Fatigue & Fracture of

Engineering Materials & Structures

doi: 10.1111/j.1460-2695.2007.01207.x

Fatigue damage analysis in a duplex stainless steel by digital image

correlation technique
A. EL BARTALI, V. AUBIN and S. DEGALLAIX
Laboratoire de Mécanique de Lille (LML, UMR CNRS 8107) Ecole Centrale de Lille, BP 48, 59651 Villeneuve d’Ascq Cedex, France

Received in final form 11 October 2007

A B S T R A C T Strain field measurements by digital image correlation today offer new possibilities for
analysing the mechanical behaviour of materials in situ during mechanical tests. The
originality of the present study is to use this technique on the micro-structural scale,
in order to understand and to obtain quantitative values of the fatigue surface damage
in a two-phased alloy. In this paper, low-cycle fatigue damage micromechanisms in an
austenitic-ferritic stainless steel are studied. Surface damage is observed in real time,
with an in situ microscopic device, during a low-cycle fatigue test performed at room
temperature. Surface displacement and strain fields are calculated using digital image
correlation from images taken during cycling. A detailed analysis of optical images and
strain fields measured enables us to follow precisely the evolution of surface strain fields
and the damage micromechanisms. Firstly, strain heterogeneities are observed in austenitic
grains. Initially, the austenitic phase accommodates the cyclic plastic strain and is then
followed by the ferritic phase. Microcrack initiation takes place at the ferrite/ferrite grain
boundaries. Microcracks propagate to the neighbouring austenitic grains following the
slip markings. Displacement and strain gradients indicate probable microcrack initiation
sites.
Keywords damage; digital image correlation; duplex stainless steel; fatigue; strain het-
erogeneity.

structure). This two-phased structure, combined with a

INTRODUCTION
small grain size (≈10 μm), leads on the one hand to good
Today, in structural design, increased requirements in mechanical properties, (high yield stress, ultimate tensile
terms of safety, cost and weight reduction involve using strength, hardening and ductility) which are further in-
reduced safety coefficients, improving mechanical prop- creased by nitrogen alloying, and on the other hand to a
erties of materials and establishing relevant constitutive very good corrosion resistance. Nowadays, DSS are used
laws. This involves knowing and understanding precisely in wide fields of industry, especially in the oil and gas,
the behaviour of materials in service on different scales. petrochemical, paper and nuclear industries.1 Moreover,
When the structure is loaded under multiaxial cyclic me- DSS mechanical properties, perhaps still more than that
chanical loading, it is necessary to identify strain and dam- of other steels, depend on the microstructure that itself re-
age physical mechanisms on the microstructural scale in sults from chemical composition, elaboration process and
order to define a relevant macroscopic constitutive law heat treatment. This microstructure is characterized by
available in structured calculation software. the two-phased morphology, the phase volume fractions,
Austenitic-ferritic stainless steels, also called duplex the grain size, the grain orientation distribution, etc. In
stainless steels (DSS), are two-phased alloys whose chem- spite of the very good mechanical properties of DSS at
ical composition and heat treatment are optimized to room temperature, there is a detrimental effect of ageing
obtain quasi-balanced volume fractions of austenite (γ - at a temperature between 250 and 500 ◦ C, due to an em-
phase, face cubic-centred crystallographic structure) and brittlement phenomenon by spinodal decomposition of
ferrite (α-phase, body-centred cubic crystallographic the ferrite.2
In the past years, low-cycle fatigue behaviour and
Correspondence: Ahmed El Bartali. E-mail: [email protected] damage mechanisms were analysed in the literature in


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151 137
138 A . E L B A RTA L I et al.

various types of DSS with different α to γ volume frac- or block microcrack propagation compared with grain
tions, nitrogen contents, strain amplitudes, ageing treat- boundaries.
ments (solution treated or embrittled by thermal ageing) According to the above brief review, previous DSS low-
at different temperatures.3–17 One of the very first stud- cycle fatigue studies analysed and explained the mechan-
ies on DSS low-cycle fatigue was carried out by Magnin ical behaviour of DSS in low-cycle fatigue from obser-
and co-workers.3 They studied a DSS in low-cycle fa- vations at given moments of the fatigue life, and no
tigue at room temperature at imposed plastic strain am- observation coupled with strain measurements at the mi-
plitudes (10−4 < ε p /2 < 10−2 ) and showed that the fa- crostructural scale was performed throughout the lifetime.
tigue life of DSS gets closer to the austenite fatigue life at In order to go deeper into the understanding of the cyclic
low strain levels (ε p /2 < 10−3 ) while it gets closer to the plastic strain phenomena in DSS, it is necessary to fol-
ferrite one at higher strain level (ε p /2 > 10−3 ). These low the surface evolution during the cycling. The recent
first results were extended during the last two decades development of optical techniques now allows the char-
to the analysis of damage evolution and crack initiation acterisation of non-homogenous mechanical behaviour of
in low-cycle fatigue. Degallaix et al.5 studied the nitro- materials and structures, for example, by the measurement
gen influence on fatigue damage in two DSS alloyed with of non-homogenous surface strain fields. The latest wide
0.11 and 0.18% nitrogen (N) by weight respectively, in development of these techniques can be seen through the
low-cycle fatigue at room temperature, in the strain range recent conferences dedicated to the application of opti-
interval 8 × 10−3 < ε t < 2.5 × 10−2 . Nitrogen alloy- cal techniques for experimental mechanical characterisa-
ing stabilizes and strengthens the austenite while ferrite tion of materials.19 Contrary to common strain measure-
properties remain approximately unchanged. The stress ment methods—using, for instance, strain gauges which
response to cyclic strain increases with the nitrogen con- allow measurement of the mean strain on a given local
tent. The fatigue life increases with the nitrogen content area—mechanical field measurement techniques by digi-
at high strain levels while it remains almost unchanged tal image correlation (DIC) give a whole map of displace-
at low strain levels. Most of the time, microcracks initi- ments or strains or both. The robustness, the speed and
ate in the ferrite due to the more intense cyclic plastic the accuracy of these techniques have increased greatly
strain localisation in it, as observed by Scanning Electron during the past 5 years, which has enabled local informa-
Microscopy (SEM). This was confirmed by Transmission tion to be obtained on mechanical behaviour of materials
Electron Microscopy (TEM) observations performed on not only on the macroscopic scale but also on the mi-
the steel alloyed with 0.18% N.4,5 The cyclic plastic strain croscopic scale, e.g. on micromechanisms such as strain
is always more homogeneously distributed in the austen- localisation, damage localisation, crack initiation and
ite, due to the fact that nitrogen favours planar disloca- propagation.19–22
tion slip. At low plastic strain ranges, plastic deformation The study presented here consists of performing obser-
is essentially accommodated by the austenitic phase. At vations at different times during a low-cycle fatigue test
higher plastic strain ranges, both phases contribute to without taking apart the specimen, and images are then
plastic deformation. Dislocation structure evolution has used to measure surface strain fields. Images were taken
been widely studied by Mateo et al.,6 Llanes et al.,7 Kruml in real time during cycling on the surface of a standard
et al.,8 Alvarez-Armas et al.15 and Marinelli et al.16 The dis- low-cycle fatigue test specimen. Displacement and strain
location structure evolution depends on the plastic strain fields were then calculated from these images using a digi-
range level. The fatigue life can be divided in to three tal image correlation technique.23–28 The joint analysis of
stages: (i) cyclic plastic strain localisation and surface dam- both surface observations and surface displacement and
aging, (ii) short-crack initiation and propagation and (iii) strain field measurements enables the cyclic plasticity and
long-crack growth. The first two stages concern generally damage micromechanisms and their evolution to be fol-
more than 80% of the fatigue life of smooth specimens. lowed and understood during cycling, i.e. to follow strain
Cyclic plastic strain localisation favours damage and mi- heterogeneity appearance at grain scale, microcrack initi-
crocrack initiation. When orientation relationships exist ation and micropropagation.
between the two phases of the DSS, the slip glide propa- In this paper, after a presentation of the experimen-
gates easily between neighbouring austenitic and ferritic tal procedure, results of the digital image correlation
grains and no microcrack initiates at the phase bound- software used are evaluated for artificially deformed
ary.15 In other cases, grain, twin and phase boundaries images. Then results of damage micromechanism ob-
play an important role in initiation mechanisms and mi- servations and displacement or strain field measure-
crocrack propagation. Düber et al.17 and Stolarz et al.18 ments for a fully reversed tension-compression low-
reported the microstructural barrier effect on fatigue mi- cycle fatigue test at ε t /2 = 5 × 10−3 are presented and
crocrack propagation in a DSS. Phase boundaries delay discussed.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 139

