Anda di halaman 1dari 9

Engineering Structures 46 (2013) 718–726

Contents lists available at SciVerse ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Experimental accuracy of two dimensional strain measurements using


Digital Image Correlation
Neil A. Hoult a,⇑, W. Andy Take a, Chris Lee b, Michael Dutton a
a
Department of Civil Engineering, Queen’s University, Ellis Hall, Kingston, ON, Canada K7L 3N6
b
Faculty of Engineering and Applied Science, Queen’s University, Kingston, ON, Canada K7L 3N6

a r t i c l e i n f o a b s t r a c t

Article history: The measurement of strain has always been central to developments in structural engineering. As struc-
Received 2 August 2011 tures around the world reach the end of their intended service life, there is a need for more robust field
Revised 14 May 2012 monitoring techniques to determine, in conjunction with numerical models, if these structures are still fit
Accepted 16 August 2012
for purpose. Current strain gage technologies are only able to offer a limited number of discrete measure-
Available online 6 October 2012
ments, which often is not enough data to assess complex structures. This paper presents an alternative
technique for measuring strain using digital images called Digital Image Correlation (DIC) that allows
Keywords:
for the measurement of 2-D strain fields. The accuracy that can be achieved in physical experiments
Strain gages
Imaging techniques
where sources of error such as lighting and camera quality are present is investigated. A series of tension
Monitoring tests on steel plates are conducted allowing measurements from the DIC technique to be compared to
Tests measurements from conventional foil strain gages. Out-of-plane motion is identified as one of the most
Instrumentation significant potential sources of strain measurement error when using DIC. Five solutions for significantly
reducing the effects of out-of-plane motion are presented and three of these solutions are applied to the
steel plate experiment. Using these solutions it is possible to achieve mean strain errors of less than 5 le
when comparing DIC to foil strain gage measurements, which suggests that the DIC technique has the
potential to replace conventional strain gages. Areas of future research are also introduced.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction acquisition constraints. Foil gages have long-term stability issues


and so, whilst they can be used in the laboratory, they are not gen-
Strain measurements have been central to the development of erally appropriate for long-term field monitoring. Vibrating wire
structural engineering theory for decades. The ability to measure strain gages are much more stable but the cost of each gage makes
strain accurately and precisely will become increasingly more them prohibitively expensive for the type of pervasive monitoring
important for field monitoring as assets that were built during that will be required to validate complex structural assessments.
the infrastructure booms of the last century continue to deteriorate Finally, both foil and vibrating wire gages need to be bonded to
with age. Engineers will require detailed strain data taken in a vari- the structure, which can have a significant impact on the strain
ety of locations on these complex structures so that they can accu- measurements if the stiffness of the structural material is not sig-
rately assess their remaining service life and capacity. However, nificantly greater than the stiffness of the strain gage (e.g. [1,2]).
the two most commonly used methods for measuring strain, A technique known as Digital Image Correlation (DIC) repre-
vibrating wire and foil strain gages, have several significant disad- sents an alternative to conventional strain gages for measuring
vantages. First of all, they are only capable of providing linear point surface strains in that it overcomes many of the disadvantages out-
readings of strain, which means a significant number of gages are lined above. DIC compares two digital images (a reference image
required in order to measure strain distributions or fields. Also, and a deformed image) in order to determine how much move-
each one of these gages has to be wired separately, and so the ment has occurred between the two. This movement can then be
ultimate number of gages is often limited by wiring and data used to determine strains as described in Section 2. It should be
noted that since the technique requires the use of digital cameras,
it is not suitable for measuring internal strains or strains in
members that are not visually accessible. For the monitoring of
existing structures, these accessibility limitations often apply to
⇑ Corresponding author. Tel.: +1 416 533 3436; fax: +1 613 533 2128.
conventional strain gages as well, and so this technique could be
E-mail addresses: neil.hoult@gmail.com (N.A. Hoult), andy.take@civil.queensu.
ca (W. Andy Take), christopher.st.lee@gmail.com (C. Lee), m.dutton@gmail.com a potential replacement. Until now DIC has been used as an alter-
(M. Dutton). native strain measurement technique in controlled lab settings but

