Engineering Structures 86 (2015) 1–15

Contents lists available at ScienceDirect

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

Method based on digital image correlation for damage assessment

in masonry structures
Huu-Luyen Nghiem a,b,⇑, Marwan Al Heib b,1, Fabrice Emeriault a,2
a
Univ. Grenoble Alpes, 3SR, F-38000 Grenoble, France
b
INERIS Parc technologique Alata, 60550 Verneuil-En-Halatte, France

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

Article history: This paper presents a new approach based on the digital image correlation (DIC) technique for evaluating
Received 8 May 2014 damage in masonry structures on the basis of the opening of joints. An application is proposed on
Revised 29 October 2014 experiments that are performed using a large-scale physical model reproducing both the soil-structure
Accepted 9 December 2014
interaction and the masonry structure. Displacement fields of blocks obtained by DIC analysis are
Available online 7 January 2015
described in a new Discrete Element System (DES) where cracks appear near joints with openings
between blocks. A damage indicator associated with the total length of cracks is proposed for a complete
Keywords:
damage evaluation. Monte Carlo simulations are used for estimating the confidence interval of this indi-
Masonry structure
Physical modelling
cator. Recommendations are suggested for the assessment of damage in masonry structures due to
Digital image correlation underground excavations or soil subsidence.
Crack identification Ó 2014 Elsevier Ltd. All rights reserved.
Monte Carlo simulation

1. Introduction
The monitoring of crack propagation in physical modelling by
conventional means (such as a crackmeter) suffers in situations
with complex cracks. This can be overcome by using the digital
image correlation (DIC) method, which provides full-field displace-
ment measurements. In particular, for brittle materials, many algo-
rithms proposed in the literature are inspired by the eXtended
Finite Element Method (X-FEM) [1]. We can cite eXtended Digital
Image Correlation (X-DIC) based on a finite element decomposition
of displacement fields [2–4]. Another approach uses Integrated
Digital Image Correlation (I-DIC), which consists in the use of the
full description of displacement fields such as Williams’ series
and integrates Stress Intensity Factors (SIF) in a crack propagation
law [5]. These two approaches are useful for the investigation of
the development of arbitrary cracks. The main disadvantage of
these approaches, however, is the necessity of direct modifications
in the correlation algorithm. Consequently, this is not easy for the
users of commercial DIC software, who are becoming more and
more numerous. Actually, a few applications have been developed
that are based on DIC output in order to study crack propagation.
⇑ Corresponding author at: Univ. Grenoble Alpes, 3SR, F-38000 Grenoble, France.
Tel.: +33 3 44 61 82 39.
E-mail addresses: huuluyen.nghiem@3sr-grenoble.fr (H.-L. Nghiem), marwan.
alheib@ineris.fr (M. Al Heib), fabrice.emeriault@grenoble-inp.fr (F. Emeriault).
1
Tel.: +33 3 55 66 28 82.
2
Tel.: +33 4 56 52 86 47.
We can cite the DIC-F model [6], which was developed for the frac-
ture investigation of soft rock.
Masonry is considered as a brittle structure, and crack propaga-
tion can be therefore predicted using the approaches mentioned
above, but this can be simplified by considering that the cracks
only appear near joints. In fact, the failure mode could be of the I
type (failure by traction) or II type (failure by shear), and the failure
mode in compression rarely appears in reality [7]. Using the DIC
technique, the difference between Mode I and Mode II can be pro-
vided with respect to Von Mises strain and their high values can
indicate the location of cracks [8,9]. Nevertheless, crack identifica-
tion is insufficient for a full damage assessment, which must take
four important properties into account: crack location, length,
width and orientation. To begin with, the approaches mentioned
above need to be improved in order to take crack width into
account.
In this paper, we develop a new eXtended Digital Image Corre-
lation method dedicated to the physical modelling of masonry
structure in order to assess damage. This research attempts to
tackle crack identification by integrating the specific aspect of
crack width. The new point of view here for crack propagation is
the use of physical modelling combined with an experimental
damage criterion based on DIC technique. The novel key is the
decomposition of displacement fields in DIC into rotation and
translation components for each block. Consequently, the blocks
move independently as discrete elements, so cracks appear at their
interfaces. In particular, this investigation provides a specific

