International Journal of Mining Science and Technology 29 (2019) 771–776

Contents lists available at ScienceDirect

International Journal of Mining Science and Technology

journal homepage: www.elsevier.com/locate/ijmst

Assessment of digital image correlation method in determining large

scale cemented rockfill strains
Bob A. Lingga, Derek B. Apel ⇑, Mohammadali Sepehri, Yuanyuan Pu
School of Mining and Petroleum Engineering, University of Alberta, Edmonton T6G 2W2, Canada

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

Article history: A conventional contact method (using linear transducers) and a non-contact method are deployed to
Received 8 December 2017 measure the axial and lateral deformations of large scale cylindrical cemented rockfill specimens.
Received in revised form 25 June 2018 Experimental works incorporating two pinhole cameras to create one stereovision by digital image cor-
Accepted 14 December 2018
relation shows that the non-contact method is as reliable for testing large cylindrical specimens as mea-
Available online 21 December 2018
surements done by using linear variable displacement transformer and string potentiometer. Considering
this particular large specimen, the experiment resulted in the acceptable mean difference between lateral
Keywords:
strain using both methods is 5.1 percent, and 14.5 percent for the axial strain. This occurrence is inevi-
Experimental testing
Cemented rockfill
table due to the heterogeneity of the concrete system and the placement of the monitoring point in dig-
Digital image correlation ital image correlation method, although the comparison of stress-strain relationship in both methods still
Strain measurement indicates a conformity. Based on the results of the present experiments, the authors recommend the non-
Mine backfill contact method for a detailed investigation of the material behavior during the uniaxial compressive
strength tests. Full field strain measurement enables this digital method to examine local strains near
cracks at any point, a very useful tool for studying material deformation behavior.
Ó 2019 Published by Elsevier B.V. on behalf of China University of Mining & Technology. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction recommended to place strain gages on the surface of a CRF sample

Consolidated backfilling is one type of underground mining
activity where the crushed rock, e.g. waste rock is placed back to
the goaf with the addition of a binding agent. The purpose is obvi-
ous that the binder such as cement can improve the product’s
strength and modify its deformation behavior, resulting in the
better underground stability. Cemented Rock Fill (CRF) is type of
consolidated backfill with higher compressive strength compared
to other types of backfill. While the underground stability is insep-
arable with compression stresses, the accurate measurement of
CRF’s strains caused by the corresponding stress are of interest of
researchers investigating the CRF properties.
In order to produce meaningful results during the unconfined
compressive strength (UCS) testing of the CRF material the pre-
pared samples have to be large enough to maintain high ratio
between the diameter of the sample and the aggregate size [1].
However, measuring the sample deformations using the traditional
strain monitoring techniques such as the linear variable differen-
tial transformer (LVDT) is usually challenging due to the size
restrictions of the most monitoring equipment. It is also not
⇑ Corresponding author.
E-mail address: derek.apel@ualberta.ca (D.B. Apel).
to measure its overall deformations. As such measurements would
give erroneous results that highly depend on the placement of the
strain gauges due to the CRF material’s composition consisting of a
stiff aggregate material and a soft binding agent.
The strain measurement methods that use strain gages or
LVDTs are also restricted due to internal limitations and external
factors, such as gage’s length limitation or compressometer-extens
ometer’s limited ring size and temperature. In addition, mounting
misalignment is another cause of frequent errors during the defor-
mation measurements. The recent sustainable and flexible measur-
ing device working based on a digital image correlation (DIC), a
non-contact technique for strain measurements, promises to over-
come the flaws of the former methods. In this study, the DIC sys-
tem from Correlated Solutions’ instruments has been utilized [2].
DIC generates the displacement data by comparing the differ-
ence in the position of a point of interest based on two different
images taken from the same position in a global coordinate system.
Summarized by Chu et al. in 1985 [3], the displacement function
and finite strain equations in digital-correlation work after object
projection onto a plane using a linear Taylor’s expansion:

