Professional Documents
Culture Documents
ABSTRACTmThis paper presents an imaging technique de- no longer homogeneous. These localization phenomenahave
veloped to study the strain localization phenomena that oc- long been under investigation by many authors from experi-
cur during the tension of thin, flat steel samples. The data mental, theoretical and numerical points of view.2-I~ These
are processed using digital speckle image correlation to de- works call for the detailed knowledge of the constitutive equa-
rive the two in-plane components of the displacement vectors.
tions. Consequently, it is of a great importance to be able to
The authors observe that the calculation of the intercorrela-
tion function reveals a systematic error and propose a numer- detect, with even greater precision, the onset of such local-
ical method to limit its influence. Plastic incompressibility and ization mechanisms.
thin-sheet assumptions are used to derive the third displace- The computational resources currently available offer
ment component and, hence, the various strain and strain new and powerful possibilities of experimental data pro-
rate components. Numerous checks are presented at each cessing. For instance, optical methods using digital im-
step in processing the data to determine the final accuracy of age correlation provide whole-field displacement data for
the strain measurements. It is estimated that this accuracy is plane structures.]1-14 The most precise methods take into
quite sufficient to track the inception and the development of account the deformation of the surface due to straining
localization. Examples of possible application are presented between two images. ~3 Nevertheless, these methods use
for mild steels whose strain localization mechanisms appear iterative algorithms and generally require considerable
to be precocious and gradual.
computation time. In addition, to track the localization de-
KEY WORDS--Digital image correlation, strain rate measure- velopment precisely, the strain field must be determined on
ment, necking, material behavior, steel high-resolution grids. This fact led us to develop a suit-
able direct correlation method. It enables such resolutions
to be achieved within a reasonable computation time using
a standard microcomputer. The third displacement compo-
Over the past two decades, the simulation tools used in nent can be obtained by assuming material incompressibility
mechanics of materials have become increasingly powerful. and depthwise homogeneity of the strain field (thin-sheet as-
Progress made in the field of scientific computation is giv- sumption). The strain and strain rate tensors are derived by
ing rise to a new branch of mechanics called computational differentiation. Because the displacement fields are discrete
mechanics of solids.1 It allows the use of more realistic, and and noisy, a local polynomial fitting method is implemented
also more complex, models of behavior. to estimate the space and time derivatives.
Of course, the consistency of the simulation results de- This paper is divided into six sections, followed by con-
pends very much on the validity of the constitutive equations cluding comments. The first three focus on the different steps
implemented in the computational codes. These phenomeno- in the image processing. The first is devoted to computation
logical equations are, most often, identified on the basis of of the two in-plane components of the displacement vector.
experiments carried out on particular structures for which The second describes the calculation of the out-of-plane com-
stress and strain patterns are supposed to be known. For ponent. The third presents the differentiation method. The
instance, we assume that strain and stress fields are homo- fourth section deals with the tests developed to check the ac-
geneous in the gage part of a sample during a simple tensile curacy of the displacement and strain measurements. Next,
test. This homogeneity hypothesis, often implicitly assumed, the experimental setup is introduced. Finally, potential appli-
is necessary to estimate the stress and the strain from load cations of the experimental approach are proposed through
and displacement measurements. Naturally, as soon as a results describing the localization development in tensile tests
strain localization occurs--local necking, LiJders bands, performed on standard thin, fiat steel samples.
transformation bands and so on--the mechanical fields are
In-plane Components of the Displacement Vector
ExperimentalMechanics 9 29
resolution, the grid step of the mesh (G S) is generally chosen ment field spanning n = 104 points, the computation time is
between 2 and 5 pixels corresponding to about 10 4 points of around 22.103 s for the complete calculation, whereas it is
measurements per couple of images. only around 222 + 42 = 264 s when an interpolation of the
displacement field calculated on n = 102 points is used to
IntercorrelationFunction compute the whole displacement field (see Table 1). Using
the first estimate of the displacement field, R Z ranges from
To determine the displacement at each point of the grid, 3 to 5 pixels.
