Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

Contents lists available at ScienceDirect

Journal of Rock Mechanics and

Geotechnical Engineering
journal homepage: www.rockgeotech.org

Full length article

Full-eld mapping of internal strain distribution in red sandstone

specimen under compression using digital volumetric speckle
photography and X-ray computed tomography
Lingtao Mao a, b, c, *, Jianping Zuo a, Zexun Yuan a, Fu-Pen Chiang b, c
a
State Key Laboratory of Coal Resources and Safe Mining, China University of Mining & Technology (Beijing), Beijing 100083, China
b
Laboratory for Experimental Mechanics Research, Stony Brook University, Stony Brook, NY 11794-2300, USA
c
Department of Mechanical Engineering, Stony Brook University, Stony Brook, NY 11794-2300, USA

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

Article history: It is always desirable to know the interior deformation pattern when a rock is subjected to mechanical
Received 20 November 2014 load. Few experimental techniques exist that can represent full-eld three-dimensional (3D) strain
Received in revised form distribution inside a rock specimen. And yet it is crucial that this information is available for fully un-
26 January 2015
derstanding the failure mechanism of rocks or other geomaterials. In this study, by using the newly
Accepted 28 January 2015
Available online 4 March 2015
developed digital volumetric speckle photography (DVSP) technique in conjunction with X-ray computed
tomography (CT) and taking advantage of natural 3D speckles formed inside the rock due to material
impurities and voids, we can probe the interior of a rock to map its deformation pattern under load and
Keywords:
Red sandstone
shed light on its failure mechanism. We apply this technique to the analysis of a red sandstone specimen
Strain localization under increasing uniaxial compressive load applied incrementally. The full-eld 3D displacement elds
Uniaxial compression are obtained in the specimen as a function of the load, from which both the volumetric and the deviatoric
Digital volumetric speckle photography strain elds are calculated. Strain localization zones which lead to the eventual failure of the rock are
(DVSP) identied. The results indicate that both shear and tension are contributing factors to the failure
X-ray micro-tomography mechanism.
2015 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by
Elsevier B.V. All rights reserved.

1. Introduction images obtained from plane strain experiments. But the 2D ob-
servations are limited in their capability to resolve the geometric
The macroscopic deformation and failure of rock is a gradual complexities and heterogeneity in geomaterials.
process of damage accumulation, crack initiation, propagation, X-ray computed tomography (CT), as a non-destructive three-
interaction and then the eventual failure (Amitrano, 2006). Failure dimensional (3D) imaging technique, has been used to investigate
rst manifests itself with the appearance of strain localization and the internal structures, deformation localization and failure of
then the creation of a damage zone. The localization of damage and geomaterials. In applications of the X-ray CT, the loading test and
strain will result in the stress redistribution and thus weaken the the scanning were not conducted simultaneously (Desrues et al.,
mechanical performance of the rock. In order to understand the 1996; Alshibli et al., 2000). Only the density of the specimen or
mechanisms and evolution of damage or strain localization in the CT number distribution inside the specimen was used to reveal
rocks, full-eld deformation measurement methods, such as ster- the localized zones (Bsuelle et al., 2000; Viggiani et al., 2004; Louis
eophotogrammetry (Desrues and Viggiani, 2004) and digital image et al., 2006; Suzanne et al., 2008). The 2D DIC technique has been
correlation (DIC) (Kozicki and Tejchman, 2007; Hall et al., 2010a, b; employed to analyze radiography (Louis et al., 2007) or the
Dautriat et al., 2011; Nguyen et al., 2011; Lin and Labuz, 2013; Zhang sectional CT images (Adam et al., 2008) for assessing the internal
et al., 2013) have been used, mostly with two-dimensional (2D) deformation patterns in the geomaterials. A signicant limitation of
this approach is the fact that it only allows the quantication of 2D
displacement eld and 2D strain distribution in the sectional plane
* Corresponding author. Tel.: 86 10 82386706. while ignoring the out-of-plane displacement all together (Adam
E-mail address: [email protected] (L. Mao). et al., 2013). This information is not sufcient to fully evaluate the
Peer review under responsibility of Institute of Rock and Soil Mechanics, onset and evolution of localized deformation. Bay et al. (1999)
Chinese Academy of Sciences.
developed a 3D strain mapping technique using 3D digital image
1674-7755 2015 Institute of Rock and Soil Mechanics, Chinese Academy of Sci-
ences. Production and hosting by Elsevier B.V. All rights reserved. volume correlation, called digital volume correlation (DVC), and
http://dx.doi.org/10.1016/j.jrmge.2015.01.003 have measured displacement and strain elds in trabecular bones
 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146 137