2 
2 
@u 1 @u @v
exx ffi þ þ ð1Þ
@x 2 @x @x

https://doi.org/10.1016/j.ijmst.2018.12.002
2095-2686/Ó 2019 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
772 B.A. Lingga et al. / International Journal of Mining Science and Technology 29 (2019) 771–776

"

2 2 #
@ v 1 @u @v
2
eyy ffi þ þ ð2Þ
@y 2 @y @y

where exx is the lateral strain; eyy the axial strain; u and v are the
components of an arbitrary point displacement in the lateral and
axial direction respectively.
Since the object essentially needed to be projected onto a plane,
the DIC method was then mainly focused on how to obtain the true
coordinate of any kind of object surface by using optical instru-
ment. In the early 1990s [4,5], DIC concepts finally advanced into
stereovision systems that enabled obtaining the true three-
dimensional (3D) position of each point on a non-planar object.
3D-DIC designates two or more pinhole cameras to collect and
then compare object region images from two or more viewpoints.
For in-plane surface deformations, 3D-DIC is able to extract Fig.1. Particle size distribution of CRF aggregate.
each component accurately, despite undergoing rigid body transla-
tion and rotation [6]. This final experiment conducted by Sutton
for 28 days. This time was selected so that the backfill could reach
et al. proved that one set of a stereovision system, which consisted
its targeted compressive strength values [15].
of two standard lenses interpreting matching strains (slope of lin-
In the described experiments, the CRF mixture was designed
ear best fit of eyy is 1.31  106 which basically approaches 0) on
based on the proportion of water and cement content, as shown in
the out of plane translation experiment without any load applica-
Table 1. The general-purpose type 1 cement was used as the binding
tion or deformation.
material. Common practice of the admixture usage for CRF to reduce
In 3D-DIC, monitoring is not limited to tests on flat surfaces but
the quantity of mixing water required to produce a CRF mixture of a
it has been used when testing cylindrical specimens. These tests
certain slump was applied; the experimental specimens resembled
were described by several researchers who were able to obtain
most of CRF in the field. MasterSet Delvo admixture was added to
good results [7–9]. However, it is still difficult to find literature on
the specimen mixture. Table 1 lists both unit weight and dry unit
testing large CRF specimens and investigating strain along develop-
weight, as well as UCS of experimental specimen.
ing fractures during the UCS tests. The monitoring of fracture devel-
The specimens were finished by capping the top and bottom
opment is another advantage of non-contact DIC method.
surfaces, and spray painting the white and black speckles onto
In this paper, conventional fixed mounted LVDTs and string
the samples surface. The capping process utilized high-strength
potentiometers are reassessed by the 3D-DIC method during CRF
gypsum paste that follows ASTM C617m [16] to ensure the uni-
specimens testing.
form load distribution applied on the specimen. While the black-
white speckles are used by the DIC cameras to track the deforma-
2. Materials and experimental setup tion of multiple points painted onto the specimen surface. Fig. 2
shows the selected points of interest on the tested specimen.
The UCS test was conducted for six CRF specimens of 4.5 per-
cent cement content and 304.8 mm in diameter. During the test, 2.2. Experimental setup
the deformations were recorded by DIC system using VID-3D soft-
ware [10]. In addition to two fixed mounted LVDTs, three string 2.2.1. Load frame and contact method setup
potentiometers were recording the lateral and axial deformations. The UCS tests were performed under constant displacement
Detailed information and experimental setup are explained in the control using MTS loading frame with the 6230 kN capacity. Each
sections below. specimen was loaded using the axial load till reaching failure.
The initial load was applied first on each sample for seating the
2.1. Cemented rockfill specimen specimen on the loading plates. The test was running on the con-
trolled 0.005 mm/s displacement rate.
The CRF mixture was created in the University of Alberta rock Two LVDTs were each placed at diametrically opposite points
mechanics laboratory to mimic the CRF mixture that is typically about mid-height of the specimen. These two LVDTs recorded the
used as a backfill in many underground mines in Canada [11]. lateral deformation to the data logger; these data were then con-
The tested CRF mixture consisted of screened <50.8 mm granite verted to the lateral strains. Similar LVDT setup was used to mea-
