Case Study

Curvature Monitoring of Beams Using

Digital Image Correlation
Michael Dutton1; W. Andy Take2; and Neil A. Hoult, M.ASCE3
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

Abstract: A method for measuring longitudinal strains with the height at a section, and thus the curvature, using a technique based on digital
image correlation (DIC), is presented. The background to this technique is introduced as well as previous work in this area. The accuracy of DIC
under ideal conditions is established using artificially generated images that represent beams with various curvatures. The practical accuracy of
DIC is established by comparing the strains measured using DIC to those predicted by elastic theory and measured using strain gauges for a steel
beam. The correlation between these results is found to be excellent. DIC is then used to measure curvatures in RC beams and these results are
compared with analytically predicted results with good agreement. The choice of an appropriate gauge length for RC is discussed and is shown
to be one of the significant advantages of using DIC as opposed to strain gauges in both laboratory testing and field monitoring of bridge struc-
tures. DOI: 10.1061/(ASCE)BE.1943-5592.0000538. © 2013 American Society of Civil Engineers.
Author keywords: Strain gauges; Imaging techniques; Monitoring; Tests; Reinforced concrete; Steel; Instrumentation.

Introduction Recent advances in the development of a technique known as

digital image correlation (DIC) have meant that this technique can
The ability to accurately measure strain has been essential in the potentially be used as an alternative and noncontact method of
development of structural engineering theories used in the design obtaining strain measurements. Specifically, a software program
and analysis of bridges, and this is a key element in many bridge called GeoPIV, an implementation of DIC originally focused on
monitoring programs. A variety of proven sensors are in use today displacement measurement in geotechnical applications (White
including electrical resistance and vibrating wire strain gauges, and Take 2002; White et al. 2003), has been further enhanced by the
which provide both accurate and repeatable measurements. Un- research team to offer microstrain-level accuracy (Lee et al. 2012).
fortunately, there are several disadvantages associated with these The measurement technique is capable of computing the dis-
devices. Most notably, they only provide point readings that are placement of a practically unlimited number of areas of interest
typically uniaxial, although strain rosettes are also available. This within a series of digital images taken during an experiment, load
results in multiple gauges being required to adequately measure the test, or field monitoring exercise. Because the captured images are
variations of strain over the material’s surface, and even in this case postprocessed, this leads to the significant advantage of being able to
areas of critical concern may not be monitored accurately because of take surface displacement and strain measurements where desired
localized behavior. The necessity of having to use several gauges can with a priori knowledge.
quickly add up for complex structural elements and limit the in- The relative movement between a pair of targets in the digital
formation available to the structural engineer as a result of data images can be used as a virtual strain gauge, which can then be
acquisition constraints. When used as a tool to measure long-term arranged to measure the horizontal strain over the height of a bridge
strain variations during field monitoring, foil gauges often suffer component such as a beam. The strain profile can then be used to
from stability issues such as drift. Vibrating wire strain gauges measure curvature. Thusyanthan et al. (2007) used this approach to
overcome this issue but are more expensive. Finally, both foil and observe the strain profile in clay beams; however, the strain in the
vibrating wire gauges need to be bonded to the material’s surface, clay was quite large in comparison with strains generally experi-
which is labor intensive because the surface must be carefully enced in structural engineering materials such as steel and RC. The
cleaned and prepared. Gauges bonded to the surface of materials are strain range of interest for these stiffer materials is often less than
also influenced by localized behavior such as cracking, which can 100 mstrain in field monitoring applications and would require
result in misleading data or debonding of the gauge. a high level of displacement measurement accuracy. However, the
implementation of new subpixel interpolation techniques (e.g., Lee
et al. 2012) could potentially make the DIC technique applicable to
structural materials and the current research explores how to opti-
1
Researcher, Dept. of Civil Engineering, Queen’s Univ., Kingston, ON, mize the use of DIC to measure strains in steel and RC beams. It
Canada K7L 3N6. E-mail: m.dutton@gmail.com should be noted that the current research presents the use of DIC in
2
Associate Professor, Dept. of Civil Engineering, Queen’s Univ., Ellis the laboratory and that there are additional challenges associated
Hall, Kingston, ON, Canada K7L 3N6. E-mail: andy.take@civil.queensu.ca
3
with field monitoring (e.g., the effect of temperature and vibrations)
Assistant Professor, Dept. of Civil Engineering, Queen’s Univ., Ellis that go beyond the scope of this paper.
Hall, Kingston, ON, Canada K7L 3N6 (corresponding author). E-mail: neil
The objective of this work is to build upon previous research
.hoult@gmail.com
Note. This manuscript was submitted on December 3, 2012; approved on using DIC to measure strains in tension members (Hoult et al.
July 22, 2013; published online on July 24, 2013. Discussion period open 2013) by investigating the applicability of the DIC technique to the
until May 3, 2014; separate discussions must be submitted for individual measurement of curvature for members in flexure for the following
papers. This paper is part of the Journal of Bridge Engineering, © ASCE, three scenarios: (1) where known imposed curvature has been ap-
ISSN 1084-0702/05013001(13)/$25.00. plied (using artificially generated images) to isolate and measure the

