Ext 717
Ext 717
Keywords Abstract
Digital Image Correlation, Strain, Fatigue,
Material Behaviors This is the fifth paper in a series on the experimental characterization of
shape memory alloy (SMA) wires. In this installment we focus on the use
Correspondence
of digital image correlation (DIC) to measure the strain field on the surface
B. Reedlunn,
of the wire. After a brief overview of the principles and mathematics behind
Department of Mechanical Engineering,
University of Michigan, 2350 Hayward St., DIC, two different thermo-mechanical tension tests using DIC are presented
Ann Arbor, MI 48109, USA to demonstrate the technique. The first experiment consists of Joule heating
Email: [email protected] a shape memory (SM) wire to induce the shape memory effect, using 2-D
DIC to measure the strain field. The second experiment measures the response
Received: August 25, 2010; accepted: of a superelastic (SE) wire to mechanical cycling at room temperature, using
November 29, 2010
3-D DIC to measure the strain field and an infrared camera to measure the
doi:10.1111/j.1747-1567.2011.00717.x temperature field. In addition to describing the experimental results, attention
is paid to specimen preparation and the two experimental setups. Many of the
challenges and precautions associated with using DIC are discussed, along with
practical recommendations for specimen speckle patterns, digital photography,
and data post processing.
then show results from two experiments using some- to measure in-plane deformations of nominally planar
what different setups, one on preconditioned SM wire samples.
and one on SE wire. The basic concept underlying DIC involves compar-
ing two digitized images of a specimen surface, taken
at different times, to obtain a quantitative, point-
Digital Image Correlation Fundamentals
by-point mapping of the surface deformation. The
Digital image correlation (DIC) is a noncontact optical surface deformation is obtained by optimizing a cross-
method used to measure full-field displacements correlation function to match the local gray-scale
on the surface of an object by tracking the intensity values of a reference image and a subse-
nonuniform random surface patterns of a specimen. quent image of the deformed specimen. Commonly,
The technique is fast, robust, scalable, and provides an a sequence of images is captured and correlated to
accurate method to determine the surface geometry, obtain the evolution of the object’s surface deforma-
displacements, and strains of a deforming object. tion with time. The reference image is usually taken
See Refs. 5 and 6 for further reading. It is to be the first image of the undeformed specimen
particularly useful to quantify strain fields that are to calculate the total displacement and Lagrangian
distinctly nonuniform, such as those arising from strain between it and any later deformed configura-
stress concentrations, and geometric or material tion. Alternatively, to aid correlation adjacent pairs
heterogeneities/discontinuities. (Note that standard of images in the sequence can be compared, and the
DIC algorithms are not well suited to sharp strain or reference image is updated in time to give incremen-
displacement discontinuities, such as kinks or cracks, tal displacements and strains between it and the next
but this is being overcome via new techniques.7,8 ) image.
For our case of uniaxially loaded SMA wire, large, The specimen surface, either covered by a natural or
but finite, strain gradients exist due to phase an artificially applied pattern, is digitized into a spatial
transformation. distribution of pixel intensities to apply a correlation
The DIC method was first developed in the function. The images are not compared pixel by pixel,
early 1980s by researchers at the University of but rather by matching local distributions of pixel
South Carolina.9 – 11 Initially, DIC was created to intensity subsets between reference and deformed
measure in-plane deformations of flat surfaces using images. For simplicity, the following describes how
a single camera (two-dimensional [2-D] DIC). As 2-D DIC is performed. Summarizing sections of
the methodology developed and computing power Chapter 5 in Sutton et al.,5 the optimal match
grew over the years, the technique was improved between subsets can be achieved by minimizing a
and expanded to three-dimensional (3-D) DIC in chosen correlation function, S, such as the simple
the late 1980s and early 1990s, using two or more
sum of squared differences criterion,
synchronized cameras to track the 3-D position of
points on nonplanar objects (developed by McNeill,
I
Luo, Faugeras, Chao, Sutton, Helm, and others12 – 14 ).