used as an aggregate. An initial sieve test of particle size distribu- sure the axial deformations. The axial strain was calculated from
tion (PSD) was conducted and the rock particle size curve is shown the average of three string potentiometers attached from the bot-
in Fig. 1. The curve shows that the aggregate size gradation in this tom to the top cap of the MTS frame. Fig. 3 shows the setup of this
experiment does not match with the one in regular concrete shown
in ASTM C33M standard [12]. However, the mixture fits Talbot
Table 1
curve [13] with value of the exponent (n) of 0.35. Generated Talbot
CRF mixture properties.
curve from Talbot general equation can be used as a reference for
the aggregate gradation or waste rock size in designing the CRF Properties Value
mixtures [14]. Mix type CRF 4.5%
The specimens were molded with the diameter to length ratio Aggregate size Waste rock < 50.8 mm
of 304.8 mm to 609.6 mm. The diameter accommodates more than Cement to aggregate ratio (%) 4.5
Water to cement ratio 2
three times the nominal maximum size of the coarse aggregate, Water to solid ratio 0.087
which is 50.8 mm. The molds in this experiment were made from Admixture Delvo (L/100 kg cement) 0.475
Sonotube. Each CRF mixture was vibrated using the vibrating rod Unit weight (kg/m3) 2358
- this was done every one third of height of the mold during the Dry unit weight (kg/m3) 2191
UCS (MPa) 2.7
pouring process. Prior to testing, specimens were cured indoors
 B.A. Lingga et al. / International Journal of Mining Science and Technology 29 (2019) 771–776
Fig. 2. Prepared specimen.
contact method; unfortunately, the third string potentiometer
could not be shown as it was placed behind the sample.
2.2.2. 3D-DIC setup
As illustrated in Fig. 4, two cameras (one stereovision system)
were set up in front of the specimen, covering approximately half
of the cylinder surface. Incorporating Fujinon 1:1.4/12.5 mm lenses
and Point Grey cameras generated a resolution of 2448  2048 pix-
els. Adequate illumination by indoor lamps and two additional task
lamps was used to depict each distinctive image. Each of the
images was captured with the rate of one image every two sec-
onds. The cameras were calibrated before the tests using a calibrat-
ing board with 12  9 grid on 25 mm spacing.
The image correlation process was run using the VIC-3D 2009
software. The procedure started with first creating an area of inter-
est (AOI) that covered the specimen surface in the reference image
and placing the start point around the bottom (fixed part) of the
specimen as the AOI. During the second step, the subset and step
size were defined. Subsets are various small regions throughout
the images which are located by the digital image system, while
the step size is spacing between subset centers both vertically
and horizontally. Dividing the image size with these subsets
together with the step size will result in particular numbers of
vertical and horizontal array of displacement data. The suggested
Fig. 3. Setup of LVDTs and string potentiometers.
773
subset size of 21 (21  21 pixels2) was used to give an optimal
match confidence of 0.01 pixel, while a step size of five pixels
was chosen with the consideration of processing time. The third
step consisted of running the correlation.
The original correlation outcome was then converted into cylin-
drical interpretation by applying cylindrical transformation option
in VIC-3D software. Next, the rigid body motion was removed - this
feature was also provided by the software. Afterwards, the strain
can be properly computed to represent both axial strain and lateral
strains. Five consecutive points along the perimeter in line with the
LVDT height became the points of interest and produced the global
axial and lateral strains to be analyzed for this experiment (see
Fig. 5).
A justification upon the 3D-DIC’s five points of interest that in
line with the LVDT height is established with some practical judge-
ments. One pair of fixed mounted LVDT in producing a lateral strain
of whole CRF is limited only to the strain at corresponding contact
points, although concrete system like CRF is in fact a heterogeneous
entity. The most laterally expanded region during the compression
test is supposedly taking place around the mid height of the speci-
men. Therefore, the mean lateral strain and axial strain from five
points from 3D-DIC which are more than two points from LVDT
at the same mid-height are assumed adequate to represent the glo-
bal strains of CRF. As with any system based on a photogrammetry
the main limitation of the 3D-DIC system is that it can only track the