E X P E R I M E N TA L P R O C E D U R E cycle fatigue test was performed at room temperature, in

a fully reversed push-pull mode at the total strain ampli-
Material
tude ε t /2 = 5 × 10−3 and with a constant total strain
The material studied is an X2 Cr Ni Mo 25-07 nitrogen rate ε̇ = 10−3 s−1 . Control and data acquisition were per-
alloyed austenitic-ferritic stainless steel. It was supplied formed using a program developed using Labview soft-
by Aubert & Duval in a cylindrical bar 70 mm in ware.
diameter. The chemical composition in weight per cent Surface damage of the fatigue specimen and its evolution
is: Cr(24.68); Ni(6.54); Mo(2.84); C(0.024); N(0.17); were observed during the low-cycle fatigue test. Obser-
Mn(0.79); S(0.62); Cu(0.07); P(0.021); S(<0.003); vations were carried out with an in situ microscopy de-
Fe(balance). The microstructure consists of austenitic vice composed of a 10-bit CCD camera, a coaxial lens and
islands (γ -phase, 40% vol.) elongated in the rolling a zoom lens (Fig. 2c). This device was mounted on the
direction, embedded in a ferritic matrix (α-phase, 60% fatigue test machine frame through micrometric tables,
vol.) (Fig. 1a). In Fig. 1a, the ferritic phase is dark and the which permitted the camera to move along three axes and
austenitic phase is light. Both phases are polycrystalline. to rotate around a vertical axis. It allowed a small zone
The mean austenite island length is approximately (120 × 90 μm2 ) of the specimen surface to be observed
500 μm. The austenitic grains are equiaxial with a mean in real time during cycling. The zone observed was small
grain size of 10 μm. The ferritic grains are elongated enough to reveal the fatigue damage of the material on the
in the longitudinal direction with approximately 50 μm microstructural scale and large enough to be representa-
length and 10 μm width. The microstructure is morpho- tive of the two-phased material.
logically transverse isotropic. The material was solution Because of the very small depth of field of the microscopy
treated for an hour at 1050 ◦ C and then water-quenched. device (<1 μm), a shallow cylindrical notch was machined
All alloying elements were, thus, put in solid solution at the centre of the specimen gauge part, in order to very
in both phases. Nitrogen atoms, as powerful austenite slightly concentrate the damage in the observed zone. The
forming interstitial elements, are randomly distributed stress concentration factor at the notch tip should be as
mainly in the austenitic phase and harden the material. small as possible in order not to disturb locally the dam-
This steel was previously studied in the laboratory in age micromechanisms, which must remain representative
terms of low-cycle fatigue and ratchetting macroscopic of those of a representative elementary volume (REV), but
behaviour, cyclic constitutive law and fatigue life at room strong enough to localize the main crack initiation in the
temperature.4,10,11 observed zone at the notch tip. A finite element calcula-
tion was performed in order to optimize the dimensions
of the notch, aiming for a stress concentration factor of
Texture measurements
1.14, as recommended by Polák et al. for the same type of
In order to reveal the possible crystallographic texture of observations.13,14 The dimensions of the notch obtained
austenite and ferrite phases, a sample taken from the rod were 40 mm in radius and 0.4 mm in depth (Fig. 2b). Be-
bar at the location where the specimens were also taken fore testing, the specimen was mechanically polished up to
was analysed by X-ray diffraction on a D500 Bruker AXS 1 μm diamond paste granularity, circumferentially before
goniometer using Cu Kα radiation. Complete pole figures notching and longitudinally after notching, in order to
were obtained for the α-phase using the (110) α , (200) α remove any surface irregularity likely to favour crack ini-
and (211) α diffraction peaks and for the γ -phase using tiation. Finally, it was lightly electrochemically etched for
the (111) γ , (200) γ , (220) γ and (311) γ , diffraction peaks. 50 s in a 50% diluted aqueous solution of nitric acid with
Figure 1b presents four of these pole figures. These mea- a 30 mA/cm2 current density in order to reveal the grain
surements show that there is no texture in the austenitic and phase boundaries of the DSS microstructure. This
phase while there is a weak texture of type 200 parallel electrochemical etching was sufficient to give the speci-
to the rolling direction in the ferritic phase. men surface the random surface texture necessary for a
good digital image correlation.
Testing equipment and in situ microscopy device
PERFORMANCE OF THE DIC TECHNIQUE
The fatigue test specimen was cylindrical and button-
headed, with a gauge section of 10 mm in diameter and 32 Displacement field measurements are performed by dig-
mm in length (Fig. 2a & b). It was mechanically tested on ital image correlation using CorreliQ4 software.23–26
an 8501 INSTRON computer-controlled servohydraulic Briefly speaking, two images of the same specimen sur-
testing machine equipped with a 100 kN load cell. The test face area are compared. The first image is subdivided into
was controlled in longitudinal strain measured by a strain several small squares (l × l pixels), which form a mesh.
gauge extensometer with a 10 mm gauge length. A low- The position of each of these squares is then found in the