under compression. By combining in situ CT scanning and DVC, to scan the specimen in each loading step. After scanning, the
some studies have been carried out on a number of geomaterials, reconstruction is carried out with the Feldkamp cone-beam
such as rocks (Lenoir et al., 2007; Charalampidou et al., 2011, 2014) reconstruction algorithm. The reconstructed volume images have
and granular materials (Hall et al., 2010a, b; Adam et al., 2013). In 566 voxel  566 voxel  954 voxel (where a voxel is the 3D
the DVC method, a cubic subset surrounding the interrogated point equivalent of a pixel) and cover the specimen with height of 43 mm.
located in the reference volume image is selected and correlated The physical size of a voxel is 45 mm3. Based on the projection image
with the corresponding location in the deformed volume image. of the specimen in each loading step, the global displacement along
The resulting displacement vector is obtained. The theory is simple. z-axis is measured, from which the axial strain is obtained. The
But in the practical implementation due to the vastly increased stressestrain curve from the loading history is shown in Fig. 2.
volume of data associated with the undeformed and deformed Because the metal compression disks inuence the top and bottom
images and the increased degree-of-freedom (DOF), the DVC is slice images, only the middle of the volume image with size of 566
facing some challenges, such as the implementation complexity, voxel  566 voxel  801 voxel is analyzed. In Fig. 3, the recon-
the measurement accuracy and the computational efciency (Pan structed volume image of step 8, three orthogonal sections, and
et al., 2012). Digital volumetric speckle photography (DVSP) is sectional images along x 12.5 mm and y 12.5 mm of steps 7 and
another 3D strain analysis technique, which is the extension of 2D 8 are shown, respectively.
digital speckle photography (DSP) technique that offers some ad- Fig. 4 shows the gray value distribution curves throughout the
vantages over the DVC technique in the computational efciency. volumetric images at different loading steps. It is noted among
Details of the evolution of the speckle technique leading to the these curves that the distributions for steps 6 and 7 are slightly
development of DVSP can be found in Chiang and Mao (2015). narrower than those of other steps. Table 1 lists the mean gray
In this study, we apply the DVSP technique in conjunction with values and the standard deviations. It can be seen that the mean
X-ray micro-tomography to obtain the 3D interior strain elds in a gray value has a slightly higher increment from step 1 to step 4,
red sandstone specimen under uniaxial compression, and then indicating globally a lower porosity and higher density; and then
discuss the accuracy of DVSP technique. declines after step 4, indicating a higher porosity and lower density.
This implies that the initial microcracks are closed under
2. Experimental procedure compressive loading before step 4, and microcrack development
makes the specimen dilatant after step 4. However, gray value
2.1. Experimental setup and imaging procedure distribution curves and CT gray level images alone are not very
effective in differentiating the strain localization area nor the
In this study, the main components of the industrial X-ray CT microcrack onset and development. Thus, to shed more light on the
system are a microfocus X-ray source from YXLON (FeinFocus deformation characteristics of the specimen, the newly developed
225 kV), a X-ray detector unit (1024 pixel  1024 pixel) from Per- full-eld strain measurement technique, DVSP, is employed as
kinElmer (XRD 0822AP 14), and a motorized rotation stage from follows.
Newport, USA. The X-ray has a focus with size of 3 mm  6 mm, a
voltage range of 50e225 kV, and the tube current ranging from 0 to 2.2. Elements of the DVSP technique
1440 mA. A simple uniaxial compression setup is designed and built
that would allow the operation of micro-tomography of a specimen From the sectional images shown in Fig. 3, it can be seen that
under load in situ. The setup cell is made of PMMA, which is there are several components and pores in the red sandstone, and
transparent to X-rays. The CT system and loading setup are shown different components have different gray values. This natural
in Fig. 1. structure can be regarded as a pattern of volumetric speckles
A cylindrical specimen of red sandstone of f25 mm  50 mm in pattern and carries the information of deformation. This naturally
size and a porosity of 23.3% is placed in the cell. The compaction of presented structural pattern is used in the analysis adopting the
the specimen is achieved by applying a compressive load in the DVSP technique. The volumetric image of step 1 is dened as the
axial (z) direction. The X-ray source to the specimen and source-to- reference volumetric image, and the volumetric image of step 2 is
detector distances are 139 mm and 696 mm, respectively, resulting dened as the deformed one. They are then divided into subsets
in a 5.0 times magnication. The whole compression process is with a cubic array of 64 voxel  64 voxel  64 voxel; there is a 32-
divided into 8 steps. In each step, the loading is kept constant while voxel overlap between neighboring subsets, and then compared.
the specimen is scanned. During the scanning, 720 projections are
captured and distributed at equal angles over 360 . It takes 25 min
14 F
7
12
6
Detector Loading setup X-ray source
10
Axial stress (MPa)

5
8
4 8
F
6
3
4
2
2
1
Rotation stage
0
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007
Axial strain

Fig. 1. Industrial CT and loading setup. Fig. 2. Stress-strain curve.