we choose to perform a direct correlation computation and
use a normalized intercorrelation function ft. Between two
images and I1 and 12, separated by a small strain increment, Out-of-plane Components of the Displacement
~3 is written as Vector
~(k, l) Assuming the plastic incompressibility, the out-of-plane
k,l e [ - RZ, RZ] = component of the displacement vector can be deduced from
cz/2 the in-plane components. In the finite transformations frame-
work, the local equation of the volume balance leads to
E
i,j=-CZI2
11(io + i, Jo + j) " I2(io + i + k, jo + j + l)
det(F(M)) = 1, (2)
I cz/2
where F(M) is the deformation gradient tensor at point M.
E 11(io + i, j0 + j)2
Its representative matrix is defined by
i,j=-CZ/2
1 1 Ou Ou 3u ]
l ~2 I2(io+i+k, j o + j + l ) 2 F = 7x 1 + -- ~ , (3)
i,j=-CZ/2 3w 3: y ow
(1) 1+
where CZ is the correlation zone (i.e., M's neighborhood where ~ = (u, v, w) is the displacement vector at point M
corresponding to a local optical signature of grid point M) and (x, y, z) are the Lagrangian coordinates.
and RZ is the research zone (i.e., M's neighborhood where Furthermore, if the transformation is supposed to be ho-
the optical signature is tracked). These zones correspond to mogeneous throughout the thickness of the sample (thin-
the domains of variation of (i, j) and (k, l), respectively. sheet assumption), eq (2) becomes
The intercorrelation function ~ is computed for multiple
pixel shifts (k, l). So the displacements are estimated with a -
Experimental Mechanics 9 31
0.15j : :
] |--svst. err.] , ,
0.1 ~--J s~d. dev. I--n .......
- - r......
( ~ I I I
I I 1 I I
0 . 0 5 : . . . . . . t . . . . . . . . . . . . . . .
I t d ! t
. . . . . . ~ . . . . . . 9 . . . . . . . 9 . . . . . .
-0.05 . . . . . . . . . . . . . . . . . . . .I. . . . . . . . . . I I
I I I
(a) (b) I I I
./~.1
U I ........ I I I
Fig. 2--(a) Experimental and (b) computer-generated 140 x ....... =r . . . . . .
i
"1 . . . . . . .
i I
140 pixel speckle images i
i
i
i
I
I
-0.15 I , I
I I
Translation motions applied to computer-generated 2000 ..... 7. . . . . . . . . 7 ....
speckle images show that the mean error on the displace- 0 i I i
ment measurement (eu) depends on the imposed displace-
ment (uirap). It is defined by -O.O5 0 0.O5
0 0.2 0.4 0.6 0.8 l With iterative methods, 13'2~ the correlation computation
uimp (pixel) converges gradually toward autocorrelation (zero displace-
ment). As a consequence, these methods suppress, by con-
Fig. 5--Systematic error corresponding to a direct correlation struction, the systematic error but remain time consuming.
computation (CZ= 20 pixels, n = 102, points). The size of large A faster method, developed by in Refs. 14 and 19, is based
spots is around 10 • 10 pixels, and the size of small spots is
on an approximation of the systematic error by a sinusoidal-
2 x 2 pixels
shaped function. These works correct the displacement mea-
surement by the value of the corresponding modeled system-
atic error. However, this method does not take into account
TABLE 2oMAXIMUM SYSTEMATIC ERROR AND STAN- the dependence of the systematic error on speckle properties.