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
140 A . E L B A RTA L I et al.

Fig. 1 Microstructure of the austenitic-ferritic stainless steel studied (α-phase in dark, γ -phase in light), and position of the four zones
analysed in the paper (a); pole figures measured using Cu Kα radiation (b).

second image (Fig. 3). A correlation method is used to Performance of the displacement and strain field calcu-
match a square in the initial image to the same square in lations in terms of accuracy and resolution was evaluated
the second image. This method is described in the Ap- on an image similar to those used in this study accord-
pendix and in Refs [23–26]. ing to the procedure described in Refs [21,22,29,30]. The


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 141

Fig. 2 (a) Geometry of the specimen, (b) geometry of the notch at the centre of the specimen and (c) in situ microscopy device mounted on
the testing machine.

Fig. 3 Meshing for the digital image correlation technique, reference image (left), deformed image (right), l is the mesh size, 1 is the loading
direction.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
142 A . E L B A RTA L I et al.

accuracy depends on the software itself, on its parameters

Displacement uncertainty (pixel)

0.14 l = 8 pixels
(for example the correlation window) and on the images 0.12 l = 16 pixels
on which the digital image analysis is performed. These l = 32 pixels
0.1
images depend indeed on the grey level distribution, the l = 64 pixels
0.08
camera dynamic range, the image resolution, the speci- l = 128 pixels
men lighting, etc.19 0.06
To evaluate field calculation performances, images were 0.04
numerically shifted by a known displacement ranging 0.02
from 0 to 1 pixel with 0.1 pixel increments. Then, error
0
and standard deviation on displacements and strains calcu- 0 0.2 0.4 0.6 0.8 1
lated by CorreliQ4 were compared to the prescribed dis-
(a) Prescribed displacement (pixel)
placements and strains. Figure 4a shows the standard de-
viation from displacements (or displacement uncertainty) 0.009
as a function of the prescribed displacements for different 0.008
mesh sizes (l = 8, 16, 32, 64, 128 pixels) for the reference 0.007 l = 8 pixels

Strain uncertainty
0.006 l = 16 pixels
image presented in Fig. 3. The maximum uncertainty is
0.005 l = 32 pixels
reached for a prescribed displacement equal to 0.5 pixel
0.004 l = 64 pixels
and its value is less than 0.12 pixel for a mesh size equal
0.003 l = 128 pixels
to 8 pixels.
The same image was also numerically deformed by an 0.002
artificial uniform strain in order to evaluate the accuracy 0.001
also in terms of strain. Figure 4b presents strain uncer- 0
tainties as a function of the artificial longitudinal strain 0 0.0002 0.0004 0.0006 0.0008 0.001
applied for different mesh sizes (l = 8, 16, 32, 64, 128 pix- (b) Prescribed strain
els). Displacement and strain uncertainties decrease when
the mesh size increases, but the number of measurement
Displacement uncertainty (pixel)

0.1
points decreases.
Size l was chosen to be equal to 32 pixels for the fol-
lowing work. This value was the best compromise (for CorreliQ4
the displacement and strain field calculations) between CorreliLMT
0.01
the lowest displacement and strain uncertainties on one
hand , and enough measurement points to reveal strain
heterogeneities on the grain scale with a good accu-
racy on the other, i.e. strain uncertainties less than 10−3
(Fig. 4b). 0.001
This analysis was also carried out on the same images 1 10 100 1000
with an FFT-based (Fast Fourier Transformation-based) (c) Mesh size (l , pixels)
correlation algorithm, CorreliLMT .31–33 The comparison
of displacement uncertainties obtained with the two soft- Fig. 4 Performance of the CorreliQ4 software evaluated on the
wares is given in Fig. 4c. Displacement uncertainties ob- image in Fig. 3: displacement uncertainty versus prescribed
tained with CorreliQ4 are lower than those obtained with artificial displacement (a); strain uncertainty versus prescribed
strain (b); displacement uncertainty versus mesh size l, comparison
CorreliLMT .
between the softwares CorreliQ4 and CorreliLMT in terms of
displacement uncertainty (c).

M A C R O S C O P I C FAT I G U E B E H AV I O U R
Cyclic hardening/softening behaviour
Let us recall that the strain rate imposed during the test
Figure 5 shows two cyclic hardening-softening curves ob- extensively studied here was 10−3 s−1 . The two hardening-
tained for the duplex stainless steel studied, at a total strain softening curves are similar; this proves that the notch and
amplitude equal to 5 × 10−3 . The plastic strain amplitude the breaks for observations during the test have no influ-
is equal to 2 × 10−3 . The curve with full circles was ob- ence on the macroscopic fatigue behaviour of the material.
tained during the test presented here; it is compared to the The hardening-softening curves present a short period of
second curve obtained with a purely cylindrical specimen hardening (the maximal stress amplitude was reached after
(without a notch) at a total strain rate of 6.6 × 10−4 s−1 .11,12 four cycles) followed by a softening up for stabilisation of


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 143

700
Stress amplitude (MPa)

600
500
400
300
200
with notch
100
without notch
0
1 10 100 1000 10000 100000
Number of cycles

Fig. 5 Cyclic hardening-softening curves of the duplex stainless

steel during fatigue tests performed at ε t /2 = 0.5 10−2 and
R ε = −1, on unnotched and notched specimens.

