used with a lapping cloth, and finally a colloidal
triaxiality significantly. Further investigations on duc- suspension with 0.05 µm diameter silica particles was
tile failure under shear dominated loading conditions used to achieve the required metallographic finish. The
will be reported elsewhere. The basic material charac- specimens were then etched, first in 2% aqueous solu-
teristics are described in Sect. 2. The results of uniaxial tion of sodium hydroxide for about 55 s, and subse-
and notched tension tests on Al 6061-T6 are discussed quently color-etched with Weck’s (Weck and Leistner

Table 2 Chemical
Wt% Si Fe Cu Mn Mg Cr Zn Ti Others Al
composition of Al 6061-T6
Min. 0.4 – 0.15 – 0.8 0.04 – – –
Max. 0.8 0.7 0.40 0.15 1.2 0.35 0.25 0.15 0.15 Bal.
Source: Alcoa

123
180 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 1 Microstructure of Al (b)

(a)
6061-T6 in the a x − y, b
x − z and c y − z planes.
The (x, y, z) directions are
the rolling, transverse and
thickness directions,
respectively. The
distribution of the grain size Particle
in the three directions is x x
shown in d
y z

(c) (d) y-z, mean = 15 μm

0.3
x-z, mean = 14 μm
0.25 x-y, mean = 39 μm

Frequency
0.2

0.15

0.1
z 0.05

y 0
0 20 40 60 80 100
Grain Size - μm

1996) alkaline solution (4% potassium permanganate orientation, if any, of different grains. In addition to
+1% sodium hydroxide) for about 15 s to obtain the the grains, the optical micrographs also reveal the dis-
contrast in the microstructure. This polishing and etch- tribution of second phase particles; one such particle is
ing process allows excellent visibility of the grains and identified by the highlighted circle in Fig. 1b. The com-
grain boundaries.To our knowledge, papers published position of these particles was identified through EDS
in the literature on Al 6061-T6, have not exploited such in a scanning electron microscope to be composed pri-
an etching process to identify the grain structure or to marily of iron; these particles are added in the solidifi-
extract estimates of the strain at the grain level. About cation process. Quantitative estimates of the amount of
40–50 images of each section were taken in a Nikon second phase particles were obtained by thresholding
Eclipse optical microscope at magnification of 200× and image analysis methods in NIH ImageJ software.
in order to quantify the initial grain size distribution in The average size of equivalent ellipses is in the range
the material. The initial microstructure of the Al 6061- of 1–5 µm; the average volume fraction of the particles
T6 in three sections x − y, x − z and y − z is shown in was found to be 0.012 and the particle spacing (defined
Fig. 1a–c. The x, y, z directions correspond to the roll- as the mean distance between nearest neighbors) was
ing, transverse and thickness directions respectively. determined to be about 25 µm.
The grain dimensions in y and z directions are cal-
culated for each section as the distance between two
intersecting points of a grain with the lines parallel to y 3 Experiments
and z directions. The distributions of the grain dimen-
sions on the three sections are plotted in Fig. 1d. It 3.1 Uniaxial tension test
is evident from distributions of the grain size that the
grains are nearly equiaxed in the rolling plane (x − y Uniaxial tension tests were performed on standard dog-
plane) with a mean size of 39 µm. The average grain bone type specimens in order to characterize the elastic-
size in the z-direction is about 14 µm. This grain size plastic behavior of the materials. The geometry of the
distribution suggests planar isotropy in the microstruc- specimens used in the experiments is shown in Fig. 2.
ture; it should be noted that this anisotropy is morpho- Specimens corresponding to the test geometry were cut
logical and does not account for the crystallographic by electric discharge machining (EDM) from 2.44 mm

123
Ductile failure behavior of polycrystalline Al 6061-T6 181

R6.35 400

350
6.35 9.52

Nominal Stress - MPa

300

31.75 36.07 250

107.95 200

150

Fig. 2 Schematic diagram of the uniaxial tension specimens 100 Al6061-T6-002-0

(dimensions in mm) Al6061-T6-001-A-90
50
Al6061-T6-Tension-45-02
0
0 0.05 0.1 0.15 0.2
thick rolled sheets of Al 6061-T6. These tests were run Normalized Gage Displacement - Δ /L
on an Instron 4482 universal testing machine at a cross-
head speed of 0.254 mm/min, resulting in quasi-static Fig. 3 Nominal stress (force/initial cross-sectional area) versus
loading at a strain rate of 1.18E-04 s−1 . The gage dis- gage displacement (Δ) normalized by gage length (L) response
at 0◦ , 45◦ and 90◦ to the rolling direction for dogbone specimens
placement was measured using a variation of the digital
made of Al 6061-T6
image correlation technique called Q4-DIC (Besnard
et al. 2006). In this method the region of interest is dis-
cretized in the same way as in the finite element method
and similar continuous shape functions are used to speckle pattern for each specimen was applied to the
interpolate the displacement within the element. The surface of the specimen in the region of interest. It has
pattern matching between subsequent images is based been shown that a white background paint with a fine
on the conservation of optical flow; then the error func- black speckle gives the best results. A CCD camera was
tional is minimized using a Newton–Raphson method, used to obtain digital images of the specimen through-
incorporating the shape functions. In contrast to the out the test at different stages of the loading. The strain
traditional DIC (Sutton et al. 1983), the continuity of in the specimen was estimated as the average over a
displacement field is enforced in this version of the gage length of 25.4 mm in the uniaxial tension speci-
DIC. One key advantage of this method is that by hav- mens. Load data were acquired as a function of time
ing the same discretization scheme for the measure- and crosshead displacement using the load cell on the
ments and the numerical simulations, the error stem- Instron testing machine. The two measurements were
ming from intermediate projection steps are reduced then correlated to obtain the stress versus strain varia-
(Réthoré et al. 2007). Furthermore, the accuracy of tion in the specimen. The nominal stress (force/initial
the traditional image correlation technique degrades cross-sectional area) versus normalized gage displace-
where there is a strong gradient or discontinuity in ment is plotted in Fig. 3 for the Al 6061-T6; in an effort
the displacement field such as in the vicinity of strain to characterize the anisotropy, specimens cut at 0◦ , 45◦
localization and cracks. This issue can be mitigated and 90◦ to the rolling direction were considered. Uni-
in the Q4-DIC approach where the kinematics can be form deformation in the gage section is terminated at
enriched with discontinuous and singular functions, as the Considère point (the point of peak load in Fig. 3); for
in extended finite element method, to account for these the material considered here, this occurs at εn = 0.08.
types of irregularities in the full field analysis (Réthoré The true stress versus true strain response of Al
et al. 2007). This method was implemented in MAT- 6061-T6 up to the onset of necking can be represented
LAB and was used to analyze the displacement and by a power law as follows (in MPa)
strain fields in experiments. The surface of the speci-
σ = 322 (1 + ε/0.0419)0.141 (2)
mens was spray painted to form a random speckle pat-
tern; the speckle size, pattern, contrast, illumination, The low strain-hardening ability of the material makes
and the resolution are among the factors that affect the it susceptible to strain localization as we will see in
quality of the displacement calculation in digital image the following sections; therefore, it is clear that the
correlation. In order to obtain optimum results sev- stress-strain curve can be obtained only up to onset
eral different combinations were examined and the best of localization from these dogbone tensile specimens;