DARD DEVIATION FOR VARIOUS MEASUREMENT METH- Our proposal is to use some characteristics of the system-
ODS ON SMALL-SPOT SPECKLE IMAGES: (a) DIRECT
atic error to get rid of this error: 1-pixel periodicity and null
CORRELATION COMPUTATION, (b) DIRECT ITERATIVE
CORRELATION COMPUTATION, (c) RANDOM-SHIFTING DI- average over one period (see Figs. 4 and 5). The aim of this
RECT CORRELATION COMPUTATION AND (d) FOURIER IT- technique is to convert the systematic error into a random
ERATIVE CORRELATION COMPUTATION one to suppress it by a subsequent filtering. To that end, the
Standard CPU Time correlation zone of each point of the final image is numeri-
max (esyst) Deviation (arbitrary cally translated by a 1-pixel-amplitude random displacement
(pixel) (pixel) units) (Uira,d) using a bilinear interpolation of the image. The com-
(a) 5.10 -2 2.10 -2 1 puted displacement of the ith point can thus be written as
(b) 3.10 -2 2.10 -2 3
(c) 10 -2 6.10 -2 1 i
Ucomp (
i p q- Urand
~- Uim i )
+ esyst Uim ( ' ) d q_ erand.
i p q.. Uran '
(d) 10 -3 6.10 -3 20 (8)
Because it depends on a random displacement, the sys-
tematic error computed at each point will become indepen-
i i i i dent of that computed for its neighbors; it is randomly re-
I i I I
distributed throughout the whole image (see Fig. 7). Then,
0.04 . . . . . . I ..... }- l II...... Uran
d i is subtracted to Uco,npi, and, using the property of null
I j.~. I I ~ I
J O~rtI[ I i~ ~k average, it is possible to greatly reduce the amplitude of the
( i i \ using the
. . . . . . . . . f . . . . superimposed random error esyst ~uim p + UraM)
z ', /I ', \ smoothing method described earlier.
% /
~ ' - - ' - : . - 9 - 7 , - . 9 . 9 , Figure 7 shows the displacement profile along the sam-
ple axis between two homogeneously deformed images, ob-
- tained using direct correlation (Udirect) or random shifting
(Urand.shift) computations. We can observe the influence of
. . . .
the systematic error on the displacement measurements and
spots the efficiency of the random shifting method. Figure 8 gives
the strain profiles related to Fig. 7. We can also observe that
-0.06 ,I w I ' ~ r the maximum error for strain has been considerably reduced.
0 0.2 0.4 0.6 0.8 1 Table 2 shows the comparative performances of the dif-
ferent measurement techniques that we implemented. In the
uimp (pixel) case of translation motions, the most efficient but most time-
Fig. 6---Systematic error plus drift corresponding to a direct consuming method is the Fourier iterative computation (d).
Fourier correlation computation (CZ = 20 pixels, n = 102 Direct methods (a, b, c) are much faster, and the random-
points) shifting direct correlation computation (c) gives a good com-
promise between speed and accuracy.
ExperimentalMechanics 9 33
correlation computation with regard to rotation movements
using numedcaUy rotated images. Imposed incremental ro-
tation angles were taken between 0 deg and 20 (leg. We
016
0"8 :'
~ ~ "direct ;t ~i ~ -i - -=~ noticed that the variations of the intercorrelation function
were strongly affected by rotation motions. For instance,
LX X Z/rand shift [ II~ "~
Figs. 9(a) and 9(b) show the variations of ~ in the vicinity of
o 9 - .... -r . . . . ~ .... ~ ..... '1 . . . . . M(io, jo) using the same R Z in the case of a translation mo-
tion and of a 20 deg rotational movement, respectively. We
I i.~l" I I can clearly see that there is only one correlation peak within
= 0.4 ..... ~---~-~,-----,
I x2c!v-_v-_v-_v-_v~ I
. . . . -~ . . . . .
i
RZ for the translation, whereas several peaks can coexist in
i ...,~ff-.r162 i i i the rotation case. Besides, the peak that corresponds to the
. ! :k~r s i ! rotational movement is not necessarily the highest. To find
0.2 - - - - . - ~ a ~ - - . - ~ .... § .... -4. . . . . the fight peak, we compute a first estimate of the displace-
x-, , . ~ - I t I
4.,~d~ I I I t ment field using small values of R Z.