0141-0296/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engstruct.2012.08.018
N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726 719

has shown promise for field monitoring. However, in order to at- is used. Wattrisse et al. [9] obtained mean strain errors of 210 le,
tempt to use this technique for field monitoring, improved strain but similar to Smith et al. they used a lower resolution camera
measurement accuracy is required, which is the subject of this pa- (1317  1035 pixels) as well as a less accurate subpixel interpola-
per. Other issues that are field specific and that go beyond the tion scheme (a bicubic spline). Additionally, Wattrisse et al. used a
scope of this paper, such as varying light conditions and tempera- lens with a small focal length of 55 mm. The focal length of the lens
ture changes, still need to be addressed in greater detail, and are is an important factor in improving strain measurement error, with
the subject of ongoing research. larger focal lengths leading to smaller errors as will be discussed
The following paper examines the results of a series of tests later. Hung and Voloshin [10] performed a uniaxial test on an alu-
conducted on a steel plate in tension to determine what level of minum plate and obtained average errors of 5.8%. Unfortunately
strain resolution can practically be achieved. The test setup will the details of their experimental setup and subpixel interpolation
be introduced and the strain measurements from the first test will scheme were not presented and so the potential causes of this
be discussed. Those results indicate that the out-of-plane move- strain error relative to the proposed approach cannot be evaluated.
ment of the test specimen has a significant impact on the strain Whilst these results show promise and may be acceptable in some
measurements. An examination of the impact of out-of-plane applications, they are not comparable to what can be achieved by
movements will then be presented and five approaches to reducing conventional strain gages. Additionally, for field monitoring, where
the effects of out-of-plane motion will be discussed. Results from a structures are not taken up to the ultimate load and where the
second plate test will then be presented to illustrate the level of measured strains may well be less than 100 le, measurement
strain resolution that is possible in a lab setting once out-of-plane accuracy increases of one or even two orders of magnitude are re-
motion is compensated for. Further research challenges and con- quired over those previously reported in the literature.
siderations will then be discussed. The DIC method discretizes a digital image into smaller areas
known as subsets. The movement of these subsets in a series of dig-
ital images is then tracked between subsequent images. The strain
2. Background resolution that can be measured is a function of the software’s ability
to track the movement of these subsets with sub-pixel accuracy.
The DIC method has been used in the research community since The strain measurement resolution is a function of gage length
the early 1980s [3,4] and takes a number of different forms. The (the larger the denominator that the same subpixel measurement
technique that will be the focus of this paper is known as 2-D DIC is divided by, the smaller the percent error). However measurement
and involves the use of a single camera to measure displacements resolution is also affected by (i) the variation in surface texture of the
in a plane (e.g. [5]). This technique has the advantage that it only re- test specimen, (ii) fluctuating light conditions and (iii) the quality of
quires a single camera, thus minimizing the equipment costs and the camera (digital cameras cannot reproduce the same image of
allowing for additional measurement areas to be added without intensity values of a stationary scene perfectly between multiple
requiring these measurements areas to overlap. However, it has exposures, instead there is a slight jitter between sequential images
one significant disadvantage and that is all the displacements must [11]). The impact of texture on strain resolution is a function of the
occur within a single stationary measurement plane or else errone- variation in intensity of the texture, which in turn affects how un-
ous results will be generated. Researchers have overcome this limi- ique a particular subset is within the image [12]. The effect of texture
tation by introducing 2.5-D (e.g. [6]) and 3-D DIC systems (e.g. [7]) on strain resolution is a subject of ongoing research but the texture
that use multiple cameras to measure a 3-D displacement field. used in this study should provide close to the optimum strain reso-
These systems tend to be more expensive and specialized and are lution and was kept constant in all the experiments discussed. In or-
thus not widely available in the structural engineering research der to minimize the impact that (ii) and (iii) had on the strain
community. The goal of this research project was to modify a system measurements, two types of averaging were performed on the data.
that used readily available and relatively inexpensive digital single First of all, ten images were taken at each strain measurement step
lens reflex (DSLR) cameras with an analysis package that has been during the steel plate tests. These 10 images were then merged to-
used extensively in the geotechnical research community (GeoPIV, gether to form a single image using MATLAB, which created a series
[5]) for use in structural engineering research. The main challenge of average images on which the DIC method was used. The strain re-
of this research involves increasing the strain measurement accu- sults were also averaged together using an approach based on
racy of the system, which has previously been used to measure geo- Mohr’s circle of strain, which is discussed in Section 3.1, to further
technical strains that are typically several orders of magnitude reduce the impact of all the potential sources of error.
larger than structural strains.
The DIC method has previously been used by researchers to 3. Plate test set-up
measure axial strains for artificially and experimentally generated
images, with varying degrees of success. Smith et al. [8] reported a In order to determine the strain measurement accuracy that is
standard deviation in strain readings of 100 le. Though they used a possible with the proposed 2-D DIC approach, a uniaxial tension
concentric lens, which will minimize the effects of out-of-plane test was conducted. The test specimen was a steel plate measuring
distortion as discussed later, their camera resolution 101 mm wide by 9 mm thick. The plate was tested in a MTS
(640  480 pixels) as well as their bicubic spline subpixel interpo- hydraulic self reacting frame as illustrated in Fig. 1. The overall
lation scheme will have had a negative effect on their measure- length of the plate was 673 mm and the distance between the grips
ment accuracy. Displacement errors produced using the DIC of test frame was approximately 500 mm.
technique are generally fractions of a pixel and when measuring Foil strain gage rosettes were attached to both sides of the plate
strain, the effect of this error can be reduced by using a larger gage in the center at a height of approximately 408 mm from the bot-
length (in pixels) to produce a lower strain error. Thus higher res- tom of the plate (as illustrated in Fig. 2) and 320 mm above the
olution cameras (used in the current work) enhance measurement bottom grip when mounted in the MTS test frame. This ensured
accuracy by allowing for larger gage lengths. Smith et al. also used that the gages were not in a region that would be susceptible to
a bicubic spline to obtain their subpixel displacement measure- end effects. The use of strain gage rosettes allowed the shear strain
ments whereas in the current work an eight coefficient filter based to be determined using Eq. (1):
on the B-spline transformation, which Sutton et al. [7] present as
the optimal technique in terms of minimizing measurement error, c ¼ 2e45  e0  e90 ð1Þ
720 N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726