© ASCE 05013001-1 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 accuracy of the DIC technique in the absence of all other errors; (2) 2013) to determine the strain error and the image correction factors,
where DIC is used to measure the curvature of a linear elastic beam if necessary, to compensate for the effects of lens distortion. The
in the laboratory to assess the level of practically achievable accuracy effect of lens distortion is also a function of the total movement of
[i.e., a steel hollow structural section (HSS) beam loaded in three- the object being monitored relative to the lens; therefore, for small
point bending]; and (3) where the strain field as a result of flexure is displacements the distortion has a smaller effect (Pan et al. 2013).
highly complex to investigate if this technique could be used for Thus, it is important to determine the extent of movement expected
materials that experience cracking and a localized highly non- during laboratory or field monitoring experiments such that the
uniform strain field (i.e., a RC beam loaded in four-point bending). potential strain error caused by lens distortion can also be deter-
mined. In this study, in-plane rigid-body distortion tests were used to
determine the potential strain error caused by lens distortion during
Background the experiments.
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

During a DIC analysis, observed movement is assumed to be

acting in a single plane parallel to the lens of the camera and any out-
Digital Image Correlation
of-plane movement relative to this plane (either toward or away from
In the DIC method, digital images of a zone of interest are captured at the camera) is assumed to be zero. Thus, any actual movement out of
various deformation states and postprocessed by tracking a collec- plane appears in the image as a uniform radial strain. This error can
tion of smaller areas known as subsets. The ability to accurately be reduced by increasing the distance between the camera and the
track the subsets after the fact is dependent on both the subpixel object and by using a larger focal length lens as discussed in greater
interpolation scheme of the DIC search algorithm and the texture/ detail elsewhere (Hoult et al. 2013). The angle between the camera
uniqueness of the subset (Pan et al. 2009; White et al. 2003). The and the object being measured has a similar impact to out-of-plane
current research uses the image analysis program GeoPIV to per- displacement where parts of the object appear to be getting farther
form the DIC analysis (White et al. 2003). As noted previously, away or closer than others (although in this case the effects are not
recent advances in GeoPIV have increased the accuracy and pre- perfectly radial). Ideally, the plane of the lens and the object being
cision of its subpixel interpolation scheme to the point where measured should be parallel, and if they are not this results in
microstrain accuracy is achievable (Lee et al. 2012). The interpolation a measured strain differential that is not actually present. The effects
stage is critical because it allows the locations of the subsets to be of this can be minimized using similar techniques to those discussed
determined to a fraction of a pixel from the initial discrete locations. for out-of-plane movement (Hoult et al. 2013).
This step can be computationally intensive depending on the level of
accuracy required. In most structural engineering applications the
Previous Research
expected displacement is relatively small, and as such the required
pixel displacement accuracy has to be high. Therefore, it is critical to Sutton et al. (1983) performed experiments using digitized images
maximize the accuracy of the interpolation scheme, which can be done recorded optically using a video camera of a cantilever beam and
by using various interpolation functions as discussed elsewhere (Lee used numerical algorithms to validate the feasibility of what came to
et al. 2012). Depending on the interpolation scheme chosen, there are be known as two-dimensional DIC. Peters et al. (1983) demonstrated
small inherent errors associated with subpixel interpolation known as its applicability to estimating rigid-body motions, such as in-plane
bias errors. Using the same interpolation scheme as in this analysis, translations and rotations, while Chu et al. (1985) used DIC to
when measuring the location of a single subset, the bias error is on the measure rotations and strains. Researchers such as Smith et al.
order of magnitude of 0.001 pixels, which has been shown to impact (1998) began using DIC to measure strains; however, they reported
strain measurement accuracy for gauge lengths of less than 1,000 relatively high standard deviations of 100 mstrain. Wattrisse et al.
pixels (Lee et al. 2012). The selection of gauge length is also important (2001) also struggled to obtain the type of strain accuracies required
for physical strain gauges and their application to the material’s for structural monitoring with mean strain errors of 210 mstrain
surface. Strain readings are best taken over an area such that local using both artificial and experimental images. Hung and Voloshin
variations, such as aggregates and hardened paste in concrete, are (2003) measured uniaxial strain during tests on an aluminum plate
averaged together. over a range from 2,900 to 46,000 mstrain and obtained average
Apart from the inherent error in the DIC analysis, several other errors of 5.8%. However, as noted previously, recent improvements
sources of error are significant when working in the laboratory or in both the camera resolution and subpixel measurement algorithms
field environment. These include, but are not limited to, lighting used in GeoPIV have increased the potentially accuracy of DIC to
conditions, camera jitter, lens quality, out-of-plane movement, and microstrain accuracy (Lee et al. 2012).
the angle between the camera and the object being measured. In the structural engineering community, DIC has been primarily
Fluctuations in the lighting of the specimen will alter the appearance used to measure steel or concrete beam deflections as a noncontact
of the image texture, which may hinder the tracking of the subset in alternative to traditional linear variable displacement transducers.
the DIC analysis (Raffel et al. 2007). To reduce changes in the Destrebecq et al. (2010) investigated the behavior of a full-scale RC
natural light, artificial lights can be used; however, fluctuations in beam after 25 years in service using the DIC method. The midspan
alternating current cause subtle variations that can be seen in the deflection profile, as measured by DIC, was used to calculate the
brightness level at fast shutter speeds, and if this is the case a more curvature of the beam and was shown to be bounded by theoretical
stable light source may be required. Camera jitter refers to the fact predictions. Küntz et al. (2006) and Yoneyama et al. (2007) showed
that digital cameras cannot take an identical image twice (Luo et al. that DIC could be used as a monitoring tool to measure the deflec-
2001). tion of RC and steel girder bridges, respectively. DIC has also been
The quality of the lens can influence the level of distortion of used to monitor flexural crack growth and width in RC beams
a captured image (i.e., the image is not an exact representation of the (Lecompte et al. 2006; Barazzetti and Scaioni 2010).
actual structure). In turn, the measurement of the actual movement of As noted previously, determining the movement, and thus the
the specimen between images can be affected by lens distortion as strain, between two subsets is a straightforward calculation; how-
noted by Pan et al. (2013). One solution to this problem is to use an ever, it can lead to significant errors because of noise amplification
in-plane rigid-body distortion test (Yoneyama et al. 2006; Pan et al. (Pan et al. 2009). Instead, the computed displacement fields could be