I ) J J
0 ..,~" i i i I The choice of a small R Z allows only for the calculation of
0 100 200 300 400 500 small displacements, which correspond here to points located
in the vicinity of the rotation center. Far from this center, the
X (pixel) data processing does not provide correct values of displace-
merits. Once the rotation angle is determined, the same low
Fig. 7--Displacement profiles using either direct or random
value of R Z can be used for the whole image, providing a sin-
shifting methods (CZ= 2 0 p i x e l s )
gle correlation peak within the research zone. This method
gave correct results for incremental rotation less than 20 (leg.
In Fig. 10, the evolution of the error was plotted for angles
varying from 0 deg to 20 deg. Note that an angle of 20 (leg
x10-3 between two consecutive images can still be found with an
2.8 i i approximate mean error of 0.1 (leg and a local accuracy of
1 deg.
~direct ~_~ J
2.52 ,., , . . . . .
Checking Out-of-plane Displacement Component
The computational procedure providing the out-of-plane
2.24 . . . . .
I ,-~ Ix I . . . . .
displacement component is checked experimentally by
recording images of a steel sample mechanically loaded until
localized necking occurs. The sample was then unloaded and
removed from the testing machine, and its dimensions were
1.96 i - ~ ~ ' ~ . .]. . . . . T. . . . . . . measured using a three-axis measurement machine (Renault
Automation, resolution approximately 1 Ixm). The temporal
1.68 ~ - - - = " ~ . . . . -4--. . . . 4-. . . . -~-~L=-- evolution of the out-of-plane displacement component was
/ J I i i
/ i i I i computed using eq (4). The elastic strain being neglected
I l t t l compared with the plastic strain, the final computed geome-
1.4 t_ t , i I try was then compared to the measured one. Figure I 1 shows
0 100 200 300 400 500 the comparison between the computed and measured thick-
nesses of the sample along its longitudinal axis. Results are
x (pixel)
in good agreement and show the global consistency of the
Fig. 8--Strain profiles related to displacement profiles of hypotheses made throughout data processing.
Fig. 7 ( A Z = 5 0 p i x e l s )
Checking Strain Measurements
As mentioned earlier, we decided to distinguish situa-
tions in which the deformation is homogeneous from cases
The order of magnitude of the CPU time consumption is in which it is heterogeneous. The differential operators are
given in arbitrary units. With a Pentium MMX 166 MHz, and estimated with noisy displacement data and can be compared
for CZ = 20 pixels and R Z = 3 pixels, this unit achieves 238 with analytic results. The strain resolution mainly depends
subsets per second. Note that direct correlation and random- on the parameter AZ, the number of approximation points
shifting direct correlation algorithms have similar CPU time within AZ, the incremental strain between two images and,
consumption. This stems from the shifting operation, which for a cumulative computation, the number of strain incre-
is much faster than the subsequent correlation computation. ments, Because we choose to work with a fixed grid step
(typically, G S = 5 pixels), the number of approximation
Sensitivity to Rotation Motions points is given by the value of the parameter AZ.
Direct correlation computations can give only the local For images deformed in a homogeneous way, the reso-
translation components at any point of the grid. They are lution on the strain measurement increases with the use of
not able to determine the local rotation of CZ. Nevertheless, higher values of parameter AZ (see Fig. 12). The mean error
this can be estimated, provided the rotation angle between is the difference between the imposed strain increment and its
two images remains small. We studied the sensitivity of the mean measured values. The standard deviation of the local
20
) ~
........;:::' :;;iii...........
2 0 ii o
RZ(pixel) o o RZ (pixel) oo
RZ (pixel) RZ (pixel)
Fig. 9--Intercorrelation function variations: (a) rigid-body translation, (b) rigid-body rotation (CZ= 20 pixels, RZ= 10 pixels)
0.5 i i x 10 "3
I I 2.5 70
I I
I. .I.
0.4 i I 2 56
i I II II 1 i i I ." i
error i I_ __I 9 I I I/ I " C'~
0.3- ~ std. dev. ] r- -1- l--7--'r .... -~. . . . . I; . . . . n ..... 42 ~=
I I I ~.J
/ t I I #'1 I --~.