the stress amplitude. The stabilisation phase covers about

80% of the fatigue life. The accommodation behaviour
(the very first cycles) is consistent with the literature low-
cycle fatigue results on DSS. However, the number of cy-
cles to failure of the specimen tested here is shorter com-
pared to previous results.11 The failure of this specimen
appeared after 1693 cycles, as a result of the propagation
of a macrocrack located outside the extensometer knives,
thus, out of the notch. Fig. 6 Average longitudinal strain measured by digital image
correlation in zone 1 between cycles 0 and 1/4 (a); macroscopic
strain calculated by finite element analysis (b).
Comparison between macroscopic strain ε 11
measured by digital image correlation and that
and equal to 5 × 10−3 in a section far from the notch, and
calculated by finite element method
the specimen was considered symmetrically (Fig. 6b). The
The average strain ε 11 at the notch tip was calculated using strain calculated at the surface at the notch tip was equal
image correlation between one image taken before the test to 6.4 × 10−3 , close to the value calculated using the DIC
(undeformed state) and a second one taken at the same method (6.9 × 10−3 ).
place at the first quarter of the cycle (i.e. under 5 × 10−3
total strain in tension).
O B S E RV E D S U R FAC E DA M AG E
Figure 6a shows the results of DIC calculation carried
M I C R O M E C H A N I S M S A N D C A L C U L AT E D
out in zone 1 (see Fig. 1a). The average strain ε 11 cal-
STRAIN FIELDS
culated in this area is equal to 6.9 × 10−3 (Fig. 6a). The
difference between the applied average strain (5 × 10−3 ) Each image presented here covers an area of 120 ×
and the measured strain (6.9 × 10−3 ) results from the stress 90 μm2 , which represents 1368 × 1024 pixels (so one
concentration due to the notch. pixel corresponds to about 90 nm). A domain of 3 × 3 im-
Calculation of strain and stress fields in the useful part of ages was observed during cycling. The area of this domain,
the notched specimen was carried out by the finite element thus, represents about 300 × 250 μm2 at the notch tip (see
method. The calculation was performed using ABAQUS Fig. 1a). This area was photographed before the test, dur-
software and the following hypotheses: (i) the behaviour ing the test and after the test. During cycling, this domain
of the material was isotropic with a von Mises plastic flow was photographed at the maximal strain at cycles 1/4 , 1, 10,
criterion, (ii) the elastic-plastic behaviour was defined by 50, 100, 300, 500, 651, 1000, 1100, 1200, 1450, 1600, 1650
the monotonic tensile stress-strain curve and (iii) a small and at 1693 (number of cycles to fracture). After fracture,
displacement formulation was used. A displacement was the same zone was observed and photographed by optical
applied to one specimen end so that the strain was uniform microscope and SEM using secondary electrons.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
144 A . E L B A RTA L I et al.

In order to analyse the displacement and strain maps In the ferritic phase, the first slip markings appeared later
at the microstructural scale, the phase and grain bound- than in the austenitic phase. The first slip markings in the
aries are superimposed on the maps. The full lines corre- ferrite were observed only at cycle 50 (Fig. 7b).35 More-
spond to the grain boundaries clearly identified and the over, until fracture, they are much less numerous than
dashed ones correspond to twin or grain boundaries as- in the γ -phase. These markings present a wavy character
sumed. Displacements are given in pixels and strains in typical of an easy dislocation cross slip, as usually observed
full scale. The loading axis is Axis 1 and corresponds to in BCC crystallographic structures.5,6,8
the vertical axis on the figures (Fig. 3). Only ε 11 strain Figure 7e–g permits the analysis of strain fields induced
fields are presented here. by fatigue cycling. We discuss additional strains in order
During cycling, the image texture changes slightly, es- to underline the fact that the strains were measured twice
pecially due to the emergence of slip markings. How- during cycling and that the first time taken as the reference
ever, for a good correlation, it is necessary to keep the was not necessarily the initial time (i.e. before cycling).
same grey level distribution as long as possible and a tex- Moreover, strains measured in the periphery of images
ture as close as possible to the initial one. So, the refer- are not taken into account in the discussion because of
ence image for the displacement and strain calculations boundary effects.
is then updated for each calculation. So, only the dis- Figure 7e shows that significant strain heterogeneities
placement and strain fields calculated between the im- were already induced in γ -grains by cycling between cy-
age taken before cycling (without any deformation) and cles 1/4 and 10. The strain heterogeneities in the γ -phase
at the first quarter of cycle (i.e. under 0.5% total strain then spread and became progressively more pronounced
in monotonous tension) are true displacements and true during the next cycles (up to cycle 50) (Fig. 7f), in partic-
strains, respectively. After this first step, the displacements ular, in the large polycrystalline austenitic island on the
and strains are calculated between two deformed states right of the image, where the first slip markings were pre-
corresponding to two successive breaks and are, thus, ad- viously observed (Fig. 7a & b). After 50 cycles, the ma-
ditional displacements and strains. The actual displace- jority of γ -grains presented strain heterogeneities and,
ments or strains at a given break must, thus, be calculated locally, the strain reached 2% in tension as well as in com-
by adding all displacements or strains until the current pression. The strongly deformed zones in the γ -phase
break. were often located close to grain boundaries. Later, be-
tween cycles 651 and 1000, in some γ -grains, the ad-
ditional strain seemed more homogeneously distributed
(ε 11 values around 0.6%) while it was still heterogeneous
Damage evolution
and around ± 2% in others (Fig. 7g). Let us note that
Figure 7 shows the surface damage evolution of zone 1 (see the additional strain per cycle was much lower during the
Fig. 1a) observed, respectively, at cycles 10, 50, 1000 and cycling from cycle 651 to cycle 1000 than during the first
1693 (Fig. 7a–d) and the ε 11 strain maps obtained in the 50 cycles.
same zone between images taken, respectively, at cycles Unlike in the γ -phase, strain fields induced by cycling
1/4 and 10, 10 and 50, 651 and 1000 (Fig. 7e–g). in the α-phase during the first 50 cycles are relatively ho-
The analysis of Fig. 7a–d allows, in particular, the obser- mogeneous. Between images taken at cycles 1 /4 and 10,
vation of the progressive appearance of slip markings re- the total strain is lower than 0.67% almost everywhere
sulting from the emergence of dislocation slip bands at the (Fig. 7e) and between images taken at cycles 10 and 50, it
surface. These images show that slip band markings ap- is lower than 0.28% (Fig. 7f). Weak strain heterogeneities
peared firstly in the austenitic phase, very early during the were observed in the α-phase up to cycle 50 (Fig. 7f) and
cycling. Indeed, the first markings were observed as early intensified only after cycle 300. The cycling from cycle
as cycle 10 in some γ -grains (see arrows in Fig. 7a). Then, 651 to cycle 1000 introduced strong strain heterogeneities
slip marking number and relief increased quickly during in the α-phase (see arrows in Fig. 7g). Post-mortem ob-
cycling: slip markings were observed in the majority of servations by an optical microscope (Fig. 7d) confirm that
γ -grains at cycle 1000 (Fig. 7c). Morphologically, these this α-zone has been strongly deformed, as shown by the
markings in the γ -phase are straight, fine, parallel and presence of many slip markings. The strains induced be-
close together. These characteristics are typical of the dis- tween cycles 651 and 1000 reach ± 1.5%, that is to say, a
location planar slip, which occurs in austenitic steels (FCC level only slightly lower than that observed in the γ -phase
crystallographic structures) all the more easily when the during the same cycling range.
nitrogen content is high.5,6,8,34 These dislocation planar Figure 8a presents the surface damage in zone 2 (see
arrangements observed in the austenite of duplex stain- Fig. 1a) observed at cycle 1600. Figure 8b and c presents
less steels are similar to those observed in 316L-316LN ε 11 strain maps obtained in this zone between images
stainless steels.8 taken, respectively, at cycles 1 /4 and 10, 1000 and 1200.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 145