S= [Gi − Fi ]2 , (1)
Both 2-D and 3-D DIC methods calculate surface
i=1
displacements and their gradients, from which surface
Lagrangian strains, velocities, accelerations, and
strain rates can be calculated. (Note that 3-D DIC where Fi = F(xi ) represents the gray-scale value at
should not be confused with volumetric DIC, where pixel i at position xi = (xi , yi ) in the reference image,
displacements and other quantities inside the material and Gi = G(x∗i ) represents the gray-scale value at pixel
are measured, often using X-ray tomography,5,15 or i at position x∗i = (xi∗ , y∗i ) in the deformed image. The
confocal microscopy.16 ) The basic principles behind summation ranges from the first (i = 1) to the last
the DIC method have no inherent length scale pixel (i = I) in the subset. Frequently this criterion is
or time scale. Practical limitations on the use of improved by adjusting for lighting variations between
DIC are imposed by the accuracy of measurement the reference subset and current subset. If a subset
techniques at smaller length scales and high speeds, moves between areas of different lighting intensities,
but substantial progress has been made in expanding say from an area of high illumination to shadow, the
the limits of DIC as better microscopy and high- gray-level values should scale uniformly. Thus, we
speed photography instrumentation have become can replace G with b G in Eq. 1, where b is a scaling
available. Although 3-D DIC has obvious advantages, parameter. The optimum value of b can be found for
2-D DIC remains a very useful and cost-effective tool each subset by minimizing S with respect to b, by
calculating ∂v
∂y ) that were calculated for each subset (p2 , p3 , p4 , p5 ),
I the displacement gradients are typically recalculated
∂S
=2 [b Gi − Fi ] Gi = 0, from the displacements (u, v) at the center of each
∂b subset (a grid point). This way the displacement
i=1
I gradients at a point depend on the neighboring grid
Fi Gi
⇒ bopt = i=1 I 2
. (2) points, rather than relying entirely on the correlation
i=1 Gi analysis of one subset. From the gradients, Lagrangian
Substituting bopt Gi for Gi in Eq. 1, yields the strains (2-D here) can be determined by
normalized sum of squared differences, 2
J 2 ∂u 1 ∂u 2 ∂v
I εxx = + + ,
j=1 Fj Gj ∂x 2 ∂x ∂x
S= J Gi − Fi . (3)
j=1 Gj
2 2
i=1 ∂v 1 ∂u 2 ∂v
εyy = + + ,
At this point, it appears that a pixel in the reference ∂y 2 ∂y ∂y
image can be matched to a location in the current
1 ∂u ∂v ∂u ∂u ∂v ∂v
image only to within a pixel; however, subpixel εxy = + + + . (5)
2 ∂y ∂x ∂x ∂y ∂x ∂y
resolution is achieved by interpolating the gray-
level intensities of the subsets. Put another way, if Note that some degree of spatial smoothing is usually
the subset moves by 0.2 pixels, some of the pixels performed on the displacement gradients (and/or
will remain nearly the same gray-level intensity, strains) to remove the noise amplified by taking
but others near large gray-level gradients will not. derivatives of experimental data.
Using an interpolation scheme allows the matching The 3-D DIC method utilizes two or more cameras
algorithm to resolve this 0.2 pixel movement. that image the specimen from multiple viewpoints,
Also instead of being restricted to summing over in order to additionally capture the out-of-plane
gray-level intensities at exact pixel locations xi in motions. 3-D DIC determines point correspondences
Eq. 1, one can now sum over a list of optimally between the two images acquired from the cameras
placed sample points across a region of interest at the same instant. To measure both the initial
(ROI). shape and the deformation of an object, the relative
The reader may also notice that, as described thus positions and operating characteristics of the two
far, minimizing a chosen S will only produce accurate cameras must be determined through a calibration
results for rigid body translation of the subset. procedure using a series of images of a grid with
Subset rotation and deformation are accounted for a known pattern. Once the calibration parameters of
by introducing shape functions analogous to finite each camera and their relative orientation in space are
element shape functions. The most commonly used determined, the specimen shape can be reconstructed
shape function assumes strains within the subset are using triangulation.18
uniform: In general terms, the main parameters affecting the
accuracy of the method are the following:
∗ ∗
p0 1 + p2 p3
xi = x (xi , p) = + xi • Specimen surface specularity. Surfaces with ran-
p1 p4 1 + p5
dom, fluctuating, gray-scale intensity distributions
(4)
work best. Surfaces that have a matte color are
where p is a vector of parameters {p0 , . . . , p5 }. This much better than that those that are highly
allows us to write S = S(p). polished/reflective. Also, surfaces with monotone
The optimal match between the reference subset (uniform) or very gradually varying surface inten-
and the current subset is found by minimizing S sities may not work well. Consequently, unless the
with respect to p. The Newton-Raphson method pre-existing surface happens to be satisfactory, one
is currently the preferred means to determine p must often artificially apply a speckle pattern using
(although the recent Levenburg-Marquardt nonlinear paint or some other high contrast medium.
parameter estimation scheme also shows promise17 ). • Camera, lens, and environment. The usual rules for
Once p is found for all subsets, strains can be good digital photography apply here. One must pay
calculated as a post-processing step. The components attention to illumination, field of view (FOV), and
of the displacement u = x∗ − x are (u, v), which depth of field (DOF) throughout an experiment.
correspond to (p0 , p1 ) for the shape functions of Eq. 4. Also, it is usually best to perform DIC with direct
Instead of using the displacement gradients ( ∂u ∂u ∂v
∂x , ∂y , ∂x , line of sight to the specimen surface. Intervening
mirrors or other optics that refract light introduce are difficult to avoid. The error induced by out-of-
optical aberrations in the images, so we discourage plane motion can be minimized by either increasing
their use unless the distortions can be corrected the distance between the camera and the object
prior to correlation. when using a standard lens, using a telecentric
• Image spatial resolution versus speckle size. Image lens, or by estimating a functional form of the
quality can be improved by using a high pixel strain error from measurements of out-of-plane
resolution charge-coupled device (CCD) digital displacement at three points on the specimen.20
camera. Speckles should be large enough to An especially poignant example where ignoring
encompass multiple pixels in an image (over- out-of-plane displacements can produce inaccu-
sampled), but should not be so large that images rate measurements in the measurement of Young’s
cannot be correlated with the required spatial modulus of steels is illustrated in Appendix A of
resolution. Saved images should be of a high- Ref. 21.