123
182 A. Ghahremaninezhad, K. Ravi-Chandar

extrapolating the curve to larger strains by curve fitting Table 3 Lankford r -values at different orientations to the rolling
is fraught with inaccuracies that cannot be estimated. direction
In order to extend the characterization of material r0 r45 r90
response beyond the point of localization, a few meth-
r -values 0.57 0.60 0.58
ods, such as the Bridgman correction factor (Bridgman
1964) torsion tests, and Bridgman’s ‘refigured’ com-
pression tests, have been suggested in the literature.
Here, we choose an inverse method to calibrate the detailed grain-based analysis is required to verify this.
stress-strain behavior (Norris et al. 1978), where the It turns out that this anisotropy of the rolled sheets is
measured response in other loading configurations is very important in capturing the true response of the Al
compared to predictions from a finite element simula- 6061-T6 specimens in other loading conditions, such
tion. We note further that there are many other possibili- as the notched-tension test.
ties for fitting the measured stress-strain response, such
as the Voce law or micromechanically based models
that incorporate the different stages of hardening, but 3.2 Notched-tension tests
there is no particular advantage to these models when
a phenomenological theory of plasticity calibrated on Although the uniaxial tension test is perhaps the sim-
the basis of macroscopic measurements is used as the plest and most commonly used test for material char-
underlying basis for calibration. acterization, the onset of localized deformation at very
The sheets of aluminum used in these experiments low strain levels makes it ill-suited for constitutive and
are expected to exhibit anisotropy resulting from the failure characterization; specifically, one cannot deter-
rolling process used in their manufacture. The mea- mine the stress-strain variation beyond the Considère
sured nominal stress versus normalized gage displace- point. Furthermore, eventual failure occurs within the
ment curves from uniaxial tension tests performed localized region where strain evolves to much larger
on dogbone tensile specimens cut in the direction at levels in a more complex stress state. In order to push
0◦ , 45◦ and 90◦ to the rolling direction are shown in the evaluation of the stress-strain diagram to larger
Fig. 3; these curves for different orientations are very strain levels and to explore the failure behavior under
close to each other, implying that the planar isotropy different triaxiality conditions, we performed tension
maybe assumed. However, the local strains have to be tests on symmetrically notched flat specimens. It was
compared in all directions as well in order to verify estimated that the initial value of triaxiality (the ratio
planar isotropy and characterize the extent of anisot- of the mean stress to the equivalent stress, σm /σeq ),
ropy. In the tests reported here, the axial and transverse could be increased to about 0.577 in this configuration.
strains dεα and dεα+π/2 were measured for each ori- The geometry of the specimen is depicted in Fig. 4a.
entation α using DIC. Therefore, the ratio of the width Two notches of radius 4.06 mm were introduced in
to thickness strain increment, which is referred to as 25.40 mm wide strips cut from the aluminum sheet
Lankford’s r -value, defined as follows: in the direction parallel to the rolling direction. One
dεα+π/2 advantage of this geometry is that it enables reaching
rα = (3)
dεz significantly larger strain levels, but at the penalty of
where dεα and dεz are the strain in direction at an angle a nonuniform strain field; therefore, it is essential that
α from the rolling direction and in thickness direction, local strain fields be monitored using DIC. Another
respectively, can be measured from each test. The volu- advantage is that it could decouple strain localization
metric strain is assumed to be zero so that the increment triggered by plastic instabilities from damage processes
of strain in thickness
 direction
can be easily calculated that result in material separation so that attention may
as dεz = − dεx + dε y . The measured r -values for be focused on the latter.
the Al 6061-T6 are given in Table 3. It was observed In order to study the onset and growth of dam-
from strain measurements that the ratio dε y /dεx before age during deformation under different levels of stress
the onset of necking remains relatively constant sug- triaxiality and plastic strain, notched tension speci-
gesting that the anisotropy does not appear to evolve mens were loaded in the rolling direction and inter-
during uniaxial loading in this material. However, a rupted at different crosshead displacements. The nom-

123
Ductile failure behavior of polycrystalline Al 6061-T6 183

Fig. 4 a Schematic p2
diagram of the flat notched 25.4
specimen (dimensions in
mm). b Photograph of the
surface of the flat notched a b
specimen. p1 –p2 is used to x
define a gage length x2
(24.02 mm). DIC based
strain measurements are 152.40
R4.06
y
reported along lines labeled y1 y2
x1 − x2 and y1 − y2 .
Metallographic
examinations are performed
on sections cut along a–a
x1
and b–b
a b
p1

(a) (b)

400
between the peak load and ultimate failure, Specimens
350 A C and D corresponding to just at the brink of failure, and
B C Specimen E corresponding to a fully failed specimen.
Nominal Stress - MPa

300
E Contours of the axial and transverse strain as measured
250
D with DIC are overlaid on the speckled specimen sur-
200 face in the images shown in Fig. 6. These are from
150 Specimen E and correspond to the normalized cross-
100
head displacements of Specimens A, B and C. For the
specific implementation of DIC used in this work, the
50
displacements are measured on a square lattice with a
0
0 0.02 0.04 0.06 0.08
nodal spacing of 79.6 µm and strains are determined
Normalized Crosshead Displacement - Δ /L with a gage length of 159.2 µm. From these results, we
note the following:
Fig. 5 Nominal stress (force/initial minimum cross-sectional
area) versus crosshead displacement (Δ) normalized by gage • The surface strain over much of the specimen is
length (L) (see Fig. 4b for the definition of the gage length) quite small, on the order of 0.02, except in the vicin-
curves of Al6061-T6 flat-notched specimens interrupted at dif-
ferent stages during the deformation
ity of the notch where it increases to values in the
range of 0.35–0.40. Associated with this, there is
significant thinning of the specimen in this region.
inal stress (force/initial minimum cross-sectional area) Clearly, as expected, the strain εx at the notch is
versus normalized crosshead displacement curves from significantly higher than that in the middle region
five different tests that were interrupted at different dis- of the specimen.
placement levels are plotted in Fig. 5. Four other tests • Near the middle portion of the minimum cross sec-
under similar conditions were also performed, three to tion (x = 0, y = 0), the strain ε y ∼ = 0, indicating
complete failure and one that was unloaded just prior to a plane strain state, and thereby an elevated triaxi-
global failure. The variation in the load at final failure ality at this location.
was within 0.9%; the crosshead displacement at final • With increasing global displacement, the strain
failure also varied within 1.6%. This small variation is contours in Fig. 6 indicate the emergence of two
used as support for considering the specimens that were horizontal bands of elevated strains. For closer
unloaded just prior to failure as being at the “brink of observation of this region of large strains, this
failure”. Specimen A corresponds to unloading from experiment was repeated with the imaging sys-
the point of the peak load, Specimen B from a point tem tracking the deformation at very high spa-