0.2 .... "~I. . . . . . +I *I - -/ ~ - - - i 1 - - [ 9' - ' t " , . . . . i . . 9 . ,. .
-'1- i 28~
I I 1 , 4 '
F ,
I
, 71---
Io
mean I! -'-
/ I, L I--stddev /I
0.5 F----t---+--z- - u - - - ~ " " ~J 14
o.1 .... ~ ...... I-~~, ..... 7~ / ~ I I'--" CPU timell
.. ~ s ', / / ,..-- " I" '- ,,J I 'l ' I
O" i -~---*-- ~. I0
0 ~"~r-~---~- "/- -~'- v ":'~" ~ 0 25 50 75 100 125
0 5 10 15 20 AZ (pixel)
rotation angle (~ Fig. 12~Strain measurement between homogeneously de-
Fig. 10--Error on the computed rotation angle (n = 104 formed images (n = 8500 points, CZ = 20 pixels, A t = 0.2
points) percent)
x 10 -3
I I 3 i i i .
' I i #
I
,,=.1
r,#l
2.5
2
1-~176176
- - std. dev.
I
I
I
I
j
TT-
,
. . . .
I I I
~ 3 ....... "] .... '=-~ ;--~'---i o 1.5 1 T--7 . . . . .
I I t ',
ra~
1 b'
1 ~----,--<,
I It I I S I
I~ I
0.5 . . . . . . ~.
s I I
-.-- 3D measurement I 0 " I i ,
2' 0. o computation I 0 0.1 0.2 0.3 0.4
0 20 40 60 80 imposed strain
longitudinal axis (ram) Fig. 13--Evolution of the error on strain measurement
according to the incremental strain between the two images
Fig. 11---Comparison between computed and measured out- (CZ=20 pixels, A Z = 50 pixels, n = 104 points)
of-plane displacements
strain error is given. Figure 12 also shows that the CPU time mean error remains small (i.e., 2.10 -4) even for a~. = 30 per-
consumption varies, approximately with the square of A Z . cent, whereas the amplitude of the local error increases from
Figure 13 illustrates the dependence o f the strain measure- 2.10- 4 at At = 0.5 percent to 5.10- 3 at At = 35 percent.
ment error at fixed A Z according to the incremental strain Figure 14 illustrates a penalizing case corresponding to the
(At) between the two homogeneously deformed images. The existence of a narrow localization zone (size of around 200
ExperimentalMechanics 9 35
O.O5 6 x 10 -4
- - - imposed ~ i
0.04 - - - A Z = 15 pix. _.~__J 5 I II . 1I ~ ,-
....... A Z = 25 pix. ~'~ I /
4 'or il --
0.03 -- Az = 5o -l-' .-i
aB /
std. dev. I i / t
....... A Z = 100 pix f " l II
~ ~ . a l i-i I I
0.02 . . . . . . ~ . . . . . . -r_,tl---~.,
"= 3t I -4-'-- ....
J J/i l"J
] I ..'~' IL~ I I ,tl I
.' ...... i I j I I
0.01 . . . . . . I
. . . . . -" . . . . . . I "- . . . .
2 t__ I. . . . . . -~. . . . . . ~----
I~ 1 j ~ I I
E /% ~- ] i
1 l- ""~ 4 ....
0 200 400 600 800
x (pixel)
Fig. 14--Strain resolution for the case of a heterogeneous 0 0.1 0.2 0.3 0.4
strain field (CZ= 20 pixels)
imposed strain
Fig. 15---Evolution of the strain measurement error according
to the number of strain increments (CZ= 20 pixels, A Z = 50
pixels). As expected, we verify that increasing A Z tends to pixels, n = 104 points, As = 0.5 percent)
enlarge the size of the localization zone and to underestimate
the maximum strain, which leads to an increase of the strain
measurement error.