Camera 1 Camera 2
Steel
plate

Foil gages 500


(both sides) mm

Speckle
pattern
(both sides)

Fig. 1. Steel plate tension sample installed in testing frame to compare strain measurements from foil strain gage rosettes and digital image correlation using two cameras.

101 mm

Steel plate 2000 pixels (67.9 mm)


(9 mm thick)

265 mm
Multiaxial foil
strain gage
500

1000

133 mm 1500

2000

2500 32 x 32 pixel subsets

275 mm 500 1000 1500 2000 2500 3000 3500


Painted X (pixels)
region for
DIC analysis

Fig. 2. Steel plate test specimen.

where e0, e45 and e90 represent the strains at 0°, 45° and 90° on the each camera. It is worth noting that the MTS machine does not actu-
strain gage rosette, respectively. Once two points on the Mohr’s cir- ally hold the load perfectly constant but instead holds the load to
cle (e0, c/2 and e90, c/2) had been determined, it was then possible within a range of ±50 N. However, this load variation corresponds
to determine the principal strains using these results. These values to a strain variation of less than 1 le and was therefore not consid-
were then used to verify the accuracy of the DIC system. ered significant. The applied load and strain gage data were recorded
The digital images were acquired using two Canon EOS Rebel XTi at a set time interval of 10 Hz throughout the duration of the test.
cameras with 10.1 megapixel resolution. The cameras were placed The plate specimens were loaded to a total load of approximately
on tripods on either side of the MTS test frame as illustrated in Fig. 1. 190 kN, which corresponded to a strain of 1000 le. Because the
As discussed earlier, a texture was applied to both sides of the plate maximum load was below the yield load for the plate (272 kN),
as illustrated in Fig. 2. The region with texture on the plate was located the same test specimen could be used for multiple tests.
at a distance greater than 100 mm away from the grips of the testing
machine to avoid end affects interfering with the DIC measurements. 3.1. Initial strain measurements
The plate test was run in load stages of 19 kN, which corre-
sponded to a change of approximately 100 le per load stage. At each The strains for each plate test were calculated using the DIC
stage, the load was held constant while 10 pictures were taken with method and an approach based on Mohr’s circle of strain with a
N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726 721