Fig. 7 Surface damage evolution in zone 1 (see Fig. 1a): images taken respectively at cycles 10 (a), 50 (b), 1000 (c) and 1693 (d); ε 11 strain
maps obtained for zone 1 respectively between images taken at cycles 1 /4 and 10 (e), 10 and 50 (f), 651 and 1000 (g).


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
146 A . E L B A RTA L I et al.

Microcrack initiation and propagation

Figure 9 presents the surface damage in zone 3 (see
Fig. 1a) at cycles 1000, 1200 and 1600 (Fig. 9a–c), and the
strain maps obtained in this zone between images taken,
respectively, at cycles 1/4 and 10, 10 and 50, 651 and 1000,
1000 and 1200 (Fig. 9d–g).
Figure 9b shows a microcrack initiated in the α-phase
during cycling (see zone A in Fig. 9a). A forewarning of the
initiation of this microcrack was observed as early as cycle
1000 (Fig. 9a) through an intensification of slip markings
in this zone. At cycle 1450, some propagation of this mi-
crocrack was observed in the neighbouring γ -grain on the
right, firstly straight and then it changed its direction to
follow a slip band marking in the austenite. At cycle 1600,
this microcrack had propagated in a γ -grain on the left,
following a slip band marking (Fig. 9c). After fracture,
this crack was clearly opened by the final fracture load
(not shown here), although it did not become the main
crack that led to final fracture.
This paragraph presents the evolution of the strain fields
calculated in zone 3 using the DIC technique during cy-
cling and the links between microcrack initiation optically
observed and the displacement and strain field evolutions.
As previously observed in zone 1, strain heterogeneities
in γ -grains in zone 3, resulting from plastic deformation
during cycling, were again observed as early as cycle 10
(Fig. 9d). In α-grains, the additional strains remain rel-
atively weak up to cycle 300. Let us now focus on the
evolution of zone A, where a microcrack initiation was ob-
served. Figure 9d shows a high strain gradient in γ -grain
1 (see arrows). This strain gradient was initially located in
the left half of the grain (before cycle 10, Fig. 9d) and then
spread through the whole of γ -grain 1, up to the neigh-
bouring α-phase on the right (between cycles 10 and 50,
Fig. 9e). This high gradient split γ -grain 1 in two parts,
the upper part was in tension and the lower part in com-
pression, both with the same maximum strain value equal
Fig. 8 Surface damage evolution in zone 2 (see Fig. 1a): images to 2.5% approximately (Fig. 9d & e). Between cycles 10
taken at cycle 1600 (a), ε 11 strain maps obtained for zone 2 and 50 (Fig. 9e), γ -grain 3 also presented a high strain
respectively between images taken at cycles 1/4 and 10 (b), 1000
gradient, which split γ -grain 3 into two parts—the upper
and 1200 (c).
part was in compression and the lower part in tension. The
zone in tension spread in α-grain 2 on the right jointing
In Fig. 8a, intense rectilinear slip markings in the γ -phase the tension part of γ -grain 1 to form a single localisation
and wavy slip markings in the α-phase are still observed. band where the microcrack appeared later (Fig. 9e).
Annealing twins, highlighted by the orientation changes Between cycles 651 and 1000, the strain in γ -grain 1 was
of the slip markings, are also observed inside austenitic no longer localized and relatively homogeneous, the grain
grains. Moreover, some slip markings propagate from a was in tension and its additional strain was equal to 1.6 ×
γ -grain to an α-grain in a rectilinear way at the surface 10−2 (Fig. 9f). It must be noted that the additional strain
(see arrows in Fig. 8a). This characteristic can probably decreased from about cycle 651. In the time range (651–
be connected to the existence of a crystallographic orien- 1000 cycles), the strain was smaller than that measured be-
tation relationships between ferritic and austenitic grains. tween cycles 10 and 50. On the contrary, in the same time
In this zone, no strain localization is observed in the strain range, the additional strain in α-grain 2 increased. This
maps (Fig. 8b & c). grain presents a high strain gradient in Fig. 9f. Between


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 147

Fig. 9 Surface damage evolution in zone 3 (see Fig. 1a): images taken respectively at cycles 1000 (a), 1200 (b), 1600 (c); ε 11 strain maps
obtained for zone 3 respectively between images taken at cycles 1 /4 and 10 (d), 10 and 50, (e) 651 and 1000 (f), 1000 and 1200 (g).


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
148 A . E L B A RTA L I et al.

cycles 1000 and 1200, the additional strain in α-grain 2

stabilized but strain gradients became less intense. This
can be linked to the microcrack initiation observed in this
zone at cycle 1200 that produced relaxation in the vicinity
of the crack.
Figure 10a and b presents images of zone 4 (see Fig. 1a)
observed, respectively, at cycle 1000 by camera and af-
ter fracture by SEM. Figure 10c and d shows displace-
ment U 1 and strain ε 11 maps measured between images
taken at cycles 1000 and 1200. A microcrack initiated at
an α−α grain boundary and was observed at cycle 1200.
The high gradients of displacement (Fig. 10c) and of strain
(Fig. 10d) between the two sides of the α−α grain bound-
ary indicate a microcrack initiation.