© ASCE 05013001-2 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 smoothed before finding strain by using the penalty FEM as pro- resulted in accurate results as will be discussed subsequently. Using
posed by Sutton et al. (1991). While shown to work for uniform and a pair of subsets spaced at a known original distance and finding the
nonuniform displacement fields, the technique is computationally relative linear displacement between them will return the change in
intensive. An alternative is to use pointwise local least-squares strain as shown in the following equation:
fitting, which Wattrisse et al. (2001) used to study strain localiza-
tion in thin steel plates in tension. Dlength
ɛ¼ (1)
However, it is also possible to use the unfiltered computed dis- gauge length
placement field to measure strain with similar accuracy to a con-
ventional strain gauge by using a sufficiently large gauge length, Placing two columns of subsets on an image will then allow for
which is made possible by capturing high-resolution images (Lee the horizontal strain profile to be found. A straight line can be fitted
et al. 2012). To help further reduce the noise in strain measurements, to this profile to determine the curvature and the location of the
neutral axis. Curvature in beam theory describes the amount that
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

Lee et al. (2012) and Hoult et al. (2013) introduced and applied
a strain-averaging technique using Mohr’s circles to find the prin- a member in flexure rotates around the neutral axis. To use this
cipal strains of a thin steel plate loaded in tension. The approach was technique, a series of 3,456 3 5,154-pixel images was created with
shown to match well with foil gauge results when out-of-plane a known curvature (with the neutral axis at the midheight of the
movement was taken into account. image) ranging from 1028 to 1025 pixels21 . Four columns of subsets
that corresponded to two gauge lengths (1,000 and 4,000 pixels)
were used in this analysis. The DIC technique was then used to
Curvature Measurement Technique calculate the strain profile and determine the curvature of the image
Artificially generated images can be used to validate the measure- at each stage. By knowing the true curvature—and hence the correct
ment of strain and the associated curvature using DIC by simulating strain profile—a horizontal strain error profile can be created. The
the case of pure curvature across the image’s width. The use of measured strain profile for the 1,000-pixel gauge length is shown in
artificial images has the advantage of confirming the calculation Fig. 2(a) and the profile for the 4,000-pixel gauge length is shown in
technique without the errors induced by the digital camera that Fig. 3(a).
impact the DIC technique, including light fluctuations, camera The strain profiles for both of the chosen gauge lengths were
jitter, and lens distortion as discussed previously, and without the straight lines, as would be expected in typical beam behavior where
errors as a result of suboptimal texture. Image texture in the artificial plane sections remain plane. Examining the strain error in Figs. 2(b)
images was generated by randomly placing thousands of white dots and 3(b) (i.e., the computed DIC strain minus the theoretical strain)
on a black background. The brightness of the pixels making up shows that DIC underpredicts the tensile strain yet overpredicts the
a single dot forms a Gaussian curve of a specified diameter. Shifting compression strain. These errors are the result of how curvature is
the peak of this curve (i.e., the center of the dot) can be done to measured using the DIC technique. Curvature is meant to be de-
subpixel accuracy and allows a series of deformed images to be termined at discrete planes along the beam; however, in this approach
made. An example of an artificial image with constant curvature/ the curvature is being measured between two planes separated by
moment applied is shown in Fig. 1. a nonzero distance. Taking the curvature over a finite distance creates
As was previously described, the DIC technique postprocesses two potential issues: (1) complications as a result of nonlinear cur-
a series of images to determine where a group of pixels, referred to as vature gradients between these two planes and (2) using linear
the subset, moves. The choice of subset size affects the accuracy. measurements to determine what is actually a change in arc length.
However, increasing the subset size beyond 32 3 32 pixels has The first issue can be avoided if the change in curvature is linear or
a diminishing effect on accuracy (White et al. 2003); as the subset minimized by ensuring the distance between measurement planes is
size increases, any accuracy that is gained from better correlation has sufficiently small. The second issue can lead to significant strain errors
to be weighed against the localized averaging of displacement at higher curvatures and longer gauge lengths; however, it is also worth
behavior that comes from using such large subsets. The most ex- noting that the total strain at these curvatures is also high, resulting in
treme example of this would be the use of a subset that had the same small percent error values. For example, at a curvature of 1 3 1026 , the
height as the beam, which would mean the curvature could not be strain error as a result of the linear estimation is 2 mstrains for the
measured at all. In the current research, a minimum of 36e64 1,000-pixel gauge length; however, that represents an error of just
3 64 pixel subsets was used over the height of the beam, which 0.16% for the 1,265 mstrains being measured. For the smaller gauge

Fig. 1. Artificial image showing (a) initial and (b) deformed shapes under 2 3 1024 pixel21 pure moment (the two rows of squares represent the size
and location of subsets tracked to calculate the profile of horizontal strain)

© ASCE 05013001-3 J. Bridge Eng.