• Field of view (3-D DIC only). The same material points
quality, uncompressed format, say bitmap or tiff
(not jpeg). should be in view of multiple cameras. This can be
• Gray-scale dynamic range. A higher number of dis- problematic for specimens with large curvatures,
such as for imaging of thin wires of interest here.
crete gray levels (dynamic range) captured by the
Out-of-plane measurement accuracy is improved
CCD camera and data acquisition system is gen-
by placing cameras at relatively large angles to each
erally better, but the usable dynamic range may
other, but small camera angles may be necessary
be limited by illumination and the contrast of
to achieve a sufficient FOV overlap when imaging
the surface pattern. A high bit A/D converter is
cylindrical objects.
advisable, and, if needed, the CCD signal can be
amplified to utilize the full dynamic range of the When using 3-D DIC, the camera-lens system
A/D converter. Incidentally, color CCD cameras are must be optimized to balance DOF and FOV for
not recommended, due to the way the CCDs the specific test under consideration. An insufficient
are physically constructed, where the introduction DOF can cause pattern defocusing due to out-of-
of color introduces a non-monotonic intensity scale plane motion and introduce substantial error in the
and reduces the effective dynamic range. calculated displacement fields. The specimen must
• Frame rate. The number of images taken must be suf- stay in the cameras FOV and stay in focus for the
ficient to resolve the motion of the specimen. More entire testing period. Assuming a fixed lens size, a
images are better, provided one has sufficient stor- reduction in aperture size will improve the DOF but
will require an increase in object illumination. A
age capacity for them. One can always throw out
reduction in aperture size may also cause light to
images, if not needed during post-processing. If an
diffract through the aperture causing an out of focus
insufficient number is taken, the relative displace-
image. An increase in the focal length of a lens will
ments may be too large between sequential images,
also increase the DOF but will decrease the angular
and the cross-correlation scheme may fail. This can
FOV.5
usually be remedied by manually selecting an initial
Other sources of error can arise from poor
guess for the Newton-Raphson minimization of S
pattern quality, poor specimen illumination, lens
described above, but this may be inconvenient. distortions, environmental disturbances, and complex
• Choice of interpolation function. As mentioned pre- testing environments (e.g., submerged specimens,
viously, gray-level intensity values and their high temperature testing). Sizeable error can also be
derivatives between pixels are typically interpo- introduced by the incorrect choice of speckle size and
lated to reconstruct image intensity patterns prior subset size. Sutton et al.5 recommend each speckle
to correlation. Higher order interpolation schemes should fill a 3 × 3 pixel array in the image, as this
increase the accuracy. A dramatic error reduction is gives near optimal spatial over-sampling. They also
achieved when going from a linear to a cubic inter- recommend that each subset contain at least 3 × 3
polation, with rather diminishing improvements for speckles to ensure reasonable matching accuracy.
fifth-order interpolation.19 Typically linear shape functions are used (such as
• Out-of-plane displacements (2-D DIC only). 2-D DIC is those of Eq. 4), so the selected subset must be small
imaged by one camera oriented perpendicular to enough that the strain within the subset can be
the specimen surface and is best used for a pla- assumed constant yet large enough to encompass
nar surface under primarily in-plane deformation. an area that has a statistically different pattern from
In practice, however, out-of-plane displacements the neighboring subsets. Computations show that
if the signal contains a dominant frequency, the as certain polymeric materials, ink can be used to dye
subset size should encompass at least half of this the specimen surface.
local wavelength in order to avoid misregistration A problematic issue during dynamic or extreme
(i.e., aliasing).22 Using a combination of speckle over- temperature testing, particularly under large strains
sampling, accurate interpolation between pixels and and when paint is used to apply the speckle pattern,
image quantization with at least 8 bits can produce is the tendency of the pattern to debond from the
displacement fields with an accuracy of ±0.02 pixels specimen surface. Recent experiments have shown
or better as shown in Ref. 19. that both the maximum surface strain and the
More detailed, step-by-step, recommendations strain rate are important factors when determining
for assembling an experimental setup with DIC if a speckle pattern will debond from a specimen
measurements are provided in section Experiment 2 surface.24 This can be mitigated by cleaning and
Setup. lightly roughening the surface prior to painting,
and by applying the paint immediately prior to the
experiment.