Using this kind of example, we found that A Z values rang-
ing from 25 to 50 pixels are a good compromise between
precision and CPU time, whether or not the strain field is
homogeneous.
The accuracy of the strain measurements is also related to
the number of displacement increments. This dependence is
illustrated in Fig. 15. The strain is computed by cumulating
the incremental displacement vectors while the strain incre-
ment between two consecutive images is equal to 0.5 percent.
We can see that the mean error remains small (< 2.10 -4)
whichever the number of increment. We also observe that
the local error decreases at first, then increases due to the
accumulation of incremental errors. However, the standard
deviation reached for an imposed deformation of 35 percent
is lower for the case of an incremental calculation compared
with the case of a direct one.
Experimental Arrangement
300 ~ x
i I !
<
I I
50 mm 2~ 200 ...... r i ---
I I
z I I
----- sample with defect
Fig. 17--Heuristic sketch of the geometrical sample defects
100 ---
- - sample without defect
I 1 I
I I I
0 I I I
the strain field associated with this rigid-body motion (iden- 0 0.1 0.2 0.3 0.4
tically equal to zero) and the measured strain field enables
us to check and optimize the alignment procedure. The typ- ~ITltlcro
ical error on the strain measurements, due to the setting up, Fig. 18--Macroscopic tensile curves: nominal stress ~N =
is about 1.10 -4 or 2.10 -4 . Note that this error corresponds F/SO versus conventional strain Smacro = LXl/Io, where F is
approximately to the strain measurement resolution. the applied load, So is the initial cross section and At and to
are the elongation and the initial gage length, respectively
Examples of Potential Application
The experimental setup is designed to study localization
phenomena in mild steels. To illustrate the capabilities and
the interest of such an approach, we give an outline of results
obtained during monotone tensile tests performed at constant
cross-head velocity (vc/~ = 0.04 ram.s-l). As long as the
strain field remains homogeneous within the gage part of the
sample, this kind of loading can also be interpreted in terms of
a strain-controlled test (Smacro ~'~ 7,10 -`$ s-l). The images
corresponding to the sample gage part contain 250 x 900
pixels (scale factor ,~ 85 l~m.pixel-]), and the frame rate is
about 0.2 image.s- 1.
It is well known that initial geometrical defects in the sam-
ple can affect the inception of localization phenomena? So
we tested samples with and without geometrical defects to
evaluate the sensitivity of our experimental procedure. These
geometrical defects of machining were deliberately created
and correspond to variations of the width along the sample
axis (trapezoid shape). The amplitude of the geometrical
defect is estimated using, once more, the three-axis measure-
ment machine (see Fig. 17).
The amplitude of the geometrical defect being small Fig. 19---Spatial distributions of the three in-plane compo-
(0.2 mm), the macroscopic tensile curves of the two spec- nents of the Hencky strain tensor E~, measured on the sam-
imens are very similar (see Fig. 18). As expected, the strain ple containing an initial geometrical defect, when the applied
reached before fracture of the sample containing an initial macroscopic strain ~macro is 12 percent
defect is lower than that of the defect-free sample. Two par-
ticular points named A and B are spotted on the tensile curves.
At the end of this section, they will help us to show that al- is heterogeneous within the gage part of the sample; exttxis
though the macroscopic mechanical responses are initially higher in the lower end of the sample. Naturally, the strain
similar, the local strain and strain rate distributions are very heterogeneity is concentrated in the neighborhood of the ge-
different. ometrical defect where the initial cross section of the sample
Figure 19 shows the three in-plane components of the is the smallest.