J. Bridge Eng. 2014.19.

Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. (a) Calculated strain profile from DIC and (b) associated strain error profile for an artificial curvature image with a 1,000-pixel gauge length

Fig. 3. (a) Calculated strain profile from DIC and (b) associated strain error profile for an artificial curvature image with a 4,000-pixel gauge length

length, Fig. 2 shows increased noise in the error profile versus that normalized cross-correlation algorithm used in the DIC analyses to
presented in Fig. 3, which is expected because the strain error as a result track rotating patches. At higher rotations, this error becomes sig-
of the bias error is inversely proportional to the gauge length (Lee et al. nificant relative to the other inherent errors. Thus, this raises an
2012). The strain errors, which are larger for increased curvature, are in important question. Will this curvature threshold and the associated
the range of 1–4% as a percent of the total measured strain. error be significant when measuring curvatures in real beams? It
In Figs. 4(a–c), the imposed curvature is compared with the should be noted that the exact magnitudes of these curvature errors
measured curvature, the error in the curvature measurement, and the are unique to this analysis. Using a different image texture would
average strain error, respectively. The measured curvature appears to return a similar trend and values but not the exact results. The sudden
match quite well with the imposed curvature as evidenced by the increase in errors above a certain curvature is also seen in the average
close fit to the one-to-one line. The absolute difference between these strain error [Fig. 4(c)], which is found by taking the absolute mean
values [Fig. 4(b)] indicates that after curvatures of approximately of the residuals from a linear regression of the strain error profile.
1026 pixel21 the curvature error noticeably increases. Below this The gauge length does impact the precision of the profile, as seen
curvature threshold, the errors are relatively constant and are of the by a lower average error for the larger gauge length.
order expected because of bias errors in the DIC technique. Above Fig. 4 indicates that the error in the measured DIC curvature is
this threshold, the errors increase as a result of the ability of the defined in image space (pixel21 ) rather than in object space (mm21 ).

© ASCE 05013001-4 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 Therefore, the quality of object space measurements can be improved square 102 3 102 3 3:2 HSS with a total length of 1.2 m was
by acquiring images with a higher-resolution camera (i.e., more pixels tested in three-point bending as detailed in Fig. 5.
per a constant field of view size) or through a reduction in the field of The testing program for the steel beam experiment consisted of
view (i.e., more pixels per millimeter) although at the expense of the applying a central point load to the beam in load increments, cor-
measurement area. This can be illustrated by using a simple example. responding to 50 mɛ, up to a maximum of 500 mɛ in the maximum
Consider the curvature of a steel beam 100 mm deep in four-point fiber at the center of the field of view. This ensured that the beam
bending, captured in an image when its extreme fiber strain is remained in the linear elastic region and was undamaged for sub-
2,000 mɛ using a 1.8-megapixel camera (1,080 3 1,620 pixels) and sequent experiments. To monitor the strain and subsequently val-
with a field of view of 120 3 180 mm. At this scale factor of idate the DIC calculations, three foil gauges were applied to the
9 pixels=mm, Fig. 4 indicates that the curvature resolution expected opposite face of the HSS beam in the same area as the field of view of
in this setup in the absence of other sources of error would be the camera. The strain gauges were located on the top and bottom
27 3 1028 mm21 with a maximum strain error of 13:5 mɛ. Using an flanges and at midheight, which allowed a strain profile to be created.
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

18-megapixel camera image (3,456 3 5,184 pixels) and the same The acquisition of the digital images during the experiment was
field of view, the scale factor increases to 28:8 pixels=mm. This done with two Canon T2i digital cameras with 180-mm lenses that
improves the curvature resolution to 7:3 3 1028 mm21 with a max- were placed on tripods and triggered remotely, while artificial lights
imum strain error of 3:6 mɛ, which is similar to an electrical resistance were used to ensure more consistent lighting. Two cameras were
foil gauge. It should be reiterated that this calculated error is the upper used such that the behavior in both shear spans could be observed.
bound of accuracy because it does not include errors as a result of the The tripods were located approximately 1.6 m away from the beam’s
image capture process in a laboratory setting. face, corresponding to a field of view of the entire depth of the beam
as shown by the shaded region in Fig. 5. This distance resulted in an
Experimental Test Setup average spatial resolution of 0:036 mm=pixel. Furthermore, cen-
tering the image on the middle of the shear span avoided the dis-
To investigate the level of accuracy of the DIC curvature measure-
turbed regions of the beam. At each load stage, a series of 10 digital
ment technique achievable in a laboratory setting, an experimental
images were acquired. This was done to reduce the impact from
program was conducted for two different sections and materials,
camera jitter when the images at a given load stage are subsequently
i.e., a steel HSS and a series of RC beams. The details of the ex-
averaged together. This technique is acceptable as long as the
perimental setup for each of these materials are given subsequently.
specimen does not noticeably change over the 20-s period in which
the images were captured.
Steel Beam
The cost of the system can be broken down to the cost of a Canon
A steel HSS was chosen for the homogeneous nature of the ma- T2i digital camera (∼US$600), 180-mm lens (∼US$1,500), computer
terial, which is beneficial because it would subsequently aid (∼US$1,000, where one computer can be used to run and analyze the
in making simplifying assumptions during the DIC analysis. A data from multiple cameras), and a MATLAB license (a cost that varies

Fig. 4. Comparison of imposed curvature to (a) measured DIC curvature; (b) curvature error; (c) average strain error