Specimen Preparation Speckle pattern quality can be evaluated through
gray-scale statistics,5,25 where an image intensity
The accuracy of the DIC measurements is especially histogram is plotted with the gray-scale intensity
dependent on the quality of the surface pattern, of a pixel on the x-axis, and the number of pixels
and since the pattern is usually applied artificially, at that gray-scale value on the y-axis. A typical
specimen preparation is worth discussing in some CCD stores 8 bits per sampled pixel, allowing 256
detail. This takes some practice to obtain good possible shades of gray to be recorded, so the x-axis of
results. The speckle pattern must be sufficiently the histogram typically runs 0–255. A non-Gaussian
non-periodic, isotropic, and of high contrast to (bimodal or otherwise skewed) image intensity profile
guarantee the uniqueness of the tracking signature can indicate a poor quality pattern, areas of local
for each subset. Prior work has shown that a smooth reflection, or other issues that can degrade correlation
transition between black and white within a given results. However, a Gaussian distribution alone does
speckle is preferable (as opposed to uniformly black not indicate a high-quality pattern for DIC, as an
speckles on a white background, which fortunately overly smooth intensity could be Gaussian but not
in practice is actually difficult to achieve) for accurate specular. Baseline translation and rotation testing is
measurements, and the pixel-to-speckle ratio must another indicator of pattern quality, where a known
be sufficiently large to enable over-sampling.19,23 Of rigid body displacement is applied to the specimen
course, the pattern application should not alter the and plotted against the DIC-measured displacements
mechanical or chemical characteristics of the sample. to assess the resulting quality of the measurement.
The pattern should have minimal local areas of
reflection, so the use of nonreflective matte paints
Experimental Results
and diffuse lighting is recommended. If there are
persistent reflective or otherwise corrupted pixels in Two experiments will be described below to demon-
the image, then the user should consider removing strate how DIC can be applied to experiments on SMA
these points from the correlation analysis. wire. The first experiment was performed at General
Optimal pattern application methods vary, depend- Motors R&D on Flexinol wire� R
subjected to Joule
ing on the required magnification. For microscale heating using a 2-D DIC setup. The second experi-
measurements, high-quality patterns can readily be ment was performed on SE wire at the University of
obtained by airbrushing. Airbrushes can be purchased Michigan using a 3-D DIC setup.
with different nozzles depending on the desired size
of the pattern, where a smaller nozzle diameter pro-
duces smaller paint droplets (speckles). The pattern Experiment 1: Joule heating response of SM wire
is often applied by lightly coating the sample surface In this first experiment, the deformation result-
with a light paint, letting this dry, then spraying the ing from the martensite-to-austenite (M+ → A)
coated specimen with a dark mist of paint. Reversing temperature-induced phase transformation of SM
the sequence, first dark paint, then light paint works wire was studied with 2-D DIC. In this simple
equally well. (More detailed suggestions for airbrush- setup, a prestrained SMA wire was heated above
ing will be provided in section Specimen Preparation.) its transition temperature at constant load to induce
For materials where paint does not adhere well, such contraction of the wire due to the shape memory
the Grasshopper. Tamron CCTV 75-mm focal length the strains from the LE (δe /Le ) or DIC (ε) rather
lenses (as shown in Fig. 5) have been successfully than the global strain (δ/L). The prescribed ‘‘global
used in previous experiments, but here we used strain rate’’ was δ̇/L = ±5 × 10−4 s−1 , which should
75-mm Fujinon HF75SA lenses instead (not shown) be emphasized again, is quite different than the
which had a maximum aperture diameter of 41.6 mm ‘‘local strain rate’’ when transformation fronts are
(vs. 19.2 mm for the Tamron lenses). This meant it propagating.
was possible to rely on the diffuse fluorescent lights In setting up the cameras, many of the same tech-
described below and use very little direct lighting, niques known to any photographer were employed.
which can cause glare. Here is the procedure we used:
The specimen was clamped between hardened steel
1. Clean the lens and protective glass in front of the CCD.
plates within pneumatically actuated grips resulting
in an axial (x-direction) free length of L = 36.96 mm. If a certain area of a speckle pattern will not cor-
It was loaded uniaxially in displacement control by an relate, and there does not appear to be anything
Instron 5585 electro-mechanical, lead-screw driven, wrong with the focus, lighting, or the speckle
load frame, where the lower grip was fixed and the pattern, then the problem may be dust on the lens
upper grip displacement (δ) was controlled and mea- or protective glass in front of the CCD. However,
sured by the load frame’s high resolution optical to correct this one must touch the cameras and
encoder. A 500 N load cell monitored the resultant destroy the calibration. So, prior to camera place-
axial force (P). From the back side, an Electronic ment in the setup, we held white paper in front of
Instruments Research (model LE-05) LE measured the camera, outside of the focal range, and slowly
the elongation (δe ) between the laser tags affixed to a moved it around. The live camera images will
central gage section of initial length Le = 28.57 mm. appear as pure white if the lens and CCD protec-
The upper grip displacement δ̇ = dδ/dt was prescribed tive glass are clean. If, however, gray smudges exist
at a constant rate during loading (δ̇ > 0) and unload- that do not move as the paper is moved, then the
ing (δ̇ < 0). As we have mentioned before in previous lens should be cleaned. If the smudges remain and
parts, some disagreement between local and global they do not move when the lens is rotated inside
strain measurements is typically unavoidable due to its C-mount, then the source of the smudges is
some grip slippage as transformation fronts (localized on the protective glass of the CCD, and blowing
strains) enter or exit the free length. Artifacts from purified compressed air across the protective glass
grip slippage were (mostly) eliminated by reporting should correct the problem.