DISCUSSION AND CONCLUSIONS

A low-cycle fatigue test was performed at ε t /2 = 5 ×
10−3 at room temperature on a shallow notched test spec-
imen in a nitrogen alloyed duplex stainless steel. Using an
in situ microscope, surface damage evolution was observed
in real time during cycling. Displacement and strain field
measurements at the notch tip were performed using a
digital image correlation technique between images taken
at the same place during cycling. The results are summa-
rized and discussed below.
High levels of strain heterogeneities were observed in γ -
grains from the very first cycles, which is consistent with
slip marking appearance very soon during the cycling. Af-
terwards, the number of the heterogeneity sites increased
gradually in γ -grains. The size of these zones of intense
strain increased progressively, until they covered some γ -
grains entirely. These features were confirmed by the high
number of slip markings in these γ -grains at the end of
the cycling.
α-grains deformed plastically, later than γ -grains. In
the first instance, strains in α-grains were initially lower,
strain fields were more homogeneous and the first slip
markings were observed later than in γ -grains. Strain
heterogeneities appeared later during the cycling, their
intensities were weaker than those observed in some γ -
grains. Slip markings appeared in these heterogeneity
zones, which did not affect all α-grains.
The more highly deformed zones were most often lo-
cated close to grain or phase boundaries. Grain and
phase boundaries play a key role in the strengthening
of polycrystalline materials because of the misorienta-
tion between neighbouring grains and the slip system
difference between the two phases, which limits disloca-
tion movement and causes internal stresses.36 This im- Fig. 10 Surface damage evolution in zone 4 (see Fig. 1a): Images
plies intense strain localizations at grain or phase bound- taken respectively at cycle 1000 by CCD camera (a) and after
aries, clearly visible in strain fields shown in Fig. 7e–g. fracture by SEM (b); U 1 displacement (c) and ε 11 strain maps (d)
However, some phase boundaries did not show this phe- between images taken at cycles 1000 and 1200.
nomenon, and some continuous and linear slip markings


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 149

through α-γ boundaries were observed. It is also well stress triaxiality level existing in a grain of a polycrystal,
known that crystallographic orientation relationships ex- due to the numerous surrounding grains, could explain
ist in DSS between ferrite (BCC) and austenite (FCC) the high strain heterogeneity observed in this grain. The
grains.9,15,37 Such orientation relationships, referred to as crystallographic orientations of the grains in the aggre-
K-S relationships (Kurdjumov-Sachs relationships), typ- gate also play an important role on the strain in each grain.
ically [110](1-11)γ //[111](1-10)α , express that some easy Moreover, strain gradients shown here concern only the
slip directions in one phase quasi-coincide with easy slip ε 11 strain component, in the loading direction. It could be
directions in the other phase. Such compatibility of easy interesting to go further and analyse all the strain compo-
slip systems in both phases favours the passing on of dislo- nents.
cation slip from austenite to ferrite through phase bound- A quantification of plastic strain levels achieved in the
aries, to further the local stress concentration caused by grains of a polycrystalline material after a mechanical test
dislocation pile-ups in the austenite.36 In the zone where is very difficult to obtain. Although it is theoretically pos-
such orientation relationships were observed, no strain sible to evaluate partly the plastic strain level reached in
localization was seen (Fig. 8b & c).This agrees with re- a grain from the number and the height of slip markings
cent observations performed by Marinelli et al.38 by TEM present at the surface of this grain, by AFM for exam-
just beneath the surface of a fatigued specimen of a SAF ple,39 these measurements are difficult to carry out. That
2507 DSS alloyed with 0.26% N. At ε t = 8 × 10−3 is the reason why most authors evaluate the beginning of
(ε p = 2.4 × 10−3 ), they observed an important role of the plasticity with the first slip markings and compare the
crystallographic correlation between phases. When K-S plastic strain levels between two grains using the number
relationships between BBC and FCC grains are satisfied, of slip markings but rarely quantify it. This quantification
the slip bands easily cross phase boundaries. On the con- can nevertheless be obtained by strain field measurements
trary, when the correlation between phases is lacking, mi- using DIC. Modern equipment and recent developments
crocracks initiated at the slip band gets arrested at the in DIC allow such measurements to be performed today
phase boundary. At ε t = 1.2 × 10−2 (ε p = 6.6 × 10−3 ), on the microstructural scale. The present work shows that
they observed no role of K-S relations. Both phases ex- strain field measurements performed on the microstruc-
hibit a more homogenous distribution of plastic deforma- tural scale are a good tool for studying fatigue damage
tion; however, extrusions-intrusions and microcracks are micromechanisms, especially in regard to strain gradients
observed preferentially in the ferrite. which are forewarning signs of microcrack initiation. The
In situ observations show several microcrack initiations development of 2D, even 3D, strain field measurements
in the analysed zone (Figs 9c & 10c). Although austenitic can give hope for better performance and accuracy of such
grains deformed plastically in an intensive way first dur- studies.
ing cycling, the first microcracks initiated in the fer-
ritic phase, preferentially at ferrite/ferrite grain bound-
Acknowledgements
aries.5,13–15 Initiation in ferrite can be explained by slip
mode differences between austenite and ferrite, influenced The authors acknowledge M. François Hild, from the
by nitrogen content. The nitrogen alloying changes the ‘Laboratoire de Mécanique et Technologie’ (LMT) of
slip mode in the austenite from wavy to planar, thus, in- ENS-Cachan, for his help concerning the use of Cor-
creasing the reversibility of dislocation slip under cyclic reliQ4 software and for fruitful discussions about the re-
loading and leading to a better spatial distribution of cyclic sults. The authors are also grateful to SRMA/LA2M of
plastic strain. This, therefore, modifies the surface damage CEA for the crystallographic texture analysis.
distribution because of a less marked intrusion-extrusion
relief. On the contrary, in the ferritic phase, irreversibility REFERENCES
of the dislocation slip due to easy cross slip produces a
1 Charles, J. (1991) Superduplex stainless steels: structures and
wavy surface relief with steeper extrusions and intrusions, properties, In: Duplex Stainless Steels 91 (Edited by J. Charles
which favours the crack initiation in ferrite. and S. Bernhardsson), Editions de Physique, Les Ulis, France,
Strain intense localization zones were observed in some pp. 3–48.
places from the first cycle. Microcrack initiations were 2 Nyström, M. and Karlsson, B. (1996) Fatigue of duplex
correlated with some of these high strain gradient zones stainless steel. Influence of discontinuous, spinodally
(Figs 9e & 10d). It must be noted that these initiation zones decomposed ferrite. Mater. Sci. Engng. A 215, 26–38.
3 Magnin, T., Lardon, J.M. and Coudreuse, L. (1988) A new
appeared very early during cycling and not only just before
approach to low-cycle fatigue behavior of a duplex stainless
initiation. Moreover, it could be surprising to observe two steel based on the deformation mechanisms of the individual
parts of the same grain split by a high gradient zone—one phases. In: Low-Cycle Fatigue, American Society for Testing and
deformed in tension, the other in compression—with ap- Materials: ASTM STP 942 (Edited by H. D. Solomon, G.R.
proximately the same strain level, but reversed. The high Halford, L.R. Kaisand and B.N. Leis), pp. 812–823.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
150 A . E L B A RTA L I et al.