138 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

z
y h1 x; y; z hx; y; z (1)

x
h2 x; y; z hx  ux; y; z; y  vx; y; z; z  wx; y; z (2)

z (mm)
z (mm)

where u, v and w are the displacement components experienced by

the speckles along the x, y, and z directions, respectively. A rst-step
3D fast Fourier transform (FFT) applied to both h1 and h2 yields


H1 fx ; fy ; fz Jfh1 x; y; zg

 h
i
 
x (mm) x (mm) H1 fx ; fy ; fz exp jF1 fx ; fy ; fz (3)
y (mm) y (mm)
(a) (b)


H2 fx ; fy ; fz Jfh2 x; y; zg

 h
i
 
H2 fx ; fy ; fz exp jF2 fx ; fy ; fz

 nh

z (mm)

 
z (mm)

zH1 fx ; fy ; fz exp j F1 fx ; fy ; fz

io
 2p ufx vfy wfz 4

y (mm) x (mm) where H1(fx, fy, fz) and H2(fx, fy, fz) are the Fourier transforms of h1(x,
(c) (d) y, z) and h2(x, y, z), respectively; J stands for Fourier transform;
jH1(fx, fy, fz)j and jH2(fx, fy, fz)j are the spectral amplitudes of H1(fx, fy,
fz) and H2(fx, fy, fz), respectively; F1(fx, fy, fz) and F2(fx, fy, fz) are the
phases of H1(fx, fy, fz) and H2(fx, fy, fz), respectively.
Then, a numerical interference between the two 3D speckle
patterns is performed at the spectral domain, i.e.
z (mm)

z (mm)




 H1 fx ; fy ; fz H2* fx ; fy ; fz
F fx ; fy ; fz r



 
H1 fx ; fy ; fz H2 fx ; fy ; fz 

 nh

y (mm) x (mm)  
zH1 fx ; fy ; fz exp j F1 fx ; fy ; fz
(e) (f)
io
 F2 fx ; fy ; fz 5
Fig. 3. Reconstructed image and sectional images. (a) Reconstructed image of step 8;
(b) Three orthogonal sections; (c) Section along x 12.5 mm at step 7; (d) Section
along y 12.5 mm at step 7; (e) Section along x 12.5 mm at step 8; (f) Section along
y 12.5 mm at step 8. It is seen that

The schematic of the processing algorithm of DVSP is demonstrated



in Fig. 5. F1 fx ; fy ; fz  F2 fx ; fy ; fz 2pufx vfy wfz (6)
Let h1(x, y, z) and h2(x, y, z) be the gray intensity functions of a
pair of generic volumetric speckle subsets, before and after defor- Finally, the following function can be obtained by performing
mation, respectively, we have another 3D FFT:

n
o
1.00E-03
Gx; h; z J F fx ; fy ; fz Gx  u; h  v; z  w (7)
Step1
8.00E-04 Step2
Step3
Frequency

6.00E-04 Table 1
Step4
Statistics of gray values.
Step5
4.00E-04 Step Axial stress (MPa) Mean value Standard deviation
Step6
1 0 4227 474
2.00E-04 Step7 2 2.17 4230 487
Step8 3 4.75 4232 490
4 6.09 4232 491
0.00E+00 5 7.97 4229 471
2000 2500 3000 3500 4000 4500 5000 5500 6000
6 10.18 4227 442
16 bit gray value
7 11.98 4213 449
8 After peak 4074 540
Fig. 4. Gray value distribution curves of the volumetric images.
 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146 139

Fig. 5. Schematics demonstrating the processing algorithm of DVSP.