Fig. 5. Layout and field of views for a steel HSS beam

© ASCE 05013001-5 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 Table 1. RC Beam Specimens
Designation Shear reinforcement Total span (m) Shear span (m) Failure load (kN) Failure mode
B1 Stirrups 3.8 1.4 132.0 Flexure
B4 None 3.4 1.2 114.7 Shear
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. RC beam details: (a) long shear span (B1 and B2); (b) short shear span (B3 and B4); (c) with shear reinforcement (B1 and B3); (d) without shear
reinforcement (B2 and B4); (e) cross section

depending on the end user). GeoPIV is not available commercially; Table 2. Material Properties of Concrete for Beam Specimens
however, similar to other DIC software packages, it is freely available Specimen
from various research groups. Although this results in an initial cost that
is higher than a single strain gauge, it is worth noting the following: the Specimen property B1 B4
data acquisition system for most strain gauges costs at least as much as Compressive strength (MPa) 33.7 35.5
the entire DIC system; a single camera can measure strains at a number Split tensile strength (MPa) 3.5 2.1
of locations simultaneously; and because the system is noncontact it can Age at testing (days) 36 265
be reused, unlike an electrical resistance strain gauge.

RC Specimens Table 3. Properties of Steel Reinforcement

A series of four RC concrete beams were constructed; two were Reinforcement
designed to have a ductile flexural failure (B1 and B3) and two were
Designation property 10M 20M
designed to fail in shear (B2 and B4). Furthermore, two different
2
shear spans were tested, giving a shear span to effective depth ratio Bar area (mm ) 100 300
of 4 for B1 and B2 and 3.4 for B3 and B4. The beam designations, as Young’s modulus (MPa) 200,000 200,000
well as their spans, failure loads, and failure modes are summarized Yield strength (MPa) 478 453
in Table 1. The cross sections of Specimens B1–B4 were 200 mm Ultimate strength (MPa) 576 563
wide by 400 mm deep and were longitudinally reinforced by 10M Strain hardening (mm=m) 7 7
(where M refers to metric bar designation and the number, e.g., 10, Rupture strain (mm=m) 195 185
is the nominal bar diameter in millimeters) top reinforcement and
20M bottom reinforcement as detailed in Fig. 6. These dimensions
and the 0.75% reinforcement ratio were selected to be representative
of typical beams used in RC construction. For the sake of brevity Experimental Results and Discussion
only the results from the tests on B1 and B4 will be discussed;
however, the results from B2 and B3 were comparable, demonstrating
Steel Beam
the repeatability of the technique. The concrete compressive and
tensile strengths from the concrete cylinders tests are reported in To find the curvature of the steel HSS beam, four columns of subsets,
Table 2 and the material properties of the steel are reported in Table 3. corresponding to two virtual gauge lengths, were centered on the
Two Canon T2i digital cameras with 180-mm lenses were placed middle of the shear span. Multiple other gauge lengths were ex-
on tripods, located at approximately 5.5 m from the beam face and amined; however, the results for just two different gauges lengths are
centered on the shear span. Two cameras were used such that the presented, although it is possible to use this technique to look at the
behavior in both shear spans could be observed. This resulted in an change in curvature along the beam. The layout of the subsets on the
average spatial resolution for the digital images of 0:133 mm=pixel. steel beam can be seen in Fig. 7.
Also, by having the cameras further away from the specimen, the The strain profiles for the 960- and 3,648-pixel gauge length
effects of out-of-plane movement (Hoult et al. 2013) were minimized virtual DIC strain gauges can be seen in Figs. 8(a and b), re-
compared with the steel specimen, as will be discussed subsequently. spectively. The dashed line represents a linear best fit for each of the
Ten images were taken at each load stage to be subsequently averaged selected load stages. The strain profile for the shorter gauge length is
together to create a single image, and additional images were taken noticeably noisier in comparison with the longer gauge length. This
between each load stage at a rate of one image every 10 s. is expected because the strain error is inversely proportional to the

© ASCE 05013001-6 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 gauge length as discussed previously. To explore the effects of lens a nonlinear strain profile with strains at the top and bottom of the
distortion, an in-plane rigid-body distortion test was conducted image appearing smaller or larger depending on the type of distor-
(Yoneyama et al. 2006; Pan et al. 2013) using the same camera and tion or the angle, which is not the case. The slope of the strain profile
lens setup that was used for all beam tests. At the maximum dis- is the curvature, which can be plotted against the applied load as
placement in the steel beam test, the subsets moved by approxi- shown in Fig. 9.
mately 10 pixels, and for that amount of movement the distortion The curvature, as measured by the foil gauges, was in good
tests suggest that the maximum expected strain error as a result of agreement with the theoretical linear elastic predictions assuming
lens distortion for a 1,000-pixel gauge length would be less than a modulus of elasticity of 200 GPa for the steel. The DIC curvature
10 mstrains. It can also be seen from Fig. 8 that lens distortion and showed a similar linear trend and slope; however, a constant offset
camera angle have had little impact on the results. If lens distortion or was observed between the measured curvature from the DIC and
camera angle were issues, it would be expected that displacement foil gauges. A possible cause of this would be an initial out-of-plane
movement of the beam relative to the camera before the first load
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

measurement errors caused by either effect would be manifested as

Fig. 7. Beam detail of a 102 3 102 3 3:2 steel HSS showing the location of the DIC subsets used for curvature calculation

Fig. 8. Horizontal strain profile (solid) and best-fit line (dashed) calculated using DIC with (a) 960-pixel gauge length and (b) 3,648-pixel gauge length