Laser Extensometer
LE
Top Grip
LE Tag
Specimen
IR camera
15° 15°
(a) (b)
Figure 5 Superelastic experiment setup: (a) photograph of front view and (b) schematic of top view.
IR DIC its scale is the right-hand vertical axis. The axial strain
Upper contour plot was created by averaging the strain across
Grip
the lateral (y) direction of the wire to create a 1-D axial
ε xx (%)
LE profile from each DIC image. A selected sequence at
T ( o C) Tag 10
1.6-s intervals was then stacked side by side (≈ 200
35
images per cycle) and synchronized with time (t)
A- M+ 8
along the horizontal axis. (3-D DIC provides a wealth
30 Front
of information, but here we are mostly interested in
6
axial strains. Other plots of other strain components,
25
such as lateral (εyy ) and shear (εxy ) strains, have
4
also been processed that show similar discontinuities
20
where fronts exists, but are not shown here in the
2
x interest of space.) The temperature contour plot was
15
similarly created from the IR images at 1.6-s intervals
0
(≈ 200 images per cycle). Both fields are shown in
the Eulerian frame (current configuration), so the
y
(a) (b) top of the free length moves up during loading and
down during unloading according to the upper grip
displacement.
Figure 8 Experiment 2, snapshots at t = 121 s of temperature (a) and
axial strain (b) fields during first-cycle loading (A → M+ propagation). During the first cycle, the sharp transition between
high strain and low strain regions in Fig. 10(a)
clearly shows transformation fronts (locations of
8 high strain gradient) traversing the specimen length.
ε As discussed in Part 4, these fronts also manifest
(%) themselves as local temperature changes in the
6 specimen, which can be seen in Fig. 10(b), due
to the exothermic (A → M+ during loading) and
endothermic (M+ → A during unloading) latent
4 heat (enthalpy) exchanges with the environment.
Because stagnant air is a relatively poor heat transfer
2 LE medium, A → M+ transformation creates self-heating
and fronts show up as local hot spots (above RT);
whereas, M+ → A transformation creates self-cooling
DIC
0 and fronts show up as local cold spots (subambient
0 0.1 0.2 0.3 0.4 temperatures). We have used this to our advantage
t (103 s) in past work to track the transformation kinetics
in SMAs. At this relatively slow loading rate, a
single A → M+ transformation front starts at the
Figure 9 Experiment 2, cycle 1 comparison of average strains as
top grip where there is a stress concentration due
measured by the laser extensometer (LE) and DIC.
to clamping, and it propagates downward at nearly
constant speed until x/L = 0.1 when the crosshead
hot spot and in the DIC image as a near disconti- motion is reversed. The A → M+ strain jump is
nuity in axial strain from 0.9 to 7.5% (about 2 wire about �εxx = 7.5 − 0.9 = 6.6%. During unloading,
diameters in axial extent). As a confirmatory step, this initial front remains static momentarily until the
we compared the average axial strains between the stress reaches the lower plateau, at which point it
LE tags as measured by the LE and DIC, which track reverses direction, propagating upward until a new
together nicely (Fig. 9). M+ → A front starts from the top grip. These two
Figure 10 shows contour plots of the axial strain fronts then propagate toward each other at about half
field history, εxx (x, t), and temperature field history, the initial speed until they coalesce (annihilate each
T(x, t), during the first five cycles. In each plot the other) at x/L = 0.67, at which point the end of the
wire axial coordinate (x) is normalized by the initial lower stress plateau is reached and further unloading
free length (L), as shown on the left-hand vertical occurs via a uniform strain field. The M+ → A strain
scale. The stress history is overlaid for reference, and jump is about �εxx = 0.3 − 6.0 = −5.7%.
1
N=1 2 3 4 5
Loading Unloading
P
1 0.8 A0
x
L (GPa)
P 0.6
A0 +
M-A
Front
0.4
+
A- M
Front 0.2
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
t (103 s)
xx (%)
(a)
0 2 4 6 8 10
1
N=1 2 3 4 5
P
1 0.8 A0
x
L (GPa)
P 0.6
A0
0.4
0.2
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
t (103 s)
T ( o C)
(b)
15 20 25 30 35
Figure 10 Experiment 2, cycles 1–5: (a) axial strain field history, (b) temperature field history.