4 Seddouki, A. (1994) Endommagement en fatigue plastique 21 Hild, H., Périé J. N., Lemon, J. and Puyo-Pain, M. (2005) On
oligocyclique à température ambiante d’aciers inoxydables the use of digital image correlation to analyse the mechanical
duplex austéno-ferritiques type Z2 CND 22-05 alliés à l’azote. properties of brittle matrix composites. In: Proceedings of 107th
PhD thesis. University of Lille 1, France. Annual Meeting, Exposition, and Technology Fair of the American
5 Degallaix, S., Seddouki, A., Degallaix, G., Kruml, T. and Polák Ceramic Society Baltimore, MD, USA.
J. (1995) Fatigue damage in austenitic-ferritic duplex stainless 22 Forquin, P., Rota, L., Charles, Y. and Hild, F. (2004) A method
steels. Fatigue Fract. Engng. Mater. Struct. 18, 65–77. to determine the macroscopic toughness scatter of brittle
6 Mateo, A., Llanes, L., Iturgoyen, L. and Anglada, M. (1996) materials. Int. J. Fracture 125, 171–187.
Cyclic stress-strain response and dislocation substructure 23 Hild, F. and Roux, S. (2006) Digital image correlation: from
evolution of a ferrite-austenite stainless steel. Acta Mater. 44, displacement measurement to identification of elastic
1143–1153. properties. Strain 42, 69–80.
7 Llanes, L., Mateo, A., Iturgoyen, L. and Anglada, M. (1996) 24 Roux, S. and Hild, F. (2006) Stress intensity factor
Aging effects on the cyclic deformation mechanisms of a duplex measurements from digital image correlation: post-processing
stainless steel. Acta Mater. 10, 3967–3978. and integrated approaches. Int. J. Fracture 140, 141–
8 Kruml, T., Polàk J., Obrtlı́k K. and Degallaix, S. (1997) 157.
Dislocation structures in the bands of localized cyclic plastic 25 Hild, F. and Roux, S. (2006) Measuring stress intensity factors
strain in austenitic 316L and austenitic-ferritic duplex stainless with a camera: Integrated digital images correlation (I-DIC). C.
steels. Acta Mater. 45, 5145–5151. R. Mecanique 334, 8–12.
9 Verhaeghe, B., Louchet, F., Doisneau-Cottignies B., Bréchet, 26 Besnard, G., Hild, F. and Roux, S. (2006) “Finite-element”
Y. and Massoud, J. P. (1997) Micromechanisms of deformation displacement fields analysis from digital images: Application to
of an austenoferritic duplex stainless steel. Phil. Mag. A 76, Portevin-Le Châtelier Bands. Exp. Mech. 46, 789–803.
1079–1091. 27 Sun, Y., Pang, J. H. L., Weng, C. K. and Su, F. (2005) Finite
10 Moussavi Torshizi, S. E. (1997) Ecrouissage cyclique d’aciers element formulation for a digital image correlation method.
inoxydables austenitiques: influence de l’histoire de Appl. Opt. 44, 7357–7363.
chargement. PhD thesis. University of Lille1, France. 28 Wagne, B., Roux, S. and Hild, F. (2002) Spectral approach to
11 Aubin, V. (2001) Plasticité cyclique d’un acier inoxydable displacement evaluation from image analysis. Eur. Phy. J. Appl.
austéno-ferritique sous chargement biaxial non proportionnel. Phy. 17, 247–252.
PhD thesis. University of Lille1, France. 29 Chu, T.C., Ranson, W. F., Sutton, M. A. and Peters, W. H.
12 Aubin, V., Quaegebeur, P. and Degallaix, S. (2003) Cyclic (1985) Applications of digital image correlation techniques to
plasticity of a duplex stainless steel under non-proportional experimental mechanics. Exp. Mech. 25, 232–244.
loading. Mater. Sci. Engng. A 346, 208–215. 30 Schreier, H. W. and Sutton, M. A. (2002) Systematic errors in
13 Polák, J., Kruml, T., Degallaix, S. and Nilsson, J. O. (1999) digital image correlation due to undermatched subset shape
Fatigue damage in austenitic-ferritic duplex stainless steel. In: functions. Exp. Mech. 42, 303–310.
Proceedings of Eighth International Conference on Mechanical 31 El Bartali, A., Aubin, V. and Degallaix, S. (2006) “Surface strain
Behaviour (ICM8), Progress in Mechanical Behaviour of and damage measurement in an austenitic-ferritic stainless steel”,
Materials, Vol. 1, (Edited by F. Ellyin and J. W. Provan). Euromech-Mecamat EMMC9, MORET SUR LOING, 9-12
Canada. pp. 47–52. May 2006. Texte paru in: Local Approach to Fracture, EMMC9
14 Polák, J. and Zezulka, P. (2005) Short crack growth and fatigue (Edited by J. Besson, D. Moinereau, D. Steglich), Ecole des
life in austenitic-ferritic duplex steel. Fatigue Fract. Engng. Mines de Paris, ISBN 2-911762-73-8, pp. 347–352.
Mater. Struct. 28, 923–935. 32 El Bartali, A., Aubin, V. and Degallaix, S. (2006) Fatigue
15 Alvarez-Armas, I., Marinelli, M. C., Malarrı́a J. A., Degallaix, damage micromechanism in an austenitic-ferritic stainless
S. and Armas, A. F. (2006) Microstructure associated with crack steel. In: Proceeding of the 16th of European conference of Fracture,
initiation during low-cycle fatigue in a low nitrogen duplex Alexandroupolis, Greece, CD-ROM.
stainless steel. Int. J. Fatigue 29, 758–764. 33 El Bartali, A., Aubin, V. and Degallaix (2006) In situ surface
16 Marinelli, M. C., Degallaix, S. and Alvarez-Armas, I. (2006) strain measurements during low-cycle fatigue tests. In:
Dislocation structure developed in the austenitic-phase of SAF Proceedings of Photomechanics 2006, International Conference on
2507 duplex stainless steel. Mater. Sci. Engng. A 435–436, full-field measurement technique and their application in
305–308. experimental solid mechanics, (Edited by M. Grédiac and J.
17 Düber, O., Künkler, B., Krupp, U., Christ, H. J. and Fritzen, Huntly). Clermont-Ferrand, France.
C. P. (2006) Experimental characterization and 34 Degallaix, S., Taillard, R. and Foct, J. (1984) Role of nitrogen
two-dimensional simulation of short-crack propagation in an interstitials in plastic fatigue of austenitic stainless steels. In:
austenitic –ferritic duplex steel. Int. J. Fatigue 28, 983–992. Fatigue 84, Vol. 1, (Edited by C. J. Beevers), pp. 49–
18 Stolarz, J. and Foct, J. (2001) Specific features of two phase 59.
alloys response to cyclic deformation. Mater. Sci. Engng. A 316, 35 El Bartali, A., Aubin, V., Sabatier, L., Quaegebeur, P. and
501–505. Degallaix, S. (2005) Fatigue damage micromechanism analysis
19 Photomechanics 2006 (2006) International Conference on method in austenitic-ferritic stainless steel. In: Proceeding in
full-field measurement technique and their application in Stainless Steel World 2005, Maastricht, Netherlands,
experimental solid mechanics (Edited by M. Grédiac and J. pp. 290–294.
Huntly). Clermont-Ferrand, France. 36 Zieliński, W., Światnicki, W., Barstch, M., and Messerschmidt,
20 Bergonnier, S., Hild, F., Rieunier, J. B. and Roux, S. (2005) U. (2003) Non-uniform distribution of plastic strain in duplex
Strain heterogeneities and local anisotropy in crimped glass steel during TEM in situ deformation. Mater. Chem. Phy. 81,
wool. J. Mater. Sci. 40, 5949–5954. 476–479.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151
 FAT I G U E DA M AG E A N A LYS I S I N A D U P L E X S TA I N L E S S S T E E L 151