subset layout as illustrated in Fig. 2. By using three pairs of patches 100


(oriented at 0°, 45° and 90° from each other), the axial and shear
strains can be calculated for these pairs of points as you would εv
for a strain gauge rosette (using Eq. (1)). This calculation can then
be repeated for every group of three patch pairs seen in Fig. 2. By 50
ε
h
further realizing that all of these points (each set of axial and shear
strains) must lie on the Mohr’s circle of strain, this can be used as
an averaging technique to obtain more accurate strain measure-
0
ments by determining the circle of best fit through the points (as
seen in Fig. 3b). It can be seen that the diameter of the measure-
εv
ment circle is 2000 pixels and that 32 by 32 pixel subsets were
tracked on both sides of the steel plate using the two camera setup -50
shown in Fig. 1. The measured load–deformation response of the ε
h
steel plate captured using foil gages and DIC image analysis are Camera 1
presented in Fig. 3a. The Mohr’s circle of strain acquired from both Camera 2
cameras and the strain gages are plotted in Fig. 3b for the maxi- -100
0 50 100 150 200
mum applied load of 187 kN.
Load (kN)
It can be seen from Fig. 3 that though the general trend of the
foil strain gage behavior has been captured by the DIC method, Fig. 4. Difference in principal strain readings between foil gages and uncorrected
there is also a significant variation between the three Mohr’s cir- DIC image analysis results observed on both sides of the steel plate (foil gages on
cles as well as the load versus principal strain behavior. Interest- both sides of plate confirm no flexural strains).
ingly though, the diameters of all three Mohr’s circle in Fig. 3b
are approximately the same and the variation between the three
is in the location of the center of each circle. This suggests that circles measured by the DIC system are offset but have the correct
there is an apparent isotropic strain offset between the strains diameter. The diameter of the Mohr’s circle of strain is a function of
measured by the foil strain gages and the DIC system. This obser- the maximum measured shear strain. Thus, the fact that the diam-
vation is corroborated by Fig. 3a, where it can be seen that the eter of the circles is constant indicates that there is no variation in
DIC measurements for both principal strains can be fit by a linear the measured shear strain between the foil strain gages and the
relationship that has a different slope than that measured by the DIC system. The only way for this to be the case is if the variation
foil strain gages. The discrepancy between the DIC measurements in axial strain is the same in every direction radially around the
and the foil gages is investigated further in Fig. 4 where the error strain measurement circle. That is to say, every subset pair used
in principal strain is plotted against load stage. It is also worth not- to measure the strain moves closer together or farther apart by
ing that whilst the strain measured by one camera is greater than an equal amount (i.e. there is an isotropic change) in addition to
the strain measured using the foil gages, the strain measured by measuring the change in displacement due to the plate being
the other camera is less. The symmetry between the similar mag- loaded. There are two potential reasons for an isotropic expansion
nitude (but opposite sign) of the errors observed on both sides of or contraction of the plate. The first potential cause is temperature
the loaded steel plate suggests that there is a correction factor that, since significant variations in temperature would cause the plate to
if determined, would compensate for this variation. However, first expand or contract uniformly. However, temperature can be ruled
it is necessary to determine the cause of this variation, including out as it would also have been measured by the foil gages and it
why the cameras on opposite sides of the plate test setup give dif- would appear as either an expansion or contraction to both cam-
ferent results, before developing the required adjustment factors, eras (whereas in this case the measurements from one camera
which is the focus of the next section. indicated the plate expanded while the measurements from the
other camera indicated it contracted). The second potential cause
of a uniform change in strain would be out of plane movement
3.2. Out-of-plane movement of the plate either towards or away from the camera. As the spec-
imen gets closer to the camera, the subsets appear to get farther
In order to determine the cause of the varying strain data from apart from each other in the digital image, which is interpreted
the DIC system, it is first necessary to explore why the Mohr’s by the analysis software as a tensile strain. A similar effect can

200
600
εh εv
400
150
200

100 0 εh εv
-200
50 Camera 1 -400
Camera 2
(a) Foil gage -600 (b)
0
-500 0 500 1000 -500 0 500 1000
6 6
Principal strain, ε (x10 ) Normal strain, ε (x10 )

Fig. 3. (a) Measured load–deformation response of steel plate captured using foil and DIC strain gage rosettes, and (b) inconsistency in Mohr’s circle of strain for load 187 kN
measured using DIC image analysis on the two sides of the steel plate.
722 N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726