Hencky strain tensor (e H) for the sample containing an ini- Because the local strain state corresponds to simple ten-
tial geometrical defect when the applied macroscopic strain sion, and because the strain distribution is fairly homoge-
e,nacro is 12 percent (point A, Fig. 18). We can also remark neous along the sample width (i.e., Oy-axis), it is possible to
that the strain state within the sample is coherent with a simple summarize all the available two-dimensional information for
tension state: shear strain concentrations are observed in the a given loading state into a one-dimensional representation.
grip sections of the specimen and are nonmeasurable in the For that purpose, we plot only the axial distribution of the
gage part, whereas the lateral contraction ratio ~ = - ~xxH/ ~yyH
- tension component of the strain tensor. Figure 20 represents
is about 0.5. Apart from the grip sections, the strain field the tension component of the Hencky strain tensor (effx) of
Experimental Mechanics 9 37
0.12 i i i 348
I I I
0.1 ,~1 , ,
232 ~ B
i . . . . . . .
ml
0.08
i
0.17
i
0.25
i
0.33
x (pixel) ~macro
Fig. 20--Longitudinal profile SEx when the applied macro-
scopic strain P-macrois 12 percent Fig. 21--Time evolution of the longitudinal strain rate
F.xx(X, O, t) in s -1 for a monotone tensile test performed at
constant strain rate (~macro ~ 7.10-4s -1) on a sample of steel
containing an initial geometrical defect
the points located along the sample longitudinal axis (i.e.,
Ox-axis). With this representation, we can then compare the
amplitude of strain heterogeneity of the two samples. 337! " ' "
Strain-rate measurements will provide information about
the onset and development of the strain localization. We will
adopt the same one-dimensional description of the strain rate
state by plotting the tension component of the Lagrangian
strain rate tensor (Exx) of the points located along the sample 225
longitudinal axis. To observe, in the same graph, the time
evolution of the strain rate distribution, the strain rate profiles
are piled in chronological order. In addition, contour plots Z
are drawn and the strain rate intensity is colored in gray scale
to further simplify the data interpretation. The abscissa axis
t~ 112
stands for time, and the ordinate axis corresponds to space
(see Figs. 21 and 22). The loading curve is superimposed on
the graph to link global and local aspects of the mechanical
response. The horizontal axis (initially time) now stands for 0
the macroscopic strain, and the vertical axis stand for the
normalized load.
0.08 0.17 0.25 0.33
Figure 21 shows the time evolution of the strain rate for Emacro
the sample containing a geometrical defect. The strain rate Fig. 22mTime evolution of the longitudinal strain rate
field gradually concentrates along the sample axis, first giv- [~xx(X, O, t) in s-~ for a monotone tensile test performed at
ing rise to a diffuse necking, then to a localized necking. It is constant strain rate (~=acro ~" 7. lO-4s - 1) on a sample of steel
important to note that this concentration takes place early on without any detectable initial geometrical defect
during the loading in the sense that a strain rate heterogeneity
appears during the strain hardening long before the maximum
load is reached. Remember that according to some theories, In all the tests performed, the necking is progressive. The
this maximum, which corresponds to the so-called ConsidEre presence of a defect in the initial geometry of the sample plays
strain (~macro(B) = 26 percent), is associated with the ap- the role of a catalyst and simply brings forward the inception
parition of diffuse necking. At point B, we observe that the of the localization.
localization, in terms of strain rate, already covers 30 percent
of the gage zone. Concluding Comments
During the sample softening, the effects of localization
strongly intensify and, finally, lead to the sample fracture. This experimental approach provides the time evolution of
Figure 22 shows the response of a defect-free sample to various in-plane kinematic variables (displacement, velocity,
the same loading. The onset of the strain rate localization acceleration, strain, strain rate, rate of kinetic energy and so
emerges later on during the loading. A slight heterogene- on). Assuming plastic incompressibility allows the compu-
ity can be observed on-the lower end of the sample: point tation of the out-of-plane components. The validity check
A (8macro(A) = 12 percen0. This strain is less than half of the data processing shows that correlation techniques pro-
of the ConsidEre strain amplitude, for which emacro(B) = vide results sufficiently accurate to study the inception and
27 percent. the evolution of the localization phenomena. Furthermore,
Experimental Mechanics 9 39