123
184 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 6 Contour plots of

transverse (left column) and
axial (right column) strain
(ε y and εx , respectively) of
Al 6061-T6 flat notched
Specimen E corresponding
to three different normalized
crosshead displacements
corresponding to Specimens
A, B and C

tial resolution such that only the central 8 mm by • It is clear that strain levels as large as 0.4 develop
6 mm region of the specimen was observed (see near the notch tips (see Fig. 6). It is common prac-
dashed box in Fig. 4b). Contours of axial strain tice to take the largest strain measured with the
from this test are shown in Fig. 7. The variation of DIC to be the “failure strain” in the material. This,
the axial strain εx along the line x1 − x2 is shown however, ignores a couple of important aspects
in Fig. 8b, corresponding to different values of the of the measurement. First, DIC does not resolve
gage displacements indicated in Fig. 6. It is evi- the extremely high strains near the notch tip due
dent that a double humped variation is seen in εx to destruction of the speckles, decorrelation and
along the x direction. The double hump in this fig- other experimental effects associated with DIC.
ure is a clear indication of the formation of the Second, strain continues to evolve within localized
strain localization across the thickness direction of bands as material separation occurs with signif-
the specimen. Post-test examination of these spec- icantly larger local strains than that captured by
imens indicated that this localization occurred at the DIC. Although direct images acquired using
an angle of 56◦ to the loading direction and devel- the DIC system (at 1 frame per second) are typ-
oped in the thickness direction over the minimum ically not able to resolve the final events leading
cross section. This is in good agreement with the to the failure, visual observation clearly indicated
result from analytical calculation of Hutchinson the onset of cracking at the notch region(s), propa-
and Tvergaard (1980) for the formation of shear gating quickly towards the center of the specimen.
bands under plane strain condition in an elastic- Therefore, it is expected that the local strains at
plastic solid with strain hardening exponent of 0.1. the point of failure initiation are somewhat higher

123
Ductile failure behavior of polycrystalline Al 6061-T6 185

Fig. 7 Close-up view of axial strain (εx ) contour plots of Al 6061-T6 flat notched specimens corresponding to three different normalized
crosshead displacements of Specimens A, B and C

(a) 0.45 (b) 0.3 C

0.4
C 0.25 B
0.35
B
0.3 0.2
Strain

Strain

0.25
A
0.15
0.2
A
0.15 0.1
0.1
0.05
0.05
0 0
-10 -5 0 5 10 -3 -2 -1 0 1 2 3
y - mm x - mm
Fig. 8 Variation of the axial strain (εx ) along lines a y1 − y2 and b x1 − x2 for Specimens A, B and C

than the strain obtained by DIC at the last frame. to extend the constitutive characterization to larger
Additional local measurements must be made post- strain levels. In this process, it became evident that
mortem as discussed next in order to obtain a better plastic anisotropy played a key role in dictating the
estimate of the “strain-to-failure”. response. These are discussed in the following sections.
The notched tension response of the material is
explored further through two additional investiga-
tions; first, optical and scanning electron microscopy 4 Microscopic examination of deformation
observations and measurements were performed on and damage
specimens recovered after unloading to various gage
displacement levels in order to reveal the details of the Specimens A through D that were unloaded from dif-
deformation leading to failure. The specimens were cut ferent loading stages of the flat notched specimens and
to expose the planes y = 0 as well as y = 7.14 mm Specimen E that was taken to final failure were cut
(denoted as section a–a and b–b, respectively, in Fig. 4b along lines a–a and b–b indicated in Fig. 4b to reveal
and in the subsequent discussion) in order to evaluate the x − z planes (thickness direction), and prepared for
the local strains as well as to examine the onset and metallographic examination. Optical microscopy was
growth, if any, of damage. Second, numerical simula- used first for examining the grain evolution and scan-
tions were performed using J2 -plasticity theory in order ning electron microscopy was used next to examine

123
186 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 9 Micrographs of
x − z sections of Specimens
A–D at a–a (see
Supplementary Materials
for high resolution images
where each individual grain
can be visualized over the
entire image)

the extent of damage. The optical micrographs of the in the initial microstructure were estimated from the
section a–a of specimens A–D are shown in Fig. 9. images of the kind shown in Fig. 1. Next, the location
High magnification images of the region near the neck of each grain boundary was identified visually along
are shown in Figs. 10 and 11. An image of the bro- horizontal dashed lines such as Line-1 through 3 in
ken specimen E is shown in Fig. 12. Comparing these Figs. 10, 11, 12 and captured into a data file using a
images with the initial microstructure shown in Fig. 1d, MATLAB code. As is apparent from the micrographs
the elongation of the grains in the direction of tension in Figs. 10, 11, 12, and further verified by our analysis
and the decrease in grain size in the thickness direction of the grain measurement, grain rotation does not seem
at different stages of the deformation can be identified to be significant, and hence no corrections for grain
clearly. We develop a simple scheme to obtain a quan- rotations were performed. The variation of the grain
titative measure of deformation at the grain level and to size across the dashed lines in Figs. 10, 11, 12 was
correlate the grain level strain with the progression of obtained, and divided by the mean grain size in the z
damage or failure. The grain size measurements can be direction to obtain an estimate of the strain in the thick-
interpreted in terms of an average strain measure; note ness direction εt = ln(t/T̂ ), where T̂ is the mean initial
that this can be done only in a statistical sense since we grain size and t is the size of the deformed grain. The
have images of the grain only from interrupted tests. equivalent plastic strain can be estimated if the strain
This is accomplished using the procedure described in the transverse direction (y-direction) is determined.
by Ghahremaninezhad and Ravi-Chandar (2011): first, For section a–a, based on the DIC measurements, we
the average and standard deviation of the grain size assume that a condition of plane-strain is satisfied, and

123
Ductile failure behavior of polycrystalline Al 6061-T6 187

Fig. 10 High magnification

views of micrographs of the
neck region of section b–b
of Specimen C

hence the equivalent

√ plastic strain can be estimated to its standard deviation are obtained as ε̄t = ln(tˆ11 /T̂ )
be εeq = (2/ 3) ln(T̂ /t); for the section b–b, the DIC and εt = [(T̂ /T̂ )2 + (tˆ11 /tˆ11 )2 ]1/2 , respectively,
measurements suggest that we may assume a state of where T̂ is the standard deviation of the initial grain
uniaxial stress, and hence the equivalent plastic strain size; the corresponding equivalent plastic strain can be
can be estimated to be εeq = 1.6 ln(T̂ /t); note that estimated as indicated earlier. The red solid lines in
the effect of anisotropy of the material has been taken Fig. 13 indicate the equivalent plastic strain averaged
into account in obtaining this estimate of the equiva- over eleven grains. It is important to note a few points
lent plastic strain. The variation of the equivalent plas- regarding these strain measurements.
tic strain across lines marked as Line-1 through 3 in
Figs. 10, 11, 12 and a few other similar lines in other • There is only a small difference in the strain mag-
specimens is plotted in Fig. 13 as open symbols. Note nitudes between the sections in plane strain (a–a)
that there is quite a large scatter in the grain based and the sections in uniaxial stress (b–b), but the
strains; this, however, does not arise from errors in maximum observed strain levels in both locations
experimental measurements. Rather, this is inherent in are quite large, in the range of 0.5–1.
the process, since we divide the current grain size by • In the locations where shear bands appear, the local
the average initial grain size. In order to interpret this mean strains can be as large as 1 ± 0.3; in compari-
strain data better, the deformed grain size was aver- son, the maximum surface strains measured by DIC
aged over eleven neighboring grains, five on either just prior to failure are only about 0.5. This indi-
side, to obtain the mean deformed grain width, tˆ11 cates that the final stage of the ductile failure is a
and its standard deviation, tˆ11 . Then the best esti- localized process where large strains occur due to
mate for the true strain in the thickness direction, and flow localization in a thin layer of the material.