be observed as the specimen moves farther away, but is compres- that the horizontal viewing angle of the lens (as determined by the
sive in terms of the apparent isotropic strain. focal length) results in a physical gage length of X0 for the sample
To explore the effects of this out-of-plane movement, a series of in its original position at a distance y from the projective center.
experiments were conducted using the plate from the tensile tests For a given movement towards the camera, dy, a smaller physical
and the DSLR camera. The plate was held stationary and the cam- distance, X1, is captured within the same field of view. As a result,
era was moved toward the plate using a precision milling machine. the corresponding strain error, eerror, is a function of the gage
A picture was taken of the plate and then the camera was moved length, x. The gage length is in turn a function of the distance away
by 0.1 mm towards the plate and another picture was taken. This from the camera, y. This relationship has previously been derived
process was repeated until a total of 4 mm of camera movement by Sutton et al. [7] and is given in Eq. (2):
was reached.
Fig. 5a shows the resulting subset movement from the test. One dy
eerror ¼ ð2Þ
can see that the subsets move outward radially from the center of y
the circle during the 0.1 mm increment of movement towards to
The next section will explore methods for eliminating the strain
the camera. It should be emphasized that the plate was not
error based on this relationship.
strained during this test and it was left stationary as the camera
was moved towards it. Fig. 5b is a plot of the Mohr’s circle for three
different differential displacements from the plate movement test. 3.3. Out-of-plane movement corrections
The first important observation is that the circles are not actually
circles but a cluster of points around the center of what would There are several potential methods for correcting the mea-
be the circle. These Mohr’s circles with essentially zero radii and sured strain results for the error due to out-of-plane movement.
thus no apparent shear strain corroborate the earlier observation The first two, summarized by Sutton et al. [7], include method (i)
that the strain variations in the plate tests did not induce any addi- in which the out-of-plane strain error is calculated using Eq. (2)
tional shear strains. The second important observation is that the and direct measurements of the distance from the camera to the
plate movement creates a uniform shift in the axial strains. Both specimen and the change in that distance, and method (ii) in which
of these observations lead to the conclusion that the difference be- the magnitude of out-of-plane error is minimized by significantly
tween the DIC measured strains and the foil gage strains are caused increasing the effective distance between the camera and the spec-
by out-of-plane movement of the specimen. This conclusion is fur- imen, y. This could either be achieved by physically moving the
ther validated by the fact that the results from one camera sug- camera farther back from the test specimen or by using a telecen-
gested the plate was moving towards it while the results from tric lens in which the effective camera distance is significantly in-
the other camera suggested the plate was moving away. Having creased due to the special optics of the lens (i.e. the paths of light
identified the cause of this strain variation, if a relationship be- entering the lens are nearly parallel). In this paper, three additional
tween the amount of out-of-plane movement and the apparent potential methods for correcting out-of-plane strain error are pro-
strain can be found, this offset can be subtracted from the DIC mea- posed. These include method (iii) in which the out-of-plane strain
surements to correct for it. error is corrected using a knowledge of the material properties of
Fig. 6a is a plot of apparent axial strain or strain error, eerror, with the specimen to constrain the problem, method (iv) in which two
distance moved by the camera towards the plate. This relationship cameras (one placed on either side of the specimen) are used to
is plotted for a camera with a 55 mm focal length and three differ- measure the magnitude and direction of the out-of-plane strain
ent initial distances away from the plate. One can see that the error, and finally method (v) proposes that the out-of-plane strain
strain error is a function of both the initial distance away from error could be calculated if a fixed (unstrained) reference length is
the plate, y, and the distance moved towards the plate, dy, as the included within the image of the test specimen.
slopes of the three lines are different. In general, the closer the sub- Solution (i) is difficult to implement for the strain resolution re-
ject is to the camera, the larger the influence of the out-of-plane quired. To generate a strain error of 5 le (which as suggested is the
movement on apparent strain error. This relationship can be more required minimum accuracy for field monitoring) with a specimen
fully understood through the plotting of the ratio of camera move- to camera distance of 1000 mm, the out-of-plane movement calcu-
ment to initial distance, dy/y in Fig. 6b. This figure demonstrates lated using Eq. (2) would be just 0.005 mm. This small displace-
that this ratio is equal to the strain error. Fig. 7 illustrates why this ment is below the noise floor of most commonly used
is the case using a simple pin-hole model representation of the dig- displacement measurement instruments, such as linear resistance
ital image formation process. Using similar triangles it can be seen displacement transducers.

(a) (b)
2000
500
1000 Out of plane movement (mm)
1000 0.5 1.0 1.5 2.0 2.5
0
1500
-1000
2000
-2000
2500
500 1000 1500 2000 2500 3000 0 2000 4000 6000
X (pixels) 6
Apparent normal strain, ε (x10 )

Fig. 5. (a) Observed pattern of apparent deformations of steel plate during out of plane movement towards camera resulting in (b) linear accumulation of isotropic strain
error with out of plane movement.
N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726 723

10000 10000
(a) y=257 mm (b)
8000 336 mm 8000

6000 6000
405 mm

4000 4000

2000 2000

0 0
0 1 2 3 0 0.002 0.004 0.006 0.008 0.01
dy (mm) dy/y

Fig. 6. (a) Apparent isotropic strain due to movement of steel plate towards the camera, and (b) comparison of experimental data to simple pin-hole camera error solution.