movement of points which are located within the field of view of
the camera. Therefore, during the tests the LVDT’s were positioned
on the opposite sides of the cylindrical samples in such way that
they did not obstruct the camera’s field of view. On the other hand
positioning the LVDT’s contact points on the most lateral edge of
the camera’s field of view of the cylindrical CRF sample forced us
to offset the 3D-DIC control points away from the edge of the sam-
ple due to the difficulty of obtaining clear camera focus on the edges
of the cylindrical sample during the test.
Based on 304.8 mm to 609.6 mm specimen dimension, the opti-
mal minimum lateral and axial strains that can be captured by DIC
system are 0.001245 and 0.002976 or better, respectively. As pre-
viously mentioned, DIC has an optimal match confidence, position
accuracy of ‘0.01 pixels or better’ after distortions are removed
from the images [17]. With an assumption of 0.01 pixels accuracy
is applied on this experiment, and the image resolution generated
from Fujinon lenses is 2448  2048 pixels2, the displacement accu-
racy of specimen experimentation by DIC can be first calculated
with Eq. (3). The displacement accuracy is then processed further
to generate the minimum strain.
L
bo ¼ b ð3Þ
N l
where bo represents the displacement accuracy; L the field of view;
N the recording resolution; and bl the DIC position accuracy.
3. Results and discussion
3.1. Typical stress-strain relationship of the experimental results
Fig. 6 shows stress-lateral strains measured by the LVDT and
the 3D-DIC monitor that followed strains at the selected five points
of interest in the mid-section of the first sample, CRF-1. Each of the
recorded 3D-DIC’s stress-lateral strain curves reasonably follows
the curve recorded by the LVDT till reaching the sample peak
strength value. The post failure stress-strain curves at each of the
analyzed points show more variation between each other. The
divergence in the post failure stress-strain curves was probably
caused by creation of various fractures within the tested sample.
Some of these fractures are visible such as the fractures adjacent
to some of the analyzed points of interests but some could have
774 B.A. Lingga et al. / International Journal of Mining Science and Technology 29 (2019) 771–776
Fig. 4. Monitoring setup for UCS testing of large CRF samples.
Fig. 5. Five discs point of interest along the perimeter line of the specimen’s mid-
height, p1-p5 (left to right). exx contour from one of the deformed 1st specimen
(CRF-1) after failure.
Fig. 6. Stress-lateral strain CRF-1.
occurred inside the sample causing the divergence in the post fail-
ure stress-strain curves.
Fig. 7 shows the corresponding CRF-10 s axial stress-strain rela-
tionship graph during the test. The axial stress-strain curves
recorded by both methods also show similar trends to stress-
lateral strains transcription.
Fig. 7. Stress-axial strain CRF-1.
As illustrated in both Figs. 6 and 7, the strains from all points of
interest generally show consistency and conformity with the con-
tact method until the specimen departs from the elastic zone.
Afterwards, the deviation becomes greater especially passed the
peak strength. This particular finding corresponds with the micro
cracking behavior of concrete under compressive stress and the
heterogenic nature of the concrete itself. Asserting that the micro-
cracking behavior of concrete system [18] does also exist in the
CRF material. Accordingly, for loads up to 70 percent of the peak
strength the deviation of stress-strain curve is negligible or only
at initiation stage (aggregate-paste cracking). Above 70 percent
of the peak strength, the paste matrix cracking occurs that bridg-
ing between existing aggregate-paste cracks. Therefore, the 3D-
DIC and contact method strains follow the same trend up to
around 70 percent of the peak strength. The initiated aggregate-
paste cracks develop later in the paste matrix. These cracks prop-
agated randomly depending on the heterogeneity or complexity of
concrete system and they may be caused by factors such as settle-
ment of aggregate, bleeding of mixing water and shrinkage stres-
ses induced by the drying process. Thus, each point of the CRF’s
surface may have in fact different strain values after the load
exceeds the 70 percent peak strength.
Two figures (Figs. 6 and 7) represent the typical stress-strain
transcription result of the entire experiment. All other five experi-
mental samples show different strain deviations from one to
another different point, similar pattern to CRF-1 sample.
 B.A. Lingga et al. / International Journal of Mining Science and Technology 29 (2019) 771–776 775