123
188 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 11 High magnification

views of micrographs of the
neck region of section a–a
of Specimen D

Fig. 12 High magnification

view of micrograph of
section a–a of Specimen E
showing a shear band along
which the failure seems to
occur (see Supplementary
Materials for a high
resolution image)

• There are large local fluctuations in the strain distri- it can be seen that the maximum strains are signif-
bution, and furthermore, the specimen to specimen icantly larger in Specimen C than in Specimen D.
variation appears to be quite high. With particular • It is interesting to contrast these strain measure-
reference to Specimens C and D, both of which ments with the usually postulated strain-to-failure
were unloaded just prior to reaching final failure, models such as the Johnson–Cook and related fail-

123
Ductile failure behavior of polycrystalline Al 6061-T6 189

Fig. 13 Variation of 1.5 1.5

equivalent strain across the (a) (b)
section a–a of Specimen C

Equivalent Strain

Equivalent Strain
(a) and Specimen D (b), the 1 1
section b–b of Specimen C
(c) and Specimen D (d) and
across Line-3 on the section
a–a of Specimen E (e). 0.5 0.5
Open symbols correspond to
strain estimate for each
grain; red line corresponds
0 0
to averaging the deformed 0 500 1000 1500 2000 0 500 1000 1500 2000
grain size over eleven Position - µm Position - µm
neighboring grains, five on 1.5 1.5
either side (c) (d)
Equivalent Strain

Equivalent Strain
1 1

0.5 0.5

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
Position - µm Position - µm
1.5
(e)
Equivalent Strain

0.5

0
0 100 200 300 400
Position - µm

ure models. We will return to this aspect in Sect. 4 in Fig. 12; it should be noted that final failure is
where we establish that these local strains are sig- due to the merging of failure along the two equiv-
nificantly larger than typical values used in the alent planes of shear banding, resulting in a zig–
Johnson–Cook model where the strain-to-failure zag shape of the fracture surface observed in these
is obtained from macroscale measurements. images.
• The emergence of the macro-shear bands across
the cross-section at an angle of 56◦ to the loading The spatial resolution of the optical micrographs of
direction, with highly localized strains within the the etched specimens is not sufficient to reveal the
bands, and the associated high level of stress tri- nucleation of damage at the location of the second
axiality in the specimen triggers failure processes phase particles. In order to investigate the formation
such as nucleation, growth and coalescence within and evolution of cavities and damage in the material,
this plane of localization. The resulting fracture is the same metallographic samples were examined in
slanted1 in the x − z plane as can be seen clearly a scanning electron microscope. The micrographs of
1 We do not go into the details of the formation of slant fracture
Specimen A did not reveal any differences from the
here except to note that it is certainly influenced by the low strain unstrained material; there is a distribution of second
hardening ability of the T6 condition. phase particles, but there is no indication of fracture or

123
190 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 14 SEM micrographs

of the section a–a of
Specimen B at different
magnifications showing
sporadic nucleation of
cavities at second phase
particles

debonding of these particles from the matrix. A typi- dependent. Scanning electron micrographs of the sec-
cal micrograph of section a–a of Specimen B is shown tions a–a and b–b of Specimens C and D are shown in
in Fig. 14a. The irregularly shaped white areas repre- Figs. 15 and 16; these show the regions where the final
sent the second phase particles. Sporadic breakage and failure is expected to develop. Clearly, a few scattered
debonding of a few second phase particles, especially cavities can be identified in these micrographs. Such
the larger particles could be seen; a high magnification features in the micrographs are typically interpreted as
image of a fractured second phase particle is shown evidence for the nucleation and growth of voids. How-
inset in Fig. 14b. However, it should be emphasized ever, before confirming such a conclusion, one must
that very few of the particles had fractured with most of rule out damage resulting from metallographic speci-
the second phase particles undamaged and still bonded men preparation procedures. The typical argument that
to the matrix. If particle fracture is associated with is put forward to exclude polishing damage is that other
nucleation of cavities for a ductile void growth model regions in the specimen have also been subjected to
(Tipper 1949; Puttick 1959), then correlating the grain the same polishing procedures, but did not reveal any
based strain measurements in the optical micrographs, such damage. In order to pursue this idea further, we
and the damage identified in the scanning electron consider quantitative comparison of the optical micro-
micrographs, these observations and measurements graph that reveals grain and grain boundaries with the
indicate that cavities are not nucleated until very high scanning electron micrograph of the same region of
strain levels on the order of one are reached. In con- the specimen. This comparison is shown in Fig. 17
trast, the typical value of ε N (mean nucleation strain where Fig. 17a is an optical micrograph, Fig. 17b is
of a normal distribution model introduced by Chu and a scanning electron micrograph of the same region and
Needleman 1980) used in literature is 0.3 (Brocks et al. Fig. 17c is a composite overlay of the SEM image
1996; Zhang 1996; also see Benseddiq and Imad 2008, over the optical image, with an opacity of about 0.5.
for a table with a collection of ε N used in the literature). With the overlaid image as a guide, attention is now
Note that the Specimen B corresponds to macro- focused on regions marked by the blue, dashed circle
scopic response that is well beyond the peak load and and square in Fig. 17b; high magnification images of
approaching failure conditions; the absence of signif- these features are presented in Fig. 18 where the grain
icant fracture or debonding of the second phase parti- boundaries can be easily identified. Clearly the entire
cles indicates that much of the descending portion of grain appears to be missing in Fig. 18a; if a nucle-
the nominal stress versus normalized crosshead dis- ated void were to grow, the grain surrounding it has
placement curve corresponds to plastic deformation to deform to accommodate such a large growth and
of the material and not to damage accumulation; of should be identifiable in the micrograph. However in
course, it is important to recognize that this conclu- the regions outlined by the dashed circles in the micro-
sion is applicable to Al 6061-T6 and could be material graphs, the entire grain is lost suggesting that these

123
Ductile failure behavior of polycrystalline Al 6061-T6 191

Fig. 15 SEM micrographs

of section a–a of a
Specimen C and b
Specimen D
1 2 3 C-a-a

(a)

D-a-a

1 2 3

(b)

Fig. 16 SEM micrographs C-b-b

of section b–b of a 1 2 3
Specimen C and b (a)
Specimen D

D-b-b

1 2 3

(b)

123
192 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 17 a Optical, b SEM

and c composite overlaid
image of central portion of
section b–b of Specimen D
distinguishing cavities from
grain pull-out

grains must have been pulled out during polishing. We was removed from the surface to ensure that the new
do not observe such large pulled out grains in Spec- section is far enough away from the previous section
imens A, B and even C. Could such pull-out be the to avoid the artifacts associated with etching process.
result of damage at the grain boundaries during defor- It was subsequently fine polished as mentioned before
mation?2 In order to explore this further, section b–b of but not etched. It was verified by numerical simulation
Specimen D was polished until a layer of about 300 µm and DIC measurements that the variation of strain and
2
triaxiality between these two sections did not exceed
This raises the issue of accurate metallographic quantification
3 and 0.8%, respectively. The resulting microstructure
of porosity: since even under precise conditions the presence of
artifacts resulting from metallographic preparation is unavoid- is shown in Fig. 18c where large cavities correspond-
able, new non-destructive techniques such as X-ray tomography ing to grain pullout during metallographic preparation
are being developed as potential candidates. Attempts to obtain are not observed and the only visible cavities are those
the reconstructed 3D X-ray tomography images of the voids in
which have nucleated at second phase particles during
interrupted specimens, however, did not succeed due to insuffi-
cient spatial resolution of images. the deformation process.