Physical and pixel dimensions Position of steel plate


of CCD sensor, xCCD , uCCD,
respectively

f y

Viewing angle Viewing angle


of Camera 1 X1 X0
of Camera 2

dy
Out of plane
displacement, dy

Fig. 7. Simple pin hole camera model illustrating the effect of an increment of out of plane movement, dy, on the image formed by cameras viewing both sides of the steel
plate.

Solution (ii) can be accomplished in one of three ways: by plac- Therefore, a potential method of correcting the strains for the
ing the camera farther away from the specimen, by using a more out-of-plane movement can be calculated by rearranging Eq. (3)
advanced lens or a combination of the two. Placing the camera far- to solve for the out-of-plane strain error, eerror:
ther away can be problematic as though it reduces the error due to
out of plane movement it also reduces the strain measurement me1 þ e2
eerror ¼ ð4Þ
accuracy of the DIC method and thus trades one source of error ð1 þ mÞ
for another. One way to overcome this is to use a lens with a longer This equation indicates that a priori knowledge of the Poisson’s
focal length, which has the effect of reducing the field of view at ratio makes it possible to directly calculate the out-of-plane strain
larger distances, thus increasing the number of pixels over a given error, which can then be subtracted from the results for uniaxial
gauge length. An approach using this solution was used for the testing. This method is employed in the next section to correct
plate tests that will be discussed later. However the distance be- the strain data from the DIC method for a uniaxial test in which
tween the specimen and the camera can also be increased by using the specimen was subject to a small amount of out-of-plane move-
a telecentric lens, which uses optics to increase the effective cam- ment. It is worth noting that this method does not distinguish be-
era to specimen distance, y, in Eq. (2) without actually physically tween changes in strain due to out-of-plane movement and
moving the camera [7]. The downside of these lenses is their high changes due to temperature as they both appear as a uniform
cost (i.e. not conducive to the large number of monitoring locations change in strain. As such, it is important to record the change in
required in field applications) and their very limited field of view temperature if it will vary during the measurement period so that
(usually constrained to the diameter of the lens). Because of these thermal strain changes can be corrected for independently from
issues, the telecentric lens option was not explored further in this out-of-plane strain errors.
study. Solution (iv) uses two cameras placed on either side of the test
Solution (iii) takes advantage of the Mohr’s circle approach specimen to infer the strain within the specimen. For each camera,
proposed by the authors and of the Poisson’s ratio, m, of the the measured strain increment observed during a loading incre-
specimen. As noted previously, the principal strains, e1 and e2, ment, De1, will be a sum of the real strain increment, Dereal, and
are effectively offset by the out-of-plane strain error, eerror. As the error associated with out-of-plane distortion of each camera
a result, the ratio of the principal strains will no longer yield
the correct Poisson’s ratio. Instead, the true Poisson’s ratio, m, dy1 dy
of the specimen will be De1 ¼ Dereal þ ; De2 ¼ Dereal þ 2 ð5Þ
y1 y2
e2  eerror where the subscripts 1, and 2, relate to cameras 1 and 2,
m¼ ð3Þ
e1  eerror respectively.
724 N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726