Fig. 10. Axial strain ratio 3D-DIC to string potentiometer.

Table 2
CRF mixture properties.
Fig. 8. Axial strain ratio 3D-DIC to string potentiometer, CRF-1.
CRF 4.5% Linear-fit ratio between 3D-DIC and contact method
3.2. Analysis of strain correspondence between contact and DIC Specimen# Lateral strain ratio (gradient) Axial strain ratio (gradient)
methods CRF-1 0.925 1.0604
CRF-2 1.273 1.1827
The analysis of axial strain correspondence between 3D-DIC and CRF-3 0.852 1.1163
string potentiometer is represented in Fig. 8 which is derived from CRF-4 0.826 1.2190
CRF-5 1.083
the same specimen as shown in Fig. 5, CRF-1. In Fig. 8, the vertical CRF-6 1.349
(y) axis is the 3D-DIC eyy mean value from the five points and the Mean 1.05132 1.1446
horizontal (x) axis is the three string potentiometers exx mean
value. When theoretically the 3D-DIC strain is equal to the string
potentiometer strain, plotting them on a Cartesian plane along
the test will create a y equals to x or a 45° slope line. A mean axial mid-height, was 6.04 percent higher than total axial strain from
strain value from 5 points of interest 3D-DIC is then plotted to the string potentiometer.
string potentiometer strain until the specimen reaches the UCS. A summary of all ratios of experimental specimens’ results
An attempt was performed to verify whether the axial strain at between the two methods for lateral and axial strain respectively
five points of interest on the specific mid-height of the specimen is given in Figs. 9 and 10. The mean differences of 5.1 percent
was similar to the axial strain of whole CRF as recorded by the and 14.5 percent were found for lateral and axial strain respec-
string potentiometer. Here, the test on CRF-1mixture showed that tively (see Table 2).
the 3D-DIC axial strain from specific location measured at sample These two figures also show the range of strain until failure at
UCS (i.e., the length of axes) for this particular large 4.5 percent
CRF. The lateral strain and axial strain at failure are approximately
0.01 and 0.006. Back calculating these values to the specimen
dimension (diameter and length), then to the mean value of strain
ratio difference from both measurement methods (Table 2), will
approximate the maximum (at UCS) difference of 3D-DIC to con-
ventional method in magnitude of mm. Therefore, the maximum
difference between both methods for large scale CRF specimen in
this experiment is 156 mm for the lateral and 529 mm for the axial.