Eq. (7) is an expanded impulse function located at (u, v, w). 9

vu vv vw
This process is carried out for every corresponding pair of the 3x a1 ; 3 y b2 ; 3 z c3 >
>
>
subsets. By detecting the crest of all these impulse functions, an vx vy vz >
>
  >
>
array of displacement vectors at each and every subset is >
>
1 vv vu 1 >
>
3 xy b1 a2 >
>
obtained. 2 vx vy 2 =
Because of the discrete nature of digital volume images, the   (9)
1 vw vv 1 >
>
displacement vectors evaluated from Eq. (7) are integer multiples c2 b3 >
>
3 yz
2 vy vz 2 >
>
of one voxel. We select a cubic subset with 3 voxel  3 voxel  3 >
>
  >
>
voxel surrounding an integer voxel of the crest and a cubic 1 vu vw 1 >
>
3 xz a3 c1 >
;
spline interpolation is employed to obtain the sub-voxel accu- 2 vz vx 2
racy. From 3D displacement elds, the internal strain tensor 3 can
then be calculated using an appropriate strain-displacement In the afore described processing scheme, the volume image of
relation. step 1 is kept as the reference image and the subsequent deformed
images are compared to the reference image sequentially from
which the displacement elds and strain elds in each step are
2.3. Strain computation calculated. The size of the volume image used in the calculation has
566 voxel  566 voxel  801 voxel. The subset size has 64
The internal strain tensor 3 can be derived from the displace- voxel  64 voxel  64 voxel, and there is a 32-voxel overlap be-
ment elds. If the displacement measurements involve errors, er- tween neighboring subsets.
rors may be amplied during the strain computation. In this study,
we use the point-wise least-squares (PLS) approach (Pan et al.,
2009, 2012) to calculate strain tensors. In the PLS approach, the
3. Results
errors in local displacement elds can be signicantly reduced
during the process of local tting.
3.1. Displacement elds
To compute the local strains of each point considered, a reg-
ular cubic box with size of (2N 1)  (2N 1)  (2N 1)
The u, v and w displacement elds of the specimen along the x-,
discrete points surrounding the considered point is selected.
y- and z-axes at step 7 are illustrated in Fig. 6, respectively.
If the strain calculation window is small enough, the displace-
At the step 7, the u displacement ranges from 110 mm to 61 mm,
ments in each direction can be reasonably assumed to be
the v displacement from 95 mm to 84 mm, and the w displacement
linearly distributed, and therefore can be mathematically
from 185 mm to 376 mm. As expected, the largest displacements
expressed as
occur in the top region along the z-axis. Due to the heterogeneity of
the red sandstone specimen, the deformation is not uniform and
9
ux; y; z a0 a1 x a2 y a3 z = with different features in different sections. We select two
vx; y; z b0 b1 x b2 y b3 z (8) orthogonal sections along x 12.5 mm and y 12.5 mm at
;
wx; y; z c0 c1 x c2 y c3 z different steps, representing the motion of the material in these
sections along the x-, y- and z-axes, respectively, as shown in
where x, y, z [eN, N] are the local coordinates within the strain Figs. 7e9. The analyzed areas correspond to the rectangular areas
calculation box; u(x, y, z), v(x, y, z) and w(x, y, z) are the displace- marked with dash lines in Fig. 3e and f. In Figs. 7 and 8, it is noted
ments directly obtained by the DVSP method; and ai, bi and ci (i 0, that the regions with the negative and positive displacements
1, 2, 3) are the unknown polynomial coefcients to be determined. manifest themselves more clearly as the load increases. The inter-
With the least-squares or multiple regression analysis, the un- face between the two regions has zero displacement and is marked
known coefcients can be estimated. Then, the six Cauchy strain by a dash line in Figs. 7c and 8c. At step 8, the specimen is broken,
components 3 x, 3 y, 3 z, 3 xy, 3 xz, 3 yz at the interrogated point can thus be and the main cracks shown in Fig. 3e and f are mainly predicated by
calculated as the zero displacement interface, i.e. along the marked dash lines.
140 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

Fig. 6. Displacement elds of the specimen at step 7. (a) u elds, (b) v elds, (c) w elds.

3.2. Strain elds The evolution of deviatoric strain elds and the volumetric
strains is illustrated in Figs. 10 and 11, respectively. At the axial
In general, the strain distribution provides more information on stress level of 6.09 MPa, it can be seen from Fig. 10a that some re-
the characteristics of deformation localization and damage evolu- gions in lighter color indicate the presence of higher deviatoric
tion in the rocks. Derived from the above displacement elds, six strain value as a result of the material heterogeneity, leading to the
Cauchy strain components are computed by using PLS approach, formation of microcracks. At this stress level, the mean volumetric
and then the principal strains are obtained. The deviatoric strain 3 s strain is 770  106, indicating the volume reduction of the
and the volumetric strain 3 v are written as specimen under compression. At the axial stress level of 7.97 MPa, a
strain localization region appears as indicated by the dash line loop
p q marked in upper of Fig. 10b. As the load increases, the strain
2
3s 3 1  3 2 2 3 2  3 3 2 3 3  3 1 2 (10) localization region starts to extend upward and new regions are
3 generated as shown in upper of Fig. 10c. The location of this region
roughly corresponds to the macrocrack shown in Fig. 3e. This im-
3v 31 32 33 (11) plies that the crack is caused by shear stress. In the volumetric
strain distribution maps depicted in upper of Fig. 11b and c, the
where 3 1, 3 2 and 3 3 are the major, intermediate and minor principal volumetric strain in this region also rises. When the stress level
strains, respectively. increases from 7.97 MPa to 11.98 MPa, the volumetric strain grows

Fig. 7. Distribution of u displacement elds of two orthogonal sections at different steps. (a) Step 4 (axial stress 6.09 MPa), (b) Step 5 (axial stress 7.97 MPa), (c) Step 7 (axial
stress 11.98 MPa).
 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146 141

Fig. 8. Distribution of v displacement elds of two orthogonal sections at different steps. (a) Step 4 (axial stress 6.09 MPa), (b) Step 5 (axial stress 7.97 MPa), (c) Step 7 (axial
stress 11.98 MPa).