Because the magnitude of the out-of-plane movement, dy, and The residual error in the principal strain error after the Poisson’s
the real strain increment of the sample, Dereal, is the same for both ratio correction had been applied is presented in Fig. 10a. The
cameras, Eq. (5) can be simplified to calculate dy, mean error in the vertical strain was +11.6 le, whereas the mean
error for the horizontal strain was observed to be 3.2 le.
De1  De2
dy ¼ ð6Þ A similar validation exercise was performed using solution (iv)
1=y1 þ 1=y2 for the test with the small amount of out-of-plane displacement
which can subsequently be used to correct for out-of-plane motion based on the benefit provided through the placement of two cam-
as per Eq. (2). This method is therefore similar to method (i), how- eras (one on each side of the test specimen) to correct for out-of-
ever it has the benefit of having a much easier mechanism in which plane motion. The residual error in the principal strain error after
to calculate the very small (but highly relevant) magnitude of out- the two camera correction had been applied is presented in
of-plane movement than attempting to measure it directly with dis- Fig. 10b. The mean error in the vertical strain was +8.0 le whereas
placement sensors. If both cameras are placed at an identical effec- the mean error for the horizontal strain was observed to be
tive distance from the specimen, y, the strain errors will be equal in 6.8 le. These values of residual error are very similar to that
magnitude but opposite in sign. achieved using the Poisson’s ratio correction method. It should
Finally, solution (v) would use an object of known dimensions be noted that the residual errors associated with the two camera
in the same plane as the test specimen. This object would not be technique had a slight increase in error with load level; however,
fully bonded to the test specimen and so would not strain under this technique has a considerable advantage in that it does not re-
load, but would move in the same plane. Thus any apparent quire a priori knowledge of the Poisson’s ratio.
changes in the object’s dimensions could be used to calculate the Whilst these results are adequate for lab testing in which spec-
amount of out-of-plane movement using Eq. (2). The practicality imens tend to undergo large strain ranges, these errors are poten-
or effectiveness of this approach has not yet been explored fully tially too large for field monitoring where strain changes may be
as it is a subject of ongoing research. As a result, this potential less than 100 le. In order to determine if, in the absence of out-
method for out-of-plane strain error is mentioned in this paper of-plane motion, it is possible to achieve better strain measure-
for the sake of completeness but is not discussed further here. ment accuracy, a second tension test was conducted where the
alignment of the testing machine was adjusted to produce minimal
out-of-plane movement. Although the adjustments ensured that
4. Adjusted strain measurements
the load was applied as concentrically as possible, it was not pos-
sible to achieve ideal concentric loading. As such, there was still a
In order to validate the proposed out-of-plane correction using
small amount of out of plane loading in this test that had to be cor-
the Poisson’s ratio approach (solution iii), the variation in apparent
rected for. The second plate test was conducted with a single Ca-
Poisson’s ratio was calculated for each load increment using both
non Digital Rebel XTi camera placed at a similar distance to the
the foil gage and DIC strain measurements for the test with a small
plate (1.08 m versus 1.576 m and 1.280 m for the cameras in the
amount of out-of-plane displacement. This data, included as
previous test) with a telephoto lens (Canon EF180f3.5L macro lens)
Fig. 8a, indicates that while the foil gage reported a constant Pois-
to keep the field of view nominally the same. The steel plate was
son’s ratio of 0.283 (consistent with the material properties of
then subjected to the same loading increments and the DIC and foil
steel), the strain values derived from DIC either greatly over or
gage strain results were determined. The error in principal strain
underreported the Poisson’s ratio due to strain errors caused by
between the DIC and foil strain gages is presented as a raw com-
out-of-plane motion. As expected from Eqs. (3) and (4), this caused
parison and after applying a Poisson’s ratio based correction in
the Mohr’s circle of strain to be horizontally shifted by a strain er-
Fig. 11a and b, respectively. One can see that once the out-of-plane
ror, eerror, as illustrated in Fig. 8b. To test if the magnitude of this
movement has been minimized, the mean principal strain error
error was well captured by the Poisson’s ratio correction, the steel
using the raw uncorrected DIC strain data was only 0.7 le and
plate test results were adjusted by subtracting the strain error gi-
4.5 le for vertical and horizontal strain, respectively. Further cor-
ven by Eq. (4). The post-correction results are given in Fig. 9, which
rection using the Poisson’s ratio based approach (assuming that a
shows both the load versus the principal strains as well as the
small amount of relative movement between the camera and the
Mohr’s circle at an axial load of 180 kN in which the Poisson’s ratio
specimen is unavoidable) further reduced these values to +3.9 le
is set at 0.283. It can be seen from Fig. 9 that the DIC method re-
and +0.1 le for vertical and horizontal strain, respectively. This
sults from both cameras are now in good agreement with the foil
implies that as long as effects such as out-of-plane motion are
strain gage results after employing the Poisson’s ratio correction.

1 ε error εerror
Camera 1 600
Camera 2
0.8 Foil gage 400

200
0.6
0 ε h)c εh εv
εv)c
) )
0.4
-200

0.2 -400

(a) -600 (b)


0
0 50 100 150 200 -500 0 500 1000
6
Load (kN) Normal strain, ε (x10 )

Fig. 8. (a) Apparent Poisson’s ratio calculated from DIC and strain gage readings, and (b) description of possible correction method based on a priori knowledge of Poisson’s
ratio.
N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726 725

200
600
ε εv
h
400 - εh
150
ε v =0.283
200

100 0 εh εv

-200
50 Camera 1 -400
Camera 2
(a) Foil gage -600 (b)
0
-500 0 500 1000 -500 0 500 1000
6
Principal strain, ε (x10 )
6 Normal strain, ε (x10 )

Fig. 9. (a) Corrected load–deformation response of steel plate observed using DIC using Poisson’s ratio = 0.283, and (b) corrected Mohr’s circle of strain for load 180 kN.

20 20
εv
εv
10 10

0 0

εh
-10 -10
εh
Camera 1
-20 (a) Camera 2 -20
(b)
0 50 100 150 200 0 50 100 150 200
Load (kN) Load (kN)

Fig. 10. Residual error in principal strain data after correction using (a) a priori knowledge of the Poisson’s ratio of the material and (b) two camera out of plane motion
approach.