© ASCE 05013001-7 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 stage (Hoult et al. 2013). If the face of the beam moved toward the
camera, then an apparent radial tensile strain would be observed as
the beam would appear to be getting uniformly larger. This apparent
tensile strain can be seen by the offset of the neutral axis of the strain
profiles in Fig. 8.
To account for the out-of-plane movement, a second DIC anal-
ysis was conducted in which the averaged image from the first load
stage was used as the reference image. The adjusted strain profiles
for the selected gauge lengths are shown in Fig. 10 and the load
versus curvature relationship is shown in Fig. 11. Using the first
load stage as the new reference image corrects the apparent hori-
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

zontal shift in the neutral axis as seen for both gauge lengths in
Fig. 10. However, the location of the neutral axis is still not exactly at
midheight of the beam. This may be a result of the added weld
material located along the top flange, which would increase the top
flange area and lead to slightly lower strains at the top of the beam as
seen in Fig. 10. In addition, the correction results in a fairly good
match in the curvature predictions for both the DIC virtual gauge
lengths and foil gauges as seen in Fig. 11(a). The curvature error,
Fig. 9. Comparison of strains calculated using DIC and foil gauges during which is the difference between the DIC and theoretical curvature, is
the steel HSS beam test (the discrepancy is a result of apparent strains in the shown in Fig. 11(b). As would be expected, the foil gauge error
DIC analysis caused by out-of-plane motion during the first load step) compared with the linear elastic predictions was generally lower

Fig. 10. Horizontal strain profile and best-fit line for (a) 960-pixel gauge length and (b) 3,648-pixel gauge length for the out-of-plane adjusted DIC
analysis of the steel HSS beam test

Fig. 11. Applied load versus (a) measured curvature and (b) curvature error for the out-of-plane adjusted DIC analysis and strain gauges for the steel HSS beam test

© ASCE 05013001-8 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 than the DIC error but larger than would be estimated. For a 1-mɛ well below this value, significant curvature errors as a result of the
resolution, the associated curvature error for the steel beam would be effects of rotation on the DIC technique can be avoided, resulting in
approximately 2 3 1028 mm21 , whereas the curvature measured measurement accuracy similar to strain gauges. The DIC strain
using the strain gauge data were an order of magnitude greater. The averaging technique could be applied to other structural steel
possible sources of error in the foil gauge measurements were the members with the aim of detecting strain changes as a result of the
imperfections in the surface bond and signal noise. presence of holes or the effects of corrosion. These variations would
From the aforementioned example, the upper bound of the cur- be challenging to detect with traditional foil gauges because of the
vature error for an 18-megapixel image was approximately 7:3 sheer number of gauges that would be required.
3 1028 mm21 . In the laboratory setting, the DIC measured the
curvature with an accuracy of approximately 2 3 1027 mm21 , RC Specimens
similar to the error in the foil gauges. Furthermore, the maximum
observed curvature of approximately 3:6 3 1027 pixel21 (equiva- To determine the impact of cracking in a heterogeneous material on the
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

lent to 1025 mm21 in the object space) was less than 1026 pixel21 , proposed measurement technique, it was applied to RC beams. The
i.e., the curvature associated with significant measurement errors as layout of the four subset columns overlaid over an image of Specimen
seen in Fig. 4. As such, for typical levels of curvature observed in B1 at 120 kN can be seen in Fig. 12(a). The approximate location
steel structures, especially in the field where changes in curvature are where the cracks formed is also shown in Fig. 12(a) by black lines.

Fig. 12. Initial images: (a) B1 showing two virtual strain gauge lengths (3,200 and 1,280 pixels) and highlighted future crack locations; (b) B4 showing
two virtual strain gauge lengths (3,904 and 1,088 pixels) and highlighted future crack locations

© ASCE 05013001-9 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 For the strain profiles shown in Fig. 13, four load stages were registers only compressive strains even in the tensile region of the
selected, representing (1) the onset of cracking at a total force of 60 beam. This is a result of the fact that DIC measures surface strains
kN (shear force of 30 kN); (2) a total force of 75 kN (shear force of and once cracks form in the concrete the shrinkage strain that has
37.5 kN); (3) the strains after applying a total force of 75 kN (shear occurred over time releases and compressive strains are measured.
force of 37.5 kN) for 48 h; and (4) at a total force of 120 kN (shear Foil gauges, adhered to the specimen’s surface, can also be affected
force of 60 kN), which is close to the beam’s ultimate capacity. For by cracking, resulting in the values from the gauge being unusable
a steel beam, it would be expected that the strain profiles for each after it is intercepted by a crack or misleading if it is adhered between
gauge length would be close to identical (depending on the level of two cracks and measures similar compressive behavior. The larger
curvature as discussed previously) because they are centered on the gauge length [Fig. 13(b)], which avoids the cracks, averages the
same location of the beam; however, this is not true for a concrete strain over a larger distance and reduces the impact of the crack on
beam because of the presence of cracks and creep that create var- the measured strain. As with the steel HSS, the maximum curvature
is less than the 1026 pixel21 error threshold [as per Fig. 4(b)]. To
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

iations in the curvature. Although creep strains will not be discussed

in this paper, the changes in strain during the hold stage were not determine whether lens distortion represented a significant source of
insignificant and were of the order of 15% of the initial strain at the error in these tests, the results of the in-plane rigid-body distortion
start of the hold stage. The strain profiles for the smaller gauge length tests were once again used in conjunction with the average vertical
[Fig. 13(a)] are affected by the presence of cracking. It can be seen pixel displacement from the beam test of approximately 65 pixels to
that the shorter gauge length that was placed between two cracks determine the potential strain error. For these displacements, lens