123
Ductile failure behavior of polycrystalline Al 6061-T6 193

Fig. 18 Higher magnification images indicating differences b–b of Specimen D showing only the cavities nucleated at second
between possible grain pull-out (a) and cavity growth (b). (c) phase particles during the deformation
Micrograph of the unetched section 300 µm away from Section

In order to develop quantitative measures of dam- the Table 4 that the area fraction of cavities is larger in
age from the micrographs in Figs. 15 and 16, a grid of Specimen D than in Specimen C at the two sections a–a
square cells of size 150 µm is overlaid on the images to and b–b. This is of particular importance in that both
provide a symbolic representative area element of the Specimens C and D were unloaded at about the same
material. Discounting cavities that have been identified point in the nominal stress versus normalized cross-
to be the result of pullout of grains during polishing, head displacement response; yet the different damage
the area fractions of the remaining cavities are calcu- levels experienced by the two specimens can be attrib-
lated for three cells at the center of the section marked uted to the stochastic nature of the void nucleation,
in each of Figs. 15 and 16; these measurements of the which, in turn, stems from the variation in the micro-
void areas and spacing are given in Table 4. structure of the material. Furthermore, the local strains
The size of cavities is in the range of 1–5 µm, sim- in Specimen C exhibiting smaller amount of voids is
ilar to the size range of the second phase particles; larger than in Specimen D. The absence of significant
the spacing between these cavities is on the order of damage in Specimens A and B, and the appearance of a
35–45 µm. Comparing this to the average spacing of small number of voids in Specimens C and D that were
the second phase particles (25 µm), it is clear that not all loaded to the brink of failure indicates that nucleation
second phase particles are trigger points for voids; this of damage from the second phase particles in the form
is also identifiable from the micrographs. It is seen from of debonding and/or particle fracture occurs rather late

123
194 A. Ghahremaninezhad, K. Ravi-Chandar

Table 4 Microstructure
Mean spacing (µm) Average void area fraction
analysis of sections a–a and
b–b of Specimens C and D Cell # 1 2 3 1 2 3

C-a–a 34 39 36 0.00285 0.00467 0.00226

D-a–a 41 38 72 0.00564 0.00783 0.00204
C-b–b 36 29 37 0.00447 0.00117 0.00596
D-b–b 33 41 45 0.01785 0.0107 0.00734

in the overall straining of the material. Correlating the (only half the image was analyzed), we identify the
local grain based strain in the Specimens C and D, as centers and equivalent diameters of each cavity in the
shown in Fig. 13, to the emergence of the voids, we can fracture plane. The distribution of areas of the cavities
suggest that nucleation of damage in Al 6061-T6 occurs is shown in Fig. 19c; the histogram appears to indicate
at equivalent strain levels larger than one. Of course, three groups or clusters of void areas. First, there is a
one must bear in mind that the nucleation strain level large number of very small voids (diameter <10 µm);
will be a strong function of the properties of the matrix, these appear typically at the junctions between larger
the second phase particles, the interface characteristics cavities and are most likely the secondary voids
and stress state. Lee and Mear (1999) made an estimate that arise during coalescence between primary voids.
of the stress concentration in ellipsoidal inclusions in a Second, there seems to be a cluster of voids of size
power law hardening material. Their results indicated ∼25 µm, and another cluster ∼18 µm; these appear to
that the stress concentration at the inclusion boundaries correlate to the mean size of the grains in the y−z plane.
saturated after rising quickly with strain. This is often From these measurements, we estimate that the primary
taken as indication that cavity nucleation must occur voids are nearly of the shape of the grains, and cause
either at very low strain levels or not at all. The pres- separation of the grains; some of the voids that are of the
ent results suggest that cavity nucleation could occur average size of the grain in the y −z plane are identified
at high strain levels. in Fig. 19a by the red lines. Smaller, secondary voids
Fractographs of Specimen E at the intersection of appear in the region of coalescence of the primary voids
sections a–a and b–b with the fracture plane are shown or at grain boundaries; a few of these smaller cavities
in Fig. 19a, b. These images exhibit numerous dim- are highlighted by the blue lines in Fig. 19a. But the
ples that are characteristics of the development of most important observation that deserves reiteration is
ductile fracture through the nucleation, growth and that the entire fracture surface is tiled with the dimpled
coalescence of cavities. Almost all of the larger dim- features and, therefore, every point in the y − z plane
ples contain a second phase particle inside its boundary, belongs to a void.
indicating that these particles are, indeed, the nucle- We now contrast this with an examination of the
ation sites. The void surfaces themselves appear ‘ter- plane orthogonal to the fracture plane. SEM micro-
raced’ or stepped, suggesting crystallographic slip as graphs of the Section b–b of Specimen E’ (another bro-
the deformation process during void growth; there is no ken specimen) is shown in Fig. 20; these micrographs
evidence of any other mechanism of failure. The overall allow an examination of damage in the planes that are
fracture plane is tilted at an angle about 56◦ to the load- below the fracture plane. A depth of about 500 µm
ing direction, which is in line with the inclination of the below the fracture surface at two different locations
shear bands that developed preceding the final failure is shown in this figure; a higher magnification view of
(see Fig. 12). While dimpled images of fracture surface the region just below the fracture surface is also shown
are typically shown to provide qualitative support for inset in Fig. 20c. A small number of second phase par-
the mechanism of failure, it is possible to obtain signif- ticles can be seen to have been broken and/or debonded
icant quantitative measures of fracture. We begin with a from the matrix; the ‘voids’ resulting from such dam-
consideration of the statistics of cavities: by identifying age appear to have elongated slightly in the direction
the boundaries of each cavity, as indicated in Fig. 19a of tension. However, the damaged particles appear in

123
Ductile failure behavior of polycrystalline Al 6061-T6 195

Fig. 19 SEM fractographs

at the intersection of a (a) (b)
section b–b and b section
a–a with fracture surface in
Specimen E. c Distribution
of dimple area over half the
fractograph shown in (a)

(c) 0.08

0.06
Frequency

0.04

0.02

0
100 200 300 400 500 600

Dimple Area - µm2

clusters, leaving large regions of the material below the fracture plane. It is easy to recognize that this scale
the fracture plane completely undamaged, exhibiting must be set by the grain size by the following argument:
no systematic development of voids. Grain boundaries nucleation of voids is seldom a homogeneous process,
are clearly visible in the micrograph and these appear but governed by the heterogeneities, especially by the
to be undamaged as well. In contrast to the appearance extreme fluctuations. Therefore, we can postulate that
of the fracture surface that was tiled with voids, we the first failure of a single grain by void growth would
can see that almost no point below the fracture plane at be triggered by the weakest heterogeneity in the region
locations that are as close as 30 μm experiences nucle- of the largest stress; when failure of this grain occurs,
ation or growth of voids and therefore in contrast to the the load is shed to the neighboring grains, triggering
fracture plane, almost no point in the x − z plane is part the nucleation and growth of the next void in these
of a void/nucleation or growth process. grains, even if the heterogeneities in this grain are not
Comparison of the number and distribution of voids as weak. The cascading of such a process results in a
on the fracture plane and the planes below the fracture localization of the failure process to the “failure plane”.
plane indicates that, indeed, ductile fracture in Al 6061- Outside of this plane, plastic deformation occurs to very
T6 occurs by void nucleation, growth and coalescence; large equivalent plastic strain levels without significant
however, it occurs in an extremely localized manner, damage (see Fig. 13). Such localization of damage was
with almost no statistically significant void nucleation identified in the percolation model of Worswick et al.
or growth at locations as close as 30 µm away from (2001).