20 (a) 20 (b)

10 10

εv
0 0

-10 εh -10

-20 Camera 1 -20

0 50 100 150 200 0 50 100 150 200


Load (kN) Load (kN)

Fig. 11. (a) Initial uncorrected error in DIC results of Test 2 and (b) residual error in principal strain data after Poisson’s ratio correction applied.

accounted for, the DIC technique can produce similar results to a in civil engineering applications (e.g. beams in bending and dis-
conventional strain gauge. turbed regions) and so the method will have to be further devel-
oped to measure these strains, especially for field monitoring
applications. The effect of temperature and lighting changes on this
5. Future research
technique as well as well dynamic strain measurements are all
ongoing topics of research.
The initial results using the DIC method suggest that it can pro-
vide similar results to strain gages. Further, as image processing
technologies and camera hardware improves, the accuracy possi- 6. Conclusions
ble with these techniques will continue to improve. Based on the
testing performed, it appears as if this technology has matured to The goal of this paper was to investigate whether the proposed
a point where the level of strain accuracy would allow the system DIC-based approach could be used to measure strain with the same
to be used for field monitoring. However, further research is re- resolution as conventional strain gages. A steel plate in tension test
quired to examine the impact of a number of variables including: was used to validate the method. It was found that even small out-
texture and variable strain fields. The importance of texture has of-plane movements of the specimen can result in significant strain
been discussed earlier. Variable strain fields are also quite common measurement errors. The effect of out-of-plane movement was
726 N.A. Hoult et al. / Engineering Structures 46 (2013) 718–726

examined and the strain error was found to be a function of the [2] Warren KA, Christopher B, Howard IL. Geosynthetic strain gage installation
procedures and alternative strain measurement methods for roadway
distance between the plate and the camera as well as the out-of-
applications. Geosynth Int 2010;17(6):403–30.
plane movement. Five potential methods for compensating for [3] Peters WH, Ranson WF. Digital imaging techniques in experimental stress
these strain errors were discussed. Three of these methods analysis. Opt Eng 1982;21(3):427–31.
(increasing the effective distance between the camera and the [4] Sutton MA, Wolters WJ, Peters WH, Ranson WF, McNeill SF. Determination of
displacements using an improved digital correlation method. Image Vis
specimen, an adjustment based on the Poisson’s ratio of the mate- Comput 1983;1(3):133–9.
rial, and a correction based on using cameras to view both sides of [5] White DJ, Take WA, Bolton MD. Soil deformation measurement using particle
the loaded specimen) were used to improve the accuracy of the image velocimetry (PIV) and photogrammetry. Géotechnique 2003;50(7):
619–31.
measurements from the plate test. It was found that the DIC [6] Brachman RWI, McLeod HA, Moore ID, Take WA. Three-dimensional ground
method can be used to provide strain resolutions that are adequate displacements from static pipe bursting in stiff clay. Can Geotech J
for field monitoring, provided that out-of-plane movement is 2010;47(4):439–50.
[7] Sutton MA, Orteu J-J, Schreier HW. Image correlation for shape, motion and
accounted for. deformation measurements. New York: Springer Science+Business Media;
2009.
Acknowledgement [8] Smith BW, Li X, Tong W. Error assessment for strain mapping by digital image
correlation. Exp Tech 1998;22(4):19–21.
[9] Wattrisse B, Chrysochoos A, Muracciole J-M, Némoz-Gaillard M. Analysis of
The authors would like to thank the Natural Science and Engi- strain localization during tensile tests by digital image correlation. Exp Mech
neering Research Council of Canada for their financial support of 2001;41(1):29–39.
[10] Hung PC, Voloshin AS. In-plane strain measurement by digital image
this research.
correlation. J Braz Soc Mech Sci Eng 2003;25(3):215–21.
[11] Luo G, Chutatape O, Fang H. Experimental study on nonuniformity of line jitter
References in CCD images. Appl Opt 2001;40(26):4716–20.
[12] Pan B, Lu Z, Xie H. Mean intensity gradient: an effective global parameter for
[1] Bathurst RJ, Allen TM, Walters DL. Short-term strain and deformation behavior quality assessment of the speckle patterns used in digital image correlation.
of geosynthetic walls at working stress conditions. Geosynth Int 2002;9(5– Opt Lasers Eng 2010;48(4):469–77.
6):451–82.

Anda mungkin juga menyukai