The magnitude of the slope of each strain the local self-heating (or self-cooling) in the specimen
discontinuity (front) in Fig. 10(a) is related to its front and is energetically favorable.
speed (c). Recall from Part 4 that transformation fronts The axial strain field is useful to precisely locate
‘‘prefer’’ to propagate at the same speed according to fronts (compared to the temperature field that is
more diffuse due to axial heat conduction) and
δ̇
c= , (6) to quantify strain jumps, but the temperature field
n �εP is useful to explain the stress history, as was
where δ̇ is the elongation rate, n is the number of shown in Part 4. The thermo-mechanical coupling
front propagating, and �εP is the strain jump across in NiTi is clearly illustrated in Fig. 11, where the
the stress plateau. This is really just a statement of stress closely follows the trends of the temperatures
axial compatibility between the grips, written in rate TA→M+ and TM+ →A , consistent with the Clausius-
form, for steady-state (equilibrium) conditions. As Clapeyron relation, Eq. 1 in Ref. 4, for transformation
new fronts are generated (increasing n) they can stress versus temperature. These transformation front
slow down, since mechanical equilibrium requires temperatures were simply found from the maximum
the axial force to be constant along the length and and minimum temperatures in the gage section
they each contribute to the prescribed the prescribed during front propagation.
global elongation rate (δ̇). On the other hand, if the Returning to Fig. 10(a), one may notice three
strain jump (�εP ) decreases, as it does from cycle anomalies in the strain field history that exist
to cycle, they speed up. In the absence of boundary during all cycles. Two of the anomalies are real
effects or other inhomogeneities that would change disturbances in the strain field caused by the epoxy
the local temperature or stress conditions, all fronts tabs for the laser tags at x/L = 0.13 and 0.86. The
tend to propagate at the same rate, as this minimizes DIC strain measurements capture the fluctuations
0.6 only εavg = 7.45% (by DIC) in the second cycle due to
P superelastic shakedown, so the crosshead proceeded
P
0.5 A0 40 further during this cycle before δe /Le > 7.5%.
A0 T
The remainder of the cycles exhibited typical
(GPa) ( o C)
0.4 30 superelastic shakedown (and no further significant
TA M
+ grip slippage occurred). As the cycle number (N)
RT
0.3 20 increases, we see an increasing number of fronts
(Fig. 10(a)), probably due to accumulated defects in
TM + A the wire that reduce the energy barriers for strain
0.2 10
localization. One can see that the strains in the
‘‘austenite’’ (low strain) regions evolve to slightly
0.1
increased strains, consistent with the small amount of
residual strain accumulation (see again Fig. 7), which
0
0 0.1 0.2 0.3 0.4 results in a decrease in strain jump across the fronts.
t (103 s) With cycling the upper plateau moves downward in
stress and its length (in time, or equivalently, �εP )
Figure 11 Experiment 2, cycle 1 stress and front temperature histories. decreases. This means that transformation via front
propagation is exhausted at earlier global strains,
in strain caused by the constraint imposed by the consistent with the somewhat higher front speeds
epoxy mass and the slight stress concentration at according to Eq. 6. Further transformation involves a
these locations. The third anomaly consists of two nearly uniform strain field, and the stress rises steeply
slight horizontal streaks (discontinuities) at about to its (nearly) same value at maximum elongation.
x/L = 0.56. The bottom streak does not follow the This is the reason for the progressively larger stress
specimen displacement, while the upper streak does, ‘‘spikes’’ seen at the end of each loading increment.
which indicates a flaw in the CCD that affected both Propagating fronts during later cycles are also
the deformed images and the reference image. This accompanied by progressively less temperature devi-
was confirmed by changing lenses and observing the ations from RT than the first cycle, as shown in
same artifact on a new specimen. The bottom streak Fig. 10(b). The stress history exhibits a number of
was even in the exact same pixel location. Thus, the small jumps and drops, each of which (except for
streaks are an artifact of the CCD imperfection and the grip slippage problem at 500 s in cycle 2) can
should be disregarded. be associated (and explained) by a change in front
The transformation fronts in the second cycle kinetics, in which fronts are created, annihilated, or
propagate along similar paths as in the first cycle. change speed. By cycle 5 there are as many as four
The only significant difference is that the A → M+ A → M+ fronts (during loading) and three M+ → A
front reaches the bottom of the gage length, where, fronts (during unloading), and some fronts even start
as frequently happens, the lateral contraction at the and stop (see the unloading portion of cycle 5), appar-
front caused the wire to slip out of the bottom grip a ently due to inhomogeneous defect accumulation in
small amount. This slippage can be seen as a drop in the wire that affects the local thermodynamic driving
the stress history as the front reaches the lower grip. force for front propagation.