123
196 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 20 a, b The SEM

micrographs of the section
b–b of Specimen E. c
Higher magnification of red
box in (a)

The optical and scanning electron micrographs of – Once damage in the form of breakage or debonding
specimens that failed as well as those that were inter- of the second phase particles begins, subsequent
rupted prior to failure provide a picture of the evolution failure occurs with little further increase in macro-
of deformation, damage and failure in Al 6061-T6; we scopic strain. Specimens that were unloaded, and
summarize our observations here. specimens that were fractured, reveal that there is
not a significant number or distribution of voids
– Nucleation of damage does not appear to occur left in the wake of the fracture at distances greater
until equivalent plastic strains of at least about one; than about 30 µm from the fracture plane.
much of the post-peak response of the specimen – The fracture plane is tiled with voids indicating
can be attributed to the formation of necking and that void nucleation, growth and coalescence is
shear bands in the specimen through plastic defor- the essential mechanism of crack growth; however,
mation and not to development of material damage. dominated by the fluctuations in the distribution of

123
Ductile failure behavior of polycrystalline Al 6061-T6 197

400
second phase particles, failure by void nucleation, (a)
growth and coalescence occurs in a highly local- 350

ized region on the order of the grain width.

Nominal Stress - MPa

300

250

200
5 Numerical simulation
150
Experiment
As mentioned earlier, the variation of stress through- 100 Simulation (isotropic)
out the notched specimens in experiments can only be Simulation (anisotropic)
50
evaluated by means of numerical simulations. A com-
mercial finite element software ABAQUSTM was used 0
0.00 0.05 0.10 0.15 0.20
to perform these calculations. Finite element models Normalized Gage Displacement - Δ/L
are created for each test using two different material
450
models—an isotropic von Mises plasticity model and (b)
400
the Hill quadratic anisotropic plasticity model, both
350
available in the ABAQUS standard material library.

True Stress - MPa

The Lankford r -values identified from DIC measure- 300

ments and shown in Table 3 were used to represent 250

the material anisotropy; these parameters were used 200

in the Hill’s anisotropic plasticity material model in 150
ABAQUS/Standard. Comparisons of different material 100
models with experimental responses will be presented 50
for each specimen. 0
0 0.2 0.4 0.6 0.8 1
True Plastic Strain
5.1 Uniaxial tension test
Fig. 21 a Nominal stress versus normalized gage displacement
response from experiment and numerical simulations using iso-
The uniform region of the dog-bone specimen was tropic von Mises and Hill’s anisotropic quadratic yield functions.
modeled in ABAQUS with 3D linear continuum ele- b True stress-true plastic strain curve from experiment (solid line)
ments with reduced integration (C3D8R). A very fine and its extension by fitting to the nominal stress versus normal-
ized gage displacement response from experiment (dashed line)
mesh was adopted in the middle part where the necking
occurs, and gradually coarsened towards the ends. The
minimum mesh size was 0.20 mm; 9 elements were
used across the 2.44 mm thickness of the specimen. response from numerical simulation is compared with
Nodal displacements in the axial direction were pre- the experimental response, and iterating by decreasing
scribed at the top and bottom boundaries of the speci- the modulus until acceptable agreement of the load is
men; the transverse tractions were set to zero in order reached. This trial-and-error process is continued until
to mimic the uniaxial tension test. The elastic-plastic the entire nominal stress versus normalized gage dis-
constitutive model of the material used in the finite ele- placement response of the specimen is recovered in the
ment simulations are extracted from the uniaxial ten- simulation. Special care should be taken in this process
sile test up to the onset of necking. In order to trigger as the elastic-plastic response has a great influence in
the localization, 3 elements at the center surfaces of determining the deformation modes in different types
the model parallel to the x − z plane were given a yield of loadings. The simulated nominal stress versus nor-
strength of about 0.1 of the rest of the model to serve as malized gage displacement curve is compared to the
imperfection site. The post-necking material behavior experimental measurement in Fig. 21a; the true stress-
was obtained by an inverse method; in this method the true plastic strain curve obtained by the iterative process
hardening behavior of the material is extrapolated lin- is shown in Fig. 21b. It is seen that Hill’s anisotropic
early according to the current tangent modulus and the model captures the experimental response of the uni-
nominal stress versus normalized gage displacement axial tension test reasonably well.

123
198 A. Ghahremaninezhad, K. Ravi-Chandar

Fig. 22 Discretized model

of the flat notched
specimen; due to symmetry
only half the geometry was
considered

450
5.2.1 Plastic response
400

350 The nominal stress versus normalized gage displace-

Nominal Stress - MPa

E
A
300 B ment curve from simulations using isotropic plasticity
C D
250 (von Mises yield function) and anisotropic plasticity
200 (Hill quadratic yield function) are compared to that
150
from the experiment in Fig. 23. As can be seen from
Experiment
Fig. 23, the Hill anisotropic plasticity model captures
100
Simulation (isotropic J2 plasticity) very well the material response during the deformation
50
Simulation (anisotropic plasticity) history while the isotropic model prediction is not satis-
0
0 0.01 0.02 0.03 0.04 0.05 0.06
factory. Note that no failure model or criterion has been
Normalized Gage Displacement - Δ /L used in the simulations; thus the response is entirely
due to plastic deformation. This also means that the
Fig. 23 Comparisons of nominal stress (force/initial minimum termination point (final failure point) is not obtained in
cross-sectional area) versus gage displacement (Δ) normalized
the simulations. The nominal stress versus normalized
by the gage length (L) (see Fig. 4b for definition of the gage
length) responses from experiment and simulation using isotro- gage displacement response gives the global response
pic von Mises and Hill’s anisotropic quadratic yield functions of the material where the effects of localization are
smeared out; therefore, we consider comparison of
local field measurements to assess the capability of the
constitutive model. To this end, the variation of surface
axial strain εx along the lines y1 − y2 and x1 − x2 (see
5.2 Notched tension test Fig. 4b) from simulations and experiments for three
different normalized gage displacements (correspond-
The response of the flat, notched tension specimen is ing to Specimens A, B and C) are plotted in Fig. 24. It
considered next. Due to symmetry half the specimen is seen that the numerical simulation with Hill aniso-
was modeled with 3D continuum elements (C3D8R) tropic yield function captures the essential features of
with 31 elements through the thickness; a finer mesh the deformation observed in the experiment. From the
size of 50 µm was used in the minimum cross section Fig. 24a it is observed that the strain prediction is in
(Fig. 22). First, we discuss the plastic response of the fairly good agreement with the experiment for much
specimen; this is then followed by some observations of the line y1 − y2 . The peak near the notch tip that
regarding failure modeling. occurs at y = 7.14 mm due to the necking at this loca-