Fig. 13. Horizontal strain profile for B1: (a) 1,280-pixel gauge length; (b) 3,200-pixel gauge length

Fig. 14. Applied load versus measured curvature comparison for concrete Specimen B1

© ASCE 05013001-10 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 distortion errors of approximately 50 mstrains are possible, which technique using the two different gauge lengths for Specimen B1.
correlates well with the predicted lens distortion error results for the The curvature, as found from the larger gauge length, matches well
steel beam test where a smaller vertical displacement resulted in with the numerical model as seen in Fig. 14. The smaller gauge
smaller potential errors. This error represents 5% of the maximum length is impacted by the cracks. Although the results from the
measured tensile strains and 10% of the maximum measured smaller gauge length show excellent agreement with the numerical
compressive strains. It is worth noting that the impact of lens dis- prediction and the longer gauge length measurement in the beam’s
tortion is dependent on the distance from the center of the lens linear elastic region, after the formation of the cracks at approx-
(Pan et al. 2013); therefore, strain errors will be smaller closer to the imately 30 kN the curvature measurements represent the local
center of the lens. strain field between the cracks and not the overall behavior of the
A numerical modeling program called Response-2000 (Bentz section.
2000) was used to predict the curvature in the RC beams at the For Specimen B4, the four subset columns are shown in Fig.
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

location of the DIC analysis. Response-2000 is based upon the 12(b) superimposed on the 110-kN image of the beam. In this case,
modified compression field theory (Vecchio and Collins 1986) and the shorter gauge length was placed such that one column intercepted
is a nonlinear sectional analysis program for RC beams and col- the developing shear crack (as shown by black straight lines for
umns (Bentz 2000). Fig. 14 shows a comparison of the predicted clarity). The strain profiles for the two gauge lengths can be seen in
curvature from Response-2000 versus that calculated by the DIC Fig. 15.

Fig. 15. Horizontal strain profile for B4: (a) 1,088-pixel gauge length; (b) 3,904-pixel gauge length

Fig. 16. Applied load versus measured curvature comparison for concrete Specimen B4

© ASCE 05013001-11 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 Similar to Specimen B1, the strain profiles for B4 illustrate that the curvature. For large gauge lengths, the calculated curvature
the placement of the virtual gauges is important in achieving the matched well with the curvature predicted by a RC analysis
correct strain profile. In the region in which the subsets intercept the program.
shear crack, it is possible to notice the difference between using Possible future work could include the use of the DIC technique on
a gauge length that measures strain with the cracks included (be- an in situ RC structure to monitor curvature changes during a load
tween 100 and 350 mm from the top of the beam) and with cracks test.
excluded (between 350 and 400 mm from the top of the beam). In the
region where the cracks are included, the tensile strains are higher Acknowledgments
than those measured with the long gauge length [Fig. 15(b)]. This is
expected because the crack width has a significant influence on the The authors acknowledge the Natural Science and Engineering Re-
strain measured using this shorter gauge length. In the region where search Council of Canada for their support of this research. The
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