4. Conclusions

According to the results of this experiment, the non-contact

Fig. 9. Lateral strain ratio 3D-DIC to LVDT.
method of calculating a large scale specimen’s elastic properties
by 3D-DIC has been validated. One stereovision system with five
points of interest laid out its strain transcription in the same order
as the conventional LDVT and string potentiometer methods. Mean
differences of 5.1 percent and 14.5 percent were found for respec-
tively lateral and axial strain. Therefore, for a large scale experi-
ment with dimension of diameter to length of 304.8 mm to
609.6 mm, the both methods’ maximum measurement (at failure)
differences are approximately 156 mm and 529 mm.
776 B.A. Lingga et al. / International Journal of Mining Science and Technology 29 (2019) 771–776
Investigation of strain at multiple points by using 3D-DIC sys-
tem on experiment’s large CRF specimen discovered the strain
behavior around the failure crack. Thus, the placement of point
of interest is greatly influencing the 3D-DIC’s generated elastic
properties. The user understanding and comprehension of what
to examine in the experiment is strongly recommended. In general,
non-contact method by 3D-DIC is acceptable to be used for large
scale specimen like CRF’s elastic properties determination. The
advantage of using this method for examining local strain is mark-
edly. It should be noted that the described experimental results
were obtained from limited number of tested samples. Due to
the difficulty in handling and preparation of such large specimens
only six samples were tested for this study.
References
[1] ASTM C192/C192M. Standard Practice for Making and Curing Concrete Test
Specimens in the Laboratory. ASTM Int 2016:1-8.
[2] Correlated Solutions. Correlated Solutions-Products 2017.
http://correlatedsolutions.com/products/ (accessed November 22, 2017).
[3] Chu TC, Ranson WF, Sutton MA. Applications of digital-image-correlation
techniques to experimental mechanics. Exp Mech 1985;25:232–44.
[4] Luo PF, Chao YJ, Sutton MA, Peters WH. Accurate measurement of three-
dimensional deformations in deformable and rigid bodies using computer
vision. Exp Mech 1993;33:123–32.
[5] Helm JD, Sutton MA, McNeill SR. Improved three-dimensional image
correlation for surface displacement measurement. Soc Photo-Optical
Instrum Eng 1996;35:1911–20.
[6] Sutton MA, Yan JH, Tiwari V, Schreier HW, Orteu JJ. The effect of out-of-plane
motion on 2D and 3D digital image correlation measurements. Opt Lasers Eng
2008;46:746–57.
[7] Lu H, Vendroux G, Knauss WG. Surface deformation measurements of a
cylindrical specimen by digital image correlation. Exp Mech 1997;37:433–9.
[8] Peters III WH, Sutton MA, Ranson WF, Poplin WP, Walker DM. Whole-field
experimental displacement analysis of composite cylinders. Exp Mech
1989;29:58–62.
[9] Santos AHA, Pitangueira RLS, Ribeiro GO, Carrasco EVM. Concrete modulus of
elasticity assessment using digital image correlation. Rev IBRACON Estruturas
E Mater 2016;9:587–94.
[10] Correlated Solutions. Vic-3D 2010 Reference Manual. 2010.
[11] Sepehri M, Apel DB, Hall RA. Prediction of mining-induced surface subsidence
and ground movements at a Canadian diamond mine using an elastoplastic
finite element model. Int J Rock Mech Min Sci 2017;100:73–82.
[12] ASTM C33. Standard Specification for Concrete Aggregates. ASTM Int 2003;
i:11.
[13] Talbot AN, Richart FE. The strength of concrete, its relation to the cement
aggregates and water. Champaign: University of Illinois at Urbana; 1923.
[14] Swan G. A new approach to cemented backfill design. CIM Bull 1985;78:53–8.
[15] Belem T, Benzaazoua M, Bussière B. Mechanical behaviour of cemented paste
backfill. Proc. 53rd Can. Geotech. Conf. Montr., Montreal 2000:373–80.
[16] ASTM C617/C617M12. Standard Practice for Capping Cylindrical Concrete
Specimens. ASTM Int 2012;98:1-6.
[17] Schreier H, Orteu JJ, Sutton MA. Image correlation for shape, motion and
deformation measurements. Boston, MA: Springer, US; 2009.
[18] Relationship Meyers BL. Between time-dependent deformation and
microcracking of plain concrete. ACI J Proc 1969:66.