123
Ductile failure behavior of polycrystalline Al 6061-T6 199

(a) 0.45 anisotropy have been developed and implemented in

Experiment
0.4
C Simulation (anisotropic) finite element software (Barlat et al. 2003; Korkolis and
0.35 Kyriakides 2008a,b); however these complex models
0.3
B
require calibration of several parameters and has not
0.25 been attempted in this work. However, as mentioned
Strain

0.2 A earlier the main focus of this work is on the events

0.15
leading to the final failure in ductile materials rather
than building a plasticity constitutive relation. The Hill
0.1
anisotropic plasticity model adopted here seems to cap-
0.05
ture the essential characteristics such as onset of neck-
0
-10 -5 0 5 10 ing as well as the development of shear localization
y - mm through the thickness. Therefore, it is concluded from
(b) 0.3 comparison to the experimental results that the Hill
Experiment
Simulation (isotropic)
C
anisotropic plasticity model can be used as a reliable
Simulation (anisotropic)
0.25 B
tool to predict the evolution of stress state especially
stress triaxiality during the deformation.
0.2
Strain

0.15 5.2.2 Strain-to-failure

A
0.1 The numerical simulation allows us to correlate the
strain at the grain level with the stress triaxiality at the
0.05
critical points in the specimen. Figure 25a shows a sec-
0 tioned view from the simulation indicating contours of
-3 -2 -1 0 1 2 3 triaxiality; the variation of the stress triaxiality at differ-
x - mm
ent gage displacements across the dashed line marked
Fig. 24 Comparisons of surface axial strain εx along a line y1 − in Fig. 25a is shown in Fig. 25b. Initially, the maximum
y2 and b line x1 − x2 from experiment and numerical simulations triaxiality occurs in the center of the specimen (∼plane
strain, 0.50) and decreases gradually towards the notch
which is nearly in a uniaxial stress state. However, this
trend changes with the deformation; it is seen that at
tion is also realized; however the simulation results are the brink of failure (Specimen C) the maximum triaxi-
slightly higher with the difference increasing with the ality of about 0.82 occurs near the notch (y = 7.14 mm)
overall deformation. In Fig. 24b, the double peaks that while it is about 0.68 at the center. The change in the
correspond to the formation of shear bands across the location of maximum stress triaxiality from the mid-
thickness of the specimen are also captured with the point of the specimen to the notch tip region during
Hill plasticity model. After the localization of defor- the deformation should particularly be emphasized as
mation in the form of intense shear bands, the central it potentially determines the location of damage and
portion of the specimen outside these localized bands macro crack initiation that finally leads to final fail-
experience a smaller increase in strain. ure of the specimen; experimental observations indeed
The quantitative discrepancy between the experi- confirmed nucleation of failure in the notch tip region.
ment and the numerical simulations can be attributed This was also the main reason for selecting the sec-
to the following: first, adopting an appropriate con- tion b–b for metallographic examination and measure-
stitutive material model that takes into account dif- ments.
ferent characteristics of the material microstructure Phenomenological fracture models such as John-
such as grain morphology and crystallographic texture son–Cook and modified Mohr–Coulomb criteria have
during the multiaxial deformation is of great impor- been developed and used extensively in the numeri-
tance in order to achieve precise numerical predic- cal simulation of fracture in ductile material. In these
tions. For aluminum alloys, plasticity models based models the model parameters are usually calibrated
on a non-quadratic yield function that accounts for by measuring the strain in the gage section of a

123
200 A. Ghahremaninezhad, K. Ravi-Chandar

1.2

1.0

Strain to Failure
0.8

0.6

0.4
(a)
0.2

0.0
0 0.2 0.4 0.6 0.8 1 1.2
Triaxiality

Fig. 26 Red band representing a lower bound fracture strain in

0.9
(b) comparison with the Johnson–Cook model (Lesuer et al. 2001)
0.8 C with parameters D1 = −0.77, D2 = 1.45, D3 = −0.47 and
0.7
B D4 = 0

0.6
without any observable damage in the material; for
Triaxiality

A
0.5
comparison, the Johnson–Cook model calibrated for
0.4
Al 6061-T6 by Lesuer et al. (2001) is also shown. As
0.3
readily seen from Fig. 26 the strain-to-failure obtained
0.2
at the level of grains that are building blocks of the
0.1 material lies distinctly above the fracture locus of John-
0 son–Cook; the arrow on the red band indicates that the
0 2 4 6 8 10
y -mm true fracture strain could be even larger as we report
the strain in the specimens just before the final failure;
Fig. 25 a Contour plot of the stress triaxiality in y − z plane at therefore, we consider this to be the lower bound of
the minimum cross section (x = 0) corresponding to Specimen the strain-to-failure. The difference from the Johnson–
C. b Variation of the stress triaxiality at different stages of defor-
mation (Specimens A, B and C) along the minimum cross section Cook type models arises from the fact that we have
in the mid-plane in the thickness direction (0 < y < y2 , z = 0) used an intrinsic length scale based on the microstruc-
ture—the grain size—as the basis for the definition of
an invariant strain measure.
broken specimen or using a hybrid experimental-
numerical method; in the latter the force-gage displace-
ment response of the specimen is used as the measure 6 Summary
of fitness and the calculated strain at the critical point
corresponding to global failure is taken as the intrinsic Tension tests and interrupted flat notched tension tests
material strain-to-failure for a given level of triaxial- were performed on Al 6061-T6 in order to examine the
ity. In this section we give a conservative lower bound evolution of the microstructure at different stages of
estimate of the strain-to-failure for Al 6061-T6 that deformation. The macroscopic strain fields were mon-
is rooted in the grain based measurement of strain. It itored with a variation of the digital image correlation
was shown in Sect. 3 that strains in the range of 0.5– method called Q4-DIC which was adopted and imple-
1.0 could be measured in the specimens just before the mented in MATLAB.
final failure, but without any indication of damage. The Samples from interrupted tests were sectioned, pol-
stress triaxiality in the region b–b at this stage of defor- ished and etched following standard procedures for
mation is obtained from the finite element analysis to be metallographic examination. The initial microstructure
about 0.82. We now show this as a red band on a plot of the Al 6061-T6 was examined and the grain size was
of strain-to-failure versus triaxiality (see Fig. 26) to determined; it is evident from distributions of the grain
represent that strains of this magnitude were observed size that the grains are equiaxed in the rolling plane