the cracks are excluded, the same compressive strain behavior that authors also thank Danielle DeRosa, Adam Hoag, Keelin Scully,
was seen for Specimen B1 is observed. This compressive strain Neil Porter, and Paul Thrasher for their assistance with the experi-
behavior in what would be expected to be the tensile region of the mental program.
beam is once again because of the fact that concrete surface strains
are being measured. The surface strains become compressive as
References
a result of shrinkage in the concrete once the tensile stresses in the
surface concrete are released because of the formation of a crack.
Barazzetti, L., and Scaioni, M. (2010). “Development and implementation of
Fig. 16 gives the load versus curvature response showing both
image-based algorithms for measurement of deformations in material
DIC gauge lengths as well as the predictions from Response-2000 for testing.” Sensors, 10(8), 7469–7495.
B4. As with B1, the predictions from Response-2000 are generally in Bentz, E. C. (2000). “Sectional analysis of reinforced concrete members.”
good agreement with the DIC measurements; however, in this case Ph.D. dissertation, Univ. of Toronto, Toronto.
the shorter gauge length is in better agreement because the gauge Chu, T. C., Ranson, W. F., Sutton, M. A., and Peters, W. H. (1985).
length intercepts a crack rather than missing it entirely, leading to “Applications of digital-image correlation techniques to experimental
larger strains and higher curvatures. At 50 kN two sets of data for the mechanics.” Exp. Mech., 25(3), 232–244.
measured curvature can be seen, where the larger value (at ap- Destrebecq, J. F., Toussaint, E., and Ferrier, E. (2010). “Analysis of cracks
proximately 2:2 3 1026 mm21 ) was measured during the load cy- and deformations in a full scale reinforced concrete beam using a digital
cling of B4 (between 50 and 75 kN), and thus measures the increase image correlation technique.” Exp. Mech., 51(6), 879–890.
Hoult, N. A., Take, W. A., Lee, C., and Dutton, M. (2013). “Experimental
in curvature as a result of concrete creep during the 48-h period when
accuracy two dimensional strain measurements using digital image
the load was held constant. correlation.” Eng. Struct., 46, 718–726.
While the placement of the virtual strain gauges has been shown Hung, P. C., and Voloshin, A. S. (2003). “In-plane strain measurement by
to impact the strain profile in the presence of a crack, this can be digital image correlation.” J. Braz. Soc. Mech. Sci. Eng., 25(3), 215–221.
overcome by changing the gauge length and reanalyzing the digital Küntz, M., Jolin, M., Bastien, J., Perez, F., and Hild, F. (2006). “Digital
images. To ensure this is the case, it is recommended that the field of image correlation analysis of crack behavior in a reinforced concrete
view of the image is larger than one crack spacing. This versatility is beam during a load test.” Can. J. Civ. Eng., 33(11), 1418–1425.
not possible with physical strain gauges because they are applied to Lecompte, D., Vantomme, J., and Sol, H. (2006). “Crack detection in
the surface of the specimen before the experiment is conducted and a concrete beam using two different camera techniques.” Struct. Health
before cracks have developed, and this represents a significant Monit., 5(1), 59–68.
Lee, C., Take, W. A., and Hoult, N. A. (2012). “Optimum accuracy of two
benefit of the DIC technique.
dimensional strain measurements using digital image correlation.”
J. Comput. Civ. Eng., 10.1061/(ASCE)CP.1943-5487.0000182, 795–803.
Luo, G., Chutatape, O., and Fang, H. (2001). “Experimental study on
Conclusions and Recommendations nonuniformity of line jitter in CCD images.” Appl. Opt., 40(26), 4716–
4720.
From the presented research, the following four key conclusions can MATLAB [Computer software]. Natick, MA, Mathworks.
be drawn: Pan, B., Qian, K., Xie, H., and Asundi, A. (2009). “Two-dimensional digital
1. A technique using DIC to measure curvature from strain over image correlation for in-plane displacement and strain measurement: A
a selected gauge length has been presented. While the inherent review.” Meas. Sci. Technol., 20(6), 062001.
subset tracking error is minimized by increasing the gauge Pan, B., Yu, L., Wu, D., and Tang, L. (2013). “Systematic errors in two-
dimensional digital image correlation due to lens distortion.” Opt. Lasers
length, the slope of the strain profile remains constant.
Eng., 51(2), 140–147.
2. The measurement of strains as a result of curvature based on Peters, W. H., Ranson, W. F., Sutton, M. A., Chu, T. C., and Anderson, J.
a linear change in distance results in errors in the measured (1983). “Application of digital correlation methods to rigid body me-
curvature; however, the error has been shown to be low for chanics.” Opt. Eng., 22(6), 226738–742.
curvatures less than 1026 pixel21 , which is higher than any of Raffel, M., Willert, M., and Kompenhans, J. (2007). Particle image
the curvatures observed in the experiments. velocimetry: A practical guide, Springer, New York.
3. The calculated curvature has been shown to match well with Smith, B. W., Li, X., and Tong, W. (1998). “Error assessment for strain
both theoretical and foil gauge measurements for a steel HSS mapping by digital image correlation.” Exp. Tech., 22(4), 19–21.
beam; out-of-plane movement that occurred at the beginning Sutton, M. A., Turner, J. L., Bruck, H. A., and Chae, T. A. (1991). “Full-field
of the test was observed and corrected by changing the representation of discretely sampled surface deformation for displace-
ment and strain analysis.” Exp. Mech., 31(2), 168–177.
reference image to the first load stage image.
Sutton, M. A., Wolters, W. J., Peters, W. H., Ranson, W. F., and McNeill,
4. With the knowledge of where the cracks will develop, virtual S. R. (1983). “Determination of displacements using an improved digital
DIC strain gauges were used on RC beams to determine the correlation method.” Image Vision Comput., 1(3), 133–139.
strain profile. The effect of cracking on the DIC strain meas- Thusyanthan, N. I., Take, W. A., Madabhushi, S. P. G., and Bolton, M. D.
urements was observed and should be avoided by using (2007). “Crack initiation in clay observed in beam bending.” Geo-
a gauge length that averages the behavior when determining technique, 57(7), 581–594.

© ASCE 05013001-12 J. Bridge Eng.

J. Bridge Eng. 2014.19.

 Vecchio, F. J., and Collins, M. P. (1986). “The modified compression-field White, D. J., Take, W. A., and Bolton, M. D. (2003). “Soil deformation
theory for reinforced concrete elements subjected to shear.” ACI J., measurement using particle image velocimetry (PIV) and photogram-
83(2), 219–231. metry.” Geotechnique, 50(7), 619–631.
Wattrisse, B., Chrysochoos, A., Muracciole, J. M., and Némoz-Gaillard, M. Yoneyama, S., Kikuta, H., Kitagawa, A. and Kitamura, K. (2006). “Lens
(2001). “Analysis of strain localization during tensile tests by digital distortion correction for digital image correlation by measuring rigid
image correlation.” Exp. Mech., 41(1), 29–39. body displacement.” Opt. Eng., 45(2), 023602.
White, D. J., and Take, W. A. (2002). “GeoPIV: Particle image velocimetry Yoneyama, S., Kitagawa, A., Iwata, S., Tani, K., and Kituta, H. (2007).
(PIV) software for use in geotechnical testing.” Technical Rep., Dept. of “Bridge deflection measurement using digital image correlation.” Exp.
Engineering, Cambridge Univ., Cambridge, U.K. Tech., 31(1), 34–40.
Downloaded from ascelibrary.org by KANSAS STATE UNIV LIBRARIES on 07/10/14. Copyright ASCE. For personal use only; all rights reserved.

© ASCE 05013001-13 J. Bridge Eng.

J. Bridge Eng. 2014.19.