Fig. 9. Distribution of w displacement elds of two orthogonal sections at different steps. (a) Step 4 (axial stress 6.09 MPa), (b) Step 5 (axial stress 7.97 MPa), (c) Step 7 (axial
stress 11.98 MPa).

rapidly as shown in Fig. 11c, and the specimen dilates. The high We calculate the mean value of the deviatoric strains and the
strain region largely matches the macrocracks shown in Fig. 3e and volumetric strains throughout the specimen, and plot their values
f. Judging from the patterns of crack shape and strain distribution, it and the axial stress as a function of the axial strain, as shown in
is reasonable to state that the specimen destruction is combination Fig. 12. The distributions indicate ve phases of specimen defor-
of tension failure and shear failure. mation development. In the rst compaction phase AB, the
142 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

Fig. 10. Deviatoric strain 3s distribution of two orthogonal sections at different steps. (a) Step 4 (axial stress 6.09 MPa), (b) Step 5 (axial stress 7.97 MPa), (c) Step 7 (axial
stress 11.98 MP).

Fig. 11. Volumetric strain 3v distribution of two orthogonal sections at different steps. (a) Step 4 (axial stress 6.09 MPa), (b) Step 5 (axial stress 7.97 MPa), (c) Step 7 (axial
stress 11.98 MPa).
 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146 143

13 A B C D E F 0.005

11 z
0.004
9
v 0.003
Axia l stress (MPa )

5 s 0.002

Strain
3 0.001
1
0
-1 0 0.002 0.004 0.006
-0.001
-3 Axial strain
-5 -0.002

Fig. 12. Deviatoric strain and volumetric strain curves.

volumetric strain is negative, indicating that the pre-existing

microcracks and pores in the material are being closed. In the
elastic phase BC, the relation between the axial strain and stress is
in approximately linear relationship, whereas the volumetric strain
decreases and the deviatoric strain becomes higher. At point C, the
volumetric strain value reaches a minimum value. After that, the
volumetric strain increases sharply while the deviatoric strain
grows in a nonlinear fashion. The stress at point C is dened as the
dilation initiation point with a load of 6.07 MPa which is 50.9% of
peak load. In the phase CD, the microcracks smaller than the voxel
size (i.e. the spatial resolution limit) of the CT image are being
developed, resulting in some discrete strain localization regions
shown in upper of Fig. 10a with light color. At point D, the volu-
Fig. 13. CT Images of specimen and normalized autocorrelation function curves. (a)
metric strain is zero. Beyond point D the volumetric strain con- Red sandstone, (b) Medium sandstone, and (c) Associated centered and normalized
tinues to rise dramatically, indicating that the specimen dilation autocorrelation radius at half height.
speed increases. It is reasonable to speculate that, at this stage, the
microcracks grow rapidly and eventually coalesce into macro-
cracks. But the specimen still retains its load-bearing capacity. In images. In this study, the autocorrelation radius r is the radius at
the phase DE, the deviatoric strain also grows rapidly which half height of the normalized autocorrelation function of the CT
prompts the development of strain localization regions. We dene images (see Fig. 13c) (David, 2004). The autocorrelation radius of
point D as the initiation point of a macrocrack (larger than the the red sandstone is 1.4 voxels (1 voxel 45 mm3), and the auto-
voxel) as the result of the coalesce of microcracks parallel to the correlation radius of medium sandstone is 4.1 voxels.
loading direction. The load at point D is 8.25 MPa which is 68.9% of Two 200 voxel  200 voxel  200 voxel volume images are
peak axial load. The phase EF is the post-peak phase and the cropped from the initial volume images of these materials and
specimen fails. dened as reference images, respectively. The deformed volume
images are then obtained by the Fourier transform method
(Schreier et al., 2000). For simplicity, only a pure rigid body trans-
4. Discussions
lation is imposed. Ten different sub-voxel displacements ranging
from 0.1 voxel to 1.0 voxel are applied along the z-direction, cor-
4.1. Factors that may affect the accuracy of applying DVSP to red
responding to a shift of 0.1 voxel between any two successive im-
sandstone
ages. The displacement elds are all calculated using a 64
voxel  64 voxel  64 voxel subset. The results are depicted in
4.1.1. The effect of the quality of structure
Fig. 14. It is noted that the accuracy and precision of the red sand-
When measuring 2D surface displacements, it is fairly easy to
stone are better than those of the medium sandstone. For a smaller
create a speckle pattern on the surface of a specimen using a variety
autocorrelation radius, there are more speckles contained in each of
of methods (Kozicki and Tejchman, 2007; Hall et al., 2010a, b;
the subsets, thus improving the performance of DVSP.
Dautriat et al., 2011; Nguyen et al., 2011; Lin and Labuz, 2013;
Zhang et al., 2013). However, creating a 3D volumetric speckle
pattern is a much more complicated endeavor. One can either 4.1.2. The effect of the size of subset
articially introduce markers into the studied material (Huang With different sizes of subset, different speckle information is
et al., 2011), or just use the materials internal microstructure it- embedded within the subset thus affecting the accuracy of the
self as the speckle pattern (Bay et al., 1999; Lenoir et al., 2007; Hall obtained results. As a demonstration, we crop a block with 200
et al., 2010a, b; Charalampidou et al., 2011, 2014). In this study, we voxel  200 voxel  200 voxel from the middle part of the volume
use the natural microstructure of red sandstone as the volumetric image of the red sandstone in step 1, and dene it as the reference
speckle pattern. This pattern is not an ideal speckle pattern, thus volume image. The deformed volume images are obtained by
errors might be introduced. imposing sub-voxel displacements from 0.1 voxel to 1.0 voxel along
In Fig. 13, the CT scanned cross-sections of red sandstone and the z-direction, corresponding to a shift of 0.1 voxel between any
medium sandstone are shown in the same magnication. For two successive images. These volume images with pre-imposed
assessing the difference of their structures, we compute the auto- sub-voxel translation are used to investigate the accuracy and
correlation radius r based on the autocorrelation function of the CT precision of DVSP with different subset sizes. The subset sizes are
144 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