37 Fréchard, S., Martin, F., Clément, C. and Cousty, J. (2006) displacement is small enough in amplitude, a first order
AFM and EBSD combined studies of plastic deformation in a Taylor expansion of g is performed, and the displacement
duplex stainless steel. Mater. Sci. Engng. A 418, 312–319. is estimated through the minimisation of linearized 
38 Marinelli, M. C., Degallaix, S. and Alvarez-Armas, I. (2007)  
Short crack initiation during low-cycle fatigue in SAF 2507
(v) = [g(x) − f (x) + v(x).∇g(x)]2 dx. (A.3)
Duplex Stainless Steel. Key Engng Mater. 345–346, 343–346.
39 Villechaise, P., Sabatier, L. and Girard, J.C. (2002) On slip
band features and crack initiation in fatigued 316L austenitic The displacement field is decomposed as a linear com-
stainless steel: Part 1: Analysis by electron back-scattered bination on the basis of known functions ψ i
diffraction and atomic force microscopy. Mater. Sci. Engng. A 
323, 377–385.
v(x) = ui ψi (x), (A.4)
i

where ψ i are known functions and u i are the vector ampli-

A P P E N D I X : D I G I TA L I M A G E C O R R E L AT I O N tudes to be determined. The problem consists of solving
In this section, the principles of the digital image corre- a linear system that is written in a matrix form as
lation technique used in this work are summarized. Cor- [M] {w} = {m} , (A.5)
reliQ4 software was used to perform displacement field
measurements.23–26 Two digital images of the same anal- where {w} is a vector containing all the unknown com-
ysis area taken at two different times are considered. The ponents v i , [M] and {m} are known quantities depending
first image is called the ‘reference image’, and the second on f , g and ψ. In CorreliQ4 software, the basis functions
image is the ‘deformed image’ (Fig. 3). The grey level dis- are defined as bilinear functions for each component of
tribution of the reference image and the deformed image the displacement over square elements. They are Q4P1-
are defined by f (x) and g(x) respectively, where x is the shape functions used in the language of the finite element
position field of the pixels in the image considered, such method. The element is made up of four nodes of the
that quadrilateral. Each element is mapped onto the square
[−1, 1]2 , where the four basic functions (i.e. P1 polyno-
g(x + u) = f (x) + b(x), (A.1) mials) are (1−x)(1−y), x(1−y), (1−x)y and xy in local (x,y)
where u(x) is the unknown in-plane displacement field, frame.26
and b(x) is the noise induced by image acquisition, which is In practice, the reference image is subdivided into small
neglected either because of its low amplitude with respect zones. These small zones define the elements with size l
to those of f and g or because of its scale separation with in pixels. The corner of each of these zones is a node of
significant components of the displacement field. the mesh (Fig. 3). The elementary matrix Me ij and vector
An objective function (v) is introduced,23,24 operating me j are calculated for each element using Eqs (3) and (4).
on displacement fields v(x) Then, an assembly of elementary matrices and vectors is
  carried out as in any finite element analysis. The displace-
(v) = [g(x + v) − f (x)]2 d x. (A.2) ment field is then calculated by inversion of Eq. (5). The
displacement field obtained is globally continuous.23 The
This functional reaches its minimum value, 0, for v = u strain field is finally calculated as the discrete derivation
(see Eq. (1)). When f and g are sufficiently smooth and the of the displacement field or by using the shape functions.


c 2008 The Authors. Journal compilation 
c 2008 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 31, 137–151