It also created a rigid body translation of the material By cycle 25, as shown in Fig. 12(a), there exist
points between t = 500 s and t = 535 s. Because as many as eight fronts during loading and four
sudden stress drops do occur for other (valid) reasons fronts during unloading, with some again propagat-
in SMA wire, it can often be difficult to determine ing in ‘‘fits and starts.’’ The fronts in the IR images
whether load discontinuities are due to phase (Fig. 12(b)) are now quite muted, although still dis-
transformation or experimental problems. Here, DIC cernible upon close inspection. This occurs for two
clearly identified this as an experimental problem, reasons: (1) the smaller strain jumps and residual
not the actual material response. The transformation strain accumulation indicate that local plasticity has
did not reach the bottom grip during the first cycle locked-in micro pockets of residual martensite, such
but did during the second cycle. The reason was that a smaller fraction of the material (within a given
that the load frame was programmed to reverse wire cross-section) actually participates in the trans-
the crosshead direction when δe /Le > 7.5% (not the formation to release or absorb latent heat, and (2)
global strain). The average strain in the M+ region more fronts during the transformation effectively dis-
reached εavg = 7.53% (by DIC) in the first cycle, but tribute the self-heating and self-cooling more evenly
1 10
P
1 0.8
x A0 ε xx 8
L (GPa) (%)
0.6
P 6
t
A0
0.4
4
0.2
2
0 0
7.6 7.7 7.8 7.9 8 0
t (103 s) 0 0.2 0.4 0.6 0.8 1
(a) xL
ε xx (%)
10
(a) 0 2 4 6 8 10
1
ε xx 8
P (%)
1 0.8 A0
x 6
L (GPa)
0.6
4
0.4
P 2
A0 0.2
0
0 0
0 0.2 0.4 0.6 0.8 1
7.6 7.7 7.8 7.9 8
(b) xL
t (103 s)
T ( o C) Figure 13 Experiment 2, cycle 1 axial strain profiles during loading (a)
(b) 15 20 25 30 35 and unloading (b). Shown in the Lagrangian frame.
Figure 12 Experiment 2, cycle 25: (a) strain field history and (b)
Lastly, Fig. 14 shows a comparison of strain εxx
temperature field history.
profiles for cycles N = 1, 5, and 25 during loading at
a selected strain level (δe /Le = 4.5%) midway across
across the specimen. The persistence of only four the upper stress plateau. The three strain profiles cor-
fronts during unloading means the fronts must move respond to times t =117.4 s, 1462.2 s, and 7654.2 s,
relatively quickly to traverse the gage length, and respectively. One can again see the number of fronts
self-cooling still decreases the specimen to as low as (strain jumps) increases and the magnitude of the
18.4◦ C. This causes the stress to steadily decrease dur- strain jumps decreases with cycle number. Comparing
cycles 1 and 25, the strain jump across a A → M+
ing M+ → A transformation (unloading); whereas,
transformation front during loading decreases from
the stress remains relatively constant during A → M+
about �εxx = 6.6% to about 3.5% (= 6.1 − 2.6%),
transformation (loading) where the latent heat was
and the strain jump across a M+ → A transformation
distributed among twice as many (eight) fronts.
front during unloading (not shown) changes from
The contour maps just shown give a good overview
−5.7% to about −3.5% (= 1.5–5.0%).
of the thermo-mechanical behavior, but quantita-
tive data can be somewhat difficult to see, so Fig. 13
shows selected strain profiles at 16-s intervals dur- Summary and Conclusions
ing the first load-unload cycle, all mapped back
to the Lagrangian (reference) frame. The dashed Digital image correlation (DIC) is a modern and
arrows show the progression of profiles in time. (Note powerful experimental technique to characterize
that, as previously discussed, the sharp oscillations at the evolution of deformation in structures. It is a
x/L = 0.13 and 0.86 are due to the LE tags and the technique that is becoming increasingly common
oscillation at x/L = 0.56 is the artifact due to the CCD for macroscale investigations in testing labs, and
imperfection.) its use with optical digital cameras is now well
10 References
14. Helm, J.D., McNeill, S.R., and Sutton, M.A., Deformation Measurement of Surface and Internal
‘‘Improved Three-dimensional Image Correlation for Planes via Digital Image Correlation,’’ Experimental
Surface Displacement Measurement,” Optical Mechanics 47(1):51–62 (2007).
Engineering 35:1911–1920 (1996). 26. Churchill, C.B., and Shaw, J.A.,
15. Bay, B., Smith, T., Fyhrie, D., and Saad, M., ‘‘Digital ‘‘Thermo-electro-mechanical Shakedown Response
Volume Correlation: Three-dimensional Strain of Conditioned Shape Memory Alloy Wires,”
Mapping Using X-ray Tomography,” Experimental Proceedings of the ASME 2009 Conference on Smart
Mechanics 39:217–226 (1999). DOI: Materials, Adaptive Structures and Intelligent
10.1007/BF02323555. Systems (SMASIS 2009), ASME, Volume 1, pp.