0.06 0.06
Red sandstone
646464
0.04
0.04
323232
Medium sandstone 0.02 161616
Mean bias error (voxel)

Mean bias error (voxel)

0.02
0
0 0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1 -0.02

-0.02 -0.04

-0.04 -0.06

-0.08
-0.06
-0.1
-0.08
Subvoxel displacement (voxel) -0.12
Subvoxel displacement (voxel)
(a)
(a)
0.03
Red sandstone 0.045
0.025 0.04

Standard deviation (voxel)

646464
Standard deviation (voxel)

0.035
Medium sandstone 323232
0.02 0.03
161616
0.025
0.015 0.02
0.015
0.01
0.01
0.005 0.005
0
0 0.2 0.4 0.6 0.8 1
0
0 0.2 0.4 0.6 0.8 1 Subvoxel displacement (voxel)
Subvoxel displacement (voxel) (b)
(b)
Fig. 15. Comparisons among different sizes of subset. (a) Mean bias error, and (b)
Fig. 14. Comparisons of two different rock materials. (a) Mean bias error, and (b) Standard deviation.
Standard deviation.

To analyze the effects of the artifacts introduced by the scanner,

ACTIS 225/320 Industrial Computed Tomography System, we take a
selected as 16 voxel  16 voxel  16 voxel, 32 voxel  32 voxel  32 baseline test, in which two consecutive scans of the red sandstone
voxel and 64 voxel  64 voxel  64 voxel, respectively, and the specimen are taken, using identical settings and without moving
overlap between neighboring subsets is half of the subset size. (other than the tomographic rotation) or deforming the sample.
Fig. 15 shows the mean bias error and the standard deviation error. The volume image has the size of 900 voxel  900 voxel  600
It is noted that both the mean bias error and the standard deviation voxel, and the physical size of a voxel is 29 mm3. Ideally, the
error of DVSP depend on the size of the subset. As the subset size displacement elds computed by DVSP for this pair of volume
increases, both the accuracy and precision of DVSP are improved. images should be zero everywhere, but the above factors may affect
The calculated displacement using a large subset results in a the results. Table 2 shows the statistic errors. The measurement
smaller mean bias error as well as a smaller standard deviation uncertainty is 0.043 voxel. As can be seen, the displacements are
error. clearly greater than 0.043 voxel standard deviation. This means that
In the DVSP theory, the subset is assumed to be rigid. In reality the displacements represent an actual rigid motion along three
there is deformation and rotation of the material within the subset. directions, respectively, and are not really noise. This motion, which
While increasing the subset size results in better accuracy of DVSP, DVSP is able to capture, is likely to occur because of the tomography
the errors caused by ignoring the deformation and rotation of the rotation stage. In the compression experiment, the errors caused by
material within the subset tend to increase too. Thus the selection the above factors are quite small compared with the practical
of the proper subset size has to be judiciously taken. In practical deformation. Therefore, the measurement results are credible.
applications, we can use the above-described method of imposing Besides the above factors, the image contrast also inuences the
virtual displacements to determine the optimal parameters for measurements. In Mao et al. (2015), all these factors that would
DVSP.

Table 2
4.1.3. Baseline test
Errors caused by artifacts and imperfect motion.
CT slice images are reconstructed with the appropriate mathe-
matical algorithm from different angular radiographic projections. Displacement Minimum Maximum Mean error Standard
component error (voxel) error (voxel) (voxel) deviation (voxel)
The non-uniformity of detector elements, the polychromatic nature
of X-ray, the imperfect motion of the rotation stage, and the u 0.270 0.010 0.152 0.043
possible rigid body motion of the specimen will all contribute to v 0.130 0.010 0.059 0.030
w 0.027 0.250 0.149 0.033
different artifacts, which will affect the performance of DVSP.
 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146 145