123
Ductile failure behavior of polycrystalline Al 6061-T6 201

(x − y plane) with a mean size of 39 µm. The average N00014-01-1-A00001 and FNC project: N00014-08-1-0189.
grain size in the z-direction is about 14 µm. This grain This support is gratefully acknowledged.
size distribution suggests planar isotropy in the micro-
structure. The microstructure of this material shows a
dispersion of mostly iron-based second phase particles References
with volume fraction in the range of 0.012 and mean
particle spacing in the range of 25 µm. Babout L, Brechet Y, Maire E, Fougères R (2004) On the com-
petition between particle fracture and particle decohesion
Evolution of deformation in the microstructure was in metal matrix composites. Acta Mater 52:4517–4525
obtained by direct measurements of grain size using Barlat F, Brem JC, Yoon JW, Chung K, Dick RE, Lege DJ, Pour-
optical microscopy. Correlation of strain estimates at boghrat F, Choi S-H, Chu E (2003) Plane stress function for
the grain level with the process of damage evolution aluminum alloy sheets—part I: theory. Int J Plast 19:1297–
1319
using scanning electron microscopy was made at dif- Benseddiq N, Imad A (2008) A ductile fracture analysis using a
ferent levels of macroscopic deformation. This allows local damage model. Int J Pressure Vessels Piping 85:219–
characterization of the damage in a multi-scale frame 227
work. Benzerga AA, Leblond J-B (2010) Ductile fracture by void
growth to coalescence. Adv Appl Mech 44:169–305
It was concluded that in Al 6061-T6 under ten- Bernauer G, Brocks W (2002) Micro-mechanical modelling of
sile loading, nucleation of damage does not appear for ductile damage and tearing—results of a European numeri-
much of the deformation history until plastic strain lev- cal round robin. Fatigue Fract Eng Mater Struct 25:363–384
els of at least about one is reached. Once damage in Besnard G, Hild F, Roux S (2006) ‘Finite-element’ displacement
fields analysis from digital images: application to Portevin–
the form of particle fracture or decohesion at the inter- le châtelier bands. Exp Mech 46:789–803
face initiates, subsequent failure follows with a small Bridgman PW (1964) Studies in large plastic flow and fracture.
increase in the overall strain. The final separation, dom- Harvard University Press, Cambridge
inated by the fluctuation in the distribution of second Brocks W, Sun DZ, Honig A (1996) Verification of microme-
chanical models for ductile fracture by cell model calcula-
phase particles, occurs by the void nucleation, growth tions. Comput Mater Sci 7:235–241
and coalescence in a highly localized layer of material Chu CC, Needleman A (1980) Void nucleation effects in biaxi-
on the order of the grain size. ally stretched sheets. J Eng Mater Technol 102:249–256
Numerical simulation of the experiments using von Decamp K, Bauvineau L, Besson J, Pineau A (1997) Size and
geometry effects on ductile rupture of notched bars in a C–
Mises isotropic and Hill’s quadratic anisotropic plas- Mn steel. Experiments and modelling. Int J Fract 88:1–18
ticity models were performed in the finite element soft- Dong MJ, Berdin C, Beranger AS, Prioul C (1996) Damage effect
ware ABAQUS 6.9. Lankford r -values were extracted in the fracture toughness of nodular cast iron. In: First Euro-
by performing uniaxial tension tests in different direc- pean mechanics of materials conference on local approach
to fracture, Fontainebleau, September
tions. The hardening behavior was obtained through Ghahremaninezhad A, Ravi-Chandar K (2011) Ductile failure of
a hybrid experimental-numerical procedure. It was polycrystalline OFHC copper. Int J Solids Struct 48:3299–
shown that anisotropy plays a significant role in cap- 3311
turing the global as well as local features of the experi- Gurson A (1977) Continuum theory of ductile rupture by void
nucleation and growth: part I—yield criteria and flow rules
ment such as diffuse necking and shear banding; there- for porous ductile media. J Eng Mater Technol 99:2–15
fore, Hill quadratic anisotropic plasticity model was Hancock JW, Mackenzie AC (1976) On the mechanisms of duc-
used to obtain the variation of stress state during the tile failure in high-strength steels subjected to multi-axial
deformation. Values of stress triaxialities and the corre- stress-states. J Mech Phys Solids 24:147–169
Hutchinson JW, Tvergaard V (1980) Surface instabilities on stat-
sponding deformation at the grain level were compared ically strained plastic solids. Int J Mech Sci 22:339–354
with strain-to-failure ductile failure models such as the Johnson GR, Cook WH (1985) Fracture characteristics of three
Johnson–Cook model. It was shown that the lower- metals subject to various strains, strain rates, temperatures
bound strain-to-failure estimates measured at the grain and pressures. Eng Fract Mech 21:31–48
Korkolis YP, Kyriakides S (2008a) Inflation and burst of aniso-
level are more than double the values usually calibrated tropic aluminum tubes for hydroforming applications. Int J
into this model. Plast 24:509–543
Korkolis YP, Kyriakides S (2008b) Inflation and burst of aniso-
Acknowledgments This work was performed during the course tropic aluminum tubes, part II: an advanced yield func-
of an investigation into ductile failure under two related research tion including deformation-induced anisotropy. Int J Plast
programs funded by the Office of Naval Research: MURI project 24:1625–1637

123
202 A. Ghahremaninezhad, K. Ravi-Chandar

Lassance D, Fabregue D, Delannay F, Pardoen T (2007) Rice J, Tracy D (1969) On the ductile enlargement of voids in
Micromechanics of room and high temperature fracture in triaxial stress fields. J Mech Phys Solids 17:201–217
6xxx Al alloys. Prog Mater Sci 52:62–129 Simar A, Nielsen KL, de Meester B, Tvergaard V, Pardoen T
Lee BJ, Mear ME (1999) Stress concentration induced by an (2010) Micro-mechanical modelling of ductile failure in
elastic spheroidal particle in a plastically deforming solid. 6005A aluminium using a physics based strain hardening
J Mech Phys Solids 47:1301–1336 law including stage IV. Eng Fract Mech 77:2491–2503
Lesuer DR, Kay GJ, LeBlanc MM (2001) Modeling large strain, Steglich D, Brocks W (1998) Micromechanical modelling of
high-rate deformation in metals. Modeling the performance damage and fracture of ductile materials. Fatigue Fract Eng
of engineering structural materials II. In: Proceedings of a Mater Struct 21:1175–1188
symposium sponsored by the SMD of TMS, Indianapolis, Sutton MA, Wolters WJ, Peters WH, Ranson WF, McNeill SR
IN, pp 75–86 (1983) Determination of displacements using an improved
McClintock FA (1968) A Criterion for ductile fracture by the digital correlation method. Image Vis Comput 1:133–139
growth of holes. J Appl Mech 35:363–371 Tipper CF (1949) The fracture of metals. Metallurgia 39:133–
Nielsen KL (2008) Ductile damage development in friction 137
stir welded aluminum (AA2024) joints. Eng Fract Mech Tvergaard V (1990) Material failure by void growth to coales-
75:2795–2811 cence. Adv Appl Mech 27:83–151
Norris DMJr, Moran B, Scudder JK, Quinones DF (1978) A Weck E, Leistner E (1996) Metallographische Anleitung zum
computer simulation of the tension test. J Mech Phys Solids Farbätzen nach dem Tauchverfahren Teil III: Nichteisen-
26:1–19 metalle, Hartmetalle und Eisenwerkstoffe, Nickel-Basis-
Pardoen T, Doghri I, Delannay F (1998) Experimental and und Kobalt-Basis-Legierungen, Series of Specialist Books
numerical comparison of void growth models and void coa- on Welding Technology Band 77/III
lescence criteria for the prediction of ductile fracture in cop- Worswick MJ, Chen ZT, Pilkey AK, Llyod D, Court S (2001)
per bars. Acta Mater 46:541–552 Damage characterization and damage percolation modelin-
Puttick KE (1959) Ductile fracture in metals. Philos Mag 4:954– gin aluminum alloy sheet. Acta Mater 49:2791–2803
969 Zhang ZL (1996) A sensitivity analysis of material parameters
Réthoré J, Hild F, Roux S (2007) Extended digital image corre- for the Gurson constitutive model. Fatigue Fract Eng Mater
lation with crack shape optimization. Int J Numer Methods Struct 19:561–570
Eng 73:248–272

123