16. Franck, C., Hong, S., Maskarinec, S., Tirrell, D., and 137–148, (2009).
Ravichandran, G., ‘‘Three-dimensional Full-field 27. National Instruments (www.dasylab.com), Data
Measurements of Large Deformations in Soft Acquisition System Laboratory (DASY-Lab) Manual,
Materials Using Confocal Microscopy and Digital DASY-Lab, Austin, TX.
Volume Correlation,” Experimental Mechanics 28. Tong, W. ‘‘A Users Guide to the Yale Surface
47:427–438 (2007). DOI: Deformation Mapping Program (SDMAP),”
10.1007/s11340-007-9037-9. Technical Report, Department of Mechanical
17. Sutton, M.A., McNeill, S.R., Helm, J.D., and Chao, Engineering, Yale University, New Haven, CT
Y.J., ‘‘Advances in Two-dimensional and (2004).
Three-dimensional Computer Vision,” 29. Tong, W. ‘‘An Evaluation of Digital Image
Photomechanics, Topics Applied Physics, Volume 77, Correlation Criteria for Strain Mapping
Springer Verlag, pp. 323–372 (2000). Applications,” Strain 41(4):167–175 (2005).
18. Schreier, H.W., Garcia, D., and Sutton M.A. 30. Sutton, M.A., McNeill, S.R., Jang, J., and Babai, M.,
‘‘Advances in Light Microscope Stereo Vision,” ‘‘Effect of Subpixel Image Restoration on Digital
Experimental Mechanics 44:278–288 (2004). Correlation Error Estimates,” Optical Engineering
19. Schreier, H.W., Braasch, J.R., and Sutton, M.A., 1988:870–877 (1988).
‘‘Systematic Errors in Digital Image Correlation 31. Bruck, H., McNeill, S. Sutton, M., and Peters, W.,
Caused by Intensity Interpolation,” Optical ‘‘Digital Image Correlation Using Newton-Raphson
Engineering 39(11):2915–2921 (2000). Method of Partial Differential Correction,”
20. Sutton, M.A., Yan, J.H., Tiwari, V. Schreier, H.W., Experimental Mechanics 29(3):261–267 (1989).
and Orteu, J.J. ‘‘The Effect of Out-of-plane Motion 32. Vendroux, G., and Knauss, W.G., ‘‘Submicron
on 2D and 3D Digital Image Correlation Deformation Field Measurements: Part 2. Improved
Measurements,” Optics and Lasers in Engineering Digital Image Correlation,” Experimental Mechanics
46(10):746–757 (2008). 38:86–92 (1998).
21. Zavattieri, P.D., Savic, V. Hector, L.G., Jr., Fekete, 33. Iadicola, M.A., and Shaw, J.A., ‘‘The Effect Of
J.R., Tong, W., and Xuan, Y., ‘‘Spatio-temporal Uniaxial Cyclic Deformation on the Evolution of
Characteristics of the Portevin–Le Châtelier Effect in Phase Transformation Fronts in Pseudoelastic NiTi
Austenitic Steel with Twinning Induced Plasticity,’’ Wire,” Journal of Intelligent Material Systems and
International Journal of Plasticity 25:2298–2330 Structures 13(2):143–156 (2002).
(2009). 34. Sun, H., Pathak, A., Luntz, J., Brei, D., Alexander,
22. Knauss, W.G., Chasiotis, I., and Huang, Y., P.W., and Johnson, N.L., ‘‘Stabilizing Shape
‘‘Mechanical Measurements at the Micron and Memory Alloy Actuator Performance Through
Nanometer Scales,” Mechanics of Materials Cyclic Shakedown: An Empirical Study,” Porter
35:217–231 (2003). Davis, L., Henderson, B.K., and McMickell, M.B.,
23. Chu, T.C., Ranson, W.F., and Sutton, M.A. (eds), Proceedings of SPIE, Industrial and Commercial
‘‘Applications of Digital-image-correlation Applications of Smart Structures Technologies 2008,
Techniques to Experimental Mechanics,” Volume 6930, SPIE, pp. 69300Q–(2008).
Experimental Mechanics 25(3):232–244 (1985). 35. Correlated Solutions, Inc.
24. Reu, P.L., and Miller, T.J., ‘‘The Application of (www.correlatedsolutions.com), Vic-Snap Manual,
High-speed Digital Image Correlation,” The Journal of Correlated Solutions, Inc., Columbia, SC.
Strain Analysis for Engineering Design 43(8):673–688 36. Correlated Solutions, Inc.
(2008). (www.correlatedsolutions.com), Vic-3D Manual,
25. Berfield, T., Patel, J., Shimmin, R., Braun, P., Correlated Solutions, Inc., Columbia, SC.
Lambros, J., and Sottos, N., ‘‘Micro- and Nanoscale