inuence the performance of DVSP technique are analyzed in Experimental Mechanic Research of the Department of Mechanical
detail. Engineering at Stony Brook University. F.P. Chiang wishes to thank
Dr. Yapa Rajapakse, Director of the US Ofce of Naval Researchs
4.2. Computational efciency of DVSP Solid Mechanics Program for his support over the years for
advancement of the speckle technique.
DVSP and DVC originated from different concepts and evolved
differently as well. Compared with DVSP, DVC has only a slightly References
higher accuracy and precision considering the deformation or the
Adam J, Schreurs G, Klinkmuller M, Wieneke M. 2D/3D strain localization and fault
rotation in the subset, but it consumes much more calculation time.
simulation in analogue experiments: insights from X-ray computed tomogra-
For a search region of size n, which is equal to the subset size, the phy and tomographic image correlation. Bollettino di Geosica Teorica ed
computational complexity for searching using FFT is O2n3 log10 n, Applicata 2008;49(Supp. 2):21e2.
Adam J, Klinkmuller M, Schreurs G, Wieneke B. Quantitative 3D strain analysis in
whereas in the case of DVC the searching in the spatial domain is
analogue experiments simulating tectonic deformation: integration of X-ray
On6 , clearly demonstrating the computational efciency of the computed tomography and digital volume correlation techniques. Journal of
DVSP algorithm. This advantage is magnied when high-resolution Structural Geology 2013;55:127e49.
and high-volume images are processed as is the case in most Alshibli KA, Sture S, Costes NC, Frank ML, Lankton MR, Batiste SN, Swanson RA.
Assessment of localized deformation in sand using X-ray computed tomogra-
practical problems. phy. Geotechnical Testing Journal 2000;23(3):274e99.
In this study, DVSP is coded with MATLAB. The computation is Amitrano D. Rupture by damage accumulation in rocks. International Journal of
performed on HP workstation Z420 with Intel Xeon E5-1650 3.2 Fracture 2006;139(3e4):369e81.
Bay BK, Smith TS, Fyhie DP, Saad M. Digital volume correlation: three-dimensional
12 M 1600 6C CPU, 32 GB DDR3-1600 ECC RAM memory. The strain mapping using X-ray tomography. Experimental Mechanics 1999;39(3):
calculated volumetric image has a dimension of 566 voxel  566 217e26.
voxel  801 voxel, the size of subset is 64 voxel  64 voxel  64 Bsuelle P, Desruesb J, Raynaudc S. Experimental characterisation of the localisation
phenomenon inside a vosges sandstone in a triaxial cell. International Journal of
voxel and overlap is 32 voxels. It takes 9157 s to calculate overall Rock Mechanics and Mining Sciences 2000;37(8):1223e37.
images. For each subset, the calculation time is 1.49 s. If we would Charalampidou EM, Hall SA, Stanchits S. Characterization of shear and compaction
use the unoptimized DVC method with the size of subset being 51 bands in a porous sandstone deformed under triaxial compression. Tectono-
physics 2011;503(1e2):8e17.
voxel  51 voxel  51 voxel, overlap 25 voxels, and search area 17
Charalampidou EM, Hall SA, Stanchits S, Viggiani G, Lewis H. Shear-enhanced
voxel  17 voxel  17 voxel, it will takes 1,172,790 s to calculate. For compaction band identication at the laboratory scale using acoustic and full-
each subset, the calculation time would be 88.6 s. It is nearly 60 eld methods. International Journal of Rock Mechanics and Mining Sciences
2014;67:240e52.
times longer than using DVSP. Several approaches have been pro-
Chiang FP, Mao LT. Development of interior strain measurement techniques using
posed to reduce redundancy and improve the efciency of DVC (Pan random speckle patterns. Meccanica 2015;50(2):401e10.
et al., 2012; Huang et al., 2011; Gates et al., 2011), but the imple- Dautriat J, Bornert M, Gland N, Dimanov A, Raphanel J. Localized deformation
mentation complexity increases as well. induced by heterogeneities in porous carbonate analysed by multi-scale digital
image correlation. Tectonophysics 2011;503(1e2):100e16.
David MR. A simple autocorrelation algorithm for determining grain size from
5. Conclusions digital images of sediment. Journal of Sedimentary Research 2004;74(1):
160e5.
Desrues J, Chambon R, Mokni M, Mazerolle F. Void ratio evolution inside shear
In this study, X-ray scanning in situ is conducted to investigate bands in triaxial sand specimens studied by computed tomography. Geo-
the damage and failure process of red sandstone specimen under technique 1996;46(3):529e46.
Desrues J, Viggiani G. Strain localization in sand: an overview of the experi-
uniaxial compression tests. Analysis of 3D full-tensor strain elds,
mental results obtained in grenoble using stereophotogrammetry. Interna-
derived from the DVSP displacement elds, has revealed the tional Journal of Numerical Analytical Methods in Geomechanics 2004;28(4):
pattern of strain localization during loading, and then subsequent 279e321.
initiation of the crack inside the localized zone resulting in failure of Gates M, Lambros J, Heath MT. Towards high performance digital volume correla-
tion. Experimental Mechanics 2011;51(4):491e507.
specimen. Hall SA, Wood DM, Ibraim E, Viggiani G. Localised deformation patterning in 2D
The advantage of DVSP is its high computational efciency. In granular materials revealed by digital image correlation. Granular Matter
practical applications, millions of points would be calculated. The 2010a;12(1):1e14.
Hall SA, Bornert M, Desrues J, Pannier Y, Lenoir N, Viggiani G, Bsuelle P. Discrete
time-saving feature of DVSP is of considerable advantage. In this and continuum analysis of localised deformation in sand using X-ray mCT and
study, we just make a tentative attempt to the application of DVSP, volumetric digital image correlation. Geotechnique 2010b;60(5):315e22.
and there is much more to do to improve the performance. We Huang JY, Pan XC, Li SS. A digital volume correlation technique for 3D deformation
measurements of soft gels. International Journal of Applied Mechanics
foresee the technique to have wide applications in rock mechanics. 2011;3(2):335e54.
Kozicki J, Tejchman J. Experimental investigations of strain localization in concrete
using digital image correlation (DIC) technique. Archives of Hydro-Engineering
Conict of interest
and Environmental Mechanics 2007;54(1):3e24.
Lenoir N, Bornert M, Desrues J, Bsuelle P, Viggiani G. Volumetric digital image
The authors wish to conrm that there are no known conicts of correlation applied to X-ray microtomography images from triaxial compres-
interest associated with this publication and there has been no sion tests on argillaceous rock. Strain 2007;43(3):193e205.
Lin Q, Labuz JF. Fracture of sandstone characterized by digital image correla-
signicant nancial support for this work that could have inu- tion. International Journal of Rock Mechanics and Mining Sciences 2013;60:
enced its outcome. 235e45.
Louis L, Wong T, Baud P, Tembe S. Imaging strain localization by X-ray computed
tomography: discrete compaction bands in Diemelstadt sandstone. Journal of
Acknowledgments Structural Geology 2006;28(5):762e75.
Louis L, Wong T, Baud P. Imaging strain localization by X-ray radiography and digital
image correlation: deformation bands in Rothbach sandstone. Journal of
This work is nancially supported by National Basic Research
Structural Geology 2007;29(1):129e40.
Program of China (973 Program) (No. 2010CB732002), National Mao LT, Chiang FP, Yuan ZX. 3D displacement measurement in solid using digital
Natural Science Foundation of China (Nos. 51374211, 51374215), volumetric speckle photography and computer tomography. Acta Optica Sinica
National Key Foundation for Exploring Scientic Instrument of 2015;35(3):0312001.
Nguyen TL, Hall SA, Vacher P. Fracture mechanisms in soft rock: identication and
China (No. 2013YQ240803), Fundamental Research Funds for the quantication of evolving displacement discontinuities by extended digital
Central Universities (No. 2009QM02) and the Laboratory for image correlation. Tectonophysics 2011;503(1e2):117e28.
146 L. Mao et al. / Journal of Rock Mechanics and Geotechnical Engineering 7 (2015) 136e146

Pan B, Wu DF, Wang ZY. Internal displacement and strain measurement using Lingtao Mao, born in Shihezi, Xinjiang Uygur Autonomous
digital volume correlation: a least-squares framework. Measurement Science Region, China, is an Associate Professor at State Key Lab-
and Technology 2012;23(4):45002. oratory of Coal Resources and Safe Mining (China Univer-
Pan B, Asundi A, Xie HM, Gao JX. Digital image correlation using iterative least sity of Mining & Technology (Beijing)), and an Adjunct
squares and pointwise least squares for displacement eld and strain eld Research Associate Professor in the Department of Me-
measurements. Optical Lasers Engineering 2009;47(7e8):865e74. chanical Engineering at the State University of New York at
Schreier HW, Braasch JR, Sutton MA. Systematic errors in digital image correlation Stony Brook, U.S.A. He got the BE at Jilin University of
caused by intensity interpolation. Optical Engineering 2000;39(11):2915e21. Technology, and the M.Sc. and Ph.D. in Engineering Me-
Suzanne R, Dominique NT, Jacques D, Mazerolle F. Brittle-to-ductile transition in chanics at China University of Mining & Technology (Bei-
Beaucaire marl from triaxial tests under the CT-scanner. International Journal of jing) in 2002 and 2005, respectively. From June 2012 to
Rock Mechanics and Mining Sciences 2008;45(5):653e71. May 2013, he joined the Department of Mechanical Engi-
Viggiani G, Lenoir N, Bsuelle P, Michiel MD, Marello S, Desrues J, Kretzschmer M. X- neering at the State University of New York at Stony Brook,
ray microtomography for studying localized deformation in ne-grained geo- U.S.A, as a visiting scholar. His current research interests
materials under triaxial compression. Comptes Rendus Mcanique include the development and application of photo-
2004;332(10):819e26. mechanics on rocks and composites. He is the author or co-author of more than fty
Zhang H, Huang GY, Song HP, Kang YL. Experimental characterization of strain scientic papers and ten patents.
localization in rock. Geophysical Journal International 2013;194(3):1554e8.