http://dx.doi.org/10.1016/j.engstruct.2014.12.021
0141-0296/Ó 2014 Elsevier Ltd. All rights reserved.
2 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
indicator referring to the total length of cracks with the purpose of
damage assessment. The uncertainty/accuracy of this indicator is
also analyzed using Monte Carlo simulation.
Our work also extends to applications for damage assessment of
masonry structure due to soil settlement and ground subsidence
using the physical model developed (Fig. 1). In Fig. 1a, D is the
depth of the void (cavity), O is the layer opening, and Wc is the
width of the cavity. The structural model presented in Fig. 1b asso-
ciates a masonry structure made of block in contact with a raft
foundation made of silicon (interface C1) resting on an analogue
soil made of sand (interface C2). The first interface C1 is a perfect
collage for an easier setup of the model structure on the soil sur-
face. In contrast, the second interface C2 is a simple frictional con-
tact between the silicon foundation and the sand.
Conventional methods for damage assessment refer to different
criteria, such as tensile limit strain [10], limit slope (see [11]), and
crack width [12]. In most cases, the building is considered as an
elastic continuous beam with the same dimensions. The damage
indicators based on these criteria can refer to the angular distor-
tion [10] or the deflexion ratio [12], with the soil-structure interac-
tion being represented by the relative stiffness [13,14] or the final
structural form [15]. Such indicators allow damage to be quantified
but are insufficient for determining the location of the damage on
the structure. As a consequence, they often over- or under-esti-
mate the true potential for structural damage. Recent physical
models have been developed in order to observe crack propagation
[16,17]. However, investigations are limited to the analysis of crack
location and do not provide for a quantification of damage. Here,
the proposed method can overcome this drawback, enabling both
crack identification and quantification.
This paper is organized into three main sections. In Section 2,
we first propose a way to quantify DIC measurement errors for fur-
ther consideration in the evaluation of the different strategies for
damage assessment. The reconstruction of the blocks, discussed
in Section 3, is concerned with the reconstruction of motion based
on DIC outputs. Here, we consider the blocks to be rigid bodies and
the centers of rotation to be their centers of gravity. The last
hypothesis is applicable when the difference between calculated
and experimental displacement is less than 5%. Thus, a cost func-
tion is introduced to ensure the accuracy of the model reconstruc-
tion. In this section, we also introduce a damage indicator, which is
useful for assessing damage in masonry structures. The effects of
measurement errors are determined using the Monte Carlo
method, which is a powerful tool for the simulation of uncertainty
in complex systems [18]. Section 4 contains two illustrations of the
applications of the proposed method for damage assessment of
masonry structures in the particular case of ground movements
induced by underground excavations. A comparison with conven-
tional indicators is also included.
2. DIC measurement errors
DIC is a non-contacting method for measuring displacement
using video cameras to record images of the surface of objects. This
technique was used as early as 1975 [19] and has increasingly
gained consideration since the 1980s [20]. Nowadays, this tech-
nique is used in a wide range of disciplines, especially in mechan-
ical tests [21]. The basic idea of the method is the matching of one
point on the reference image with a corresponding point of the
deformed image. Depending on the number of cameras, different
strategies can be employed: (i) 2D version with only one camera
and the motion of specimen must be in plane, and (ii) 3D version
using two cameras, which allows for the measurement of three-
dimensional displacements [22,23].
Although DIC is a powerful technique for mechanical tests on
materials and structures, the measurement errors are not
discussed sufficiently and difficult to improve in practice. The
sources of errors are numerous and can be divided into two main
categories: quality of the measurement devices and working envi-
ronment, and correlation algorithm [24]. The first category is asso-
ciated with the materiel (e.g., mire, optical lens), and the working
environment such as epipolar constraint, process of calibration,
lighting, image contrast and the presence of out-of-plane displace-
ments. The second category concerns aspects of the correlation
algorithm, such as subset size, speckle pattern and correlation
algorithms.
Many studies focus only on optimizing algorithms and neglect
the other source errors. Practically speaking, improving the optimi-
zation algorithm is not possible for scientists or engineers using
commercial software. Few studies show how to integrate DIC mea-
surement errors into the experimental results. One way is through
the use of rigid body motions [24], but this is not applicable to brit-
tle materials because of the associated failure phenomena. Another
approach is to analyse two images of a specimen in consecutive
instants t and t + 1 while the specimen is loaded. Considering slow
variations in load, DIC results can be considered as measurement
errors. However, this approach is difficult to employ in the case
of rapid loading or a highly stiff specimen.
For brittle materials such as concrete and masonry, it is possible
to take the measurement errors into account using the DIC tech-
nique. Here, we adopt a static approach, i.e., analysing images in
static state [25]. In the initial state of the specimen, there are the-
oretically no displacements but residual values exist after DIC anal-
ysis because of working environment and optimization processes
in algorithms. Therefore, these values can be considered as system-
atic measurement errors. Briefly, once the test set-up is realized
and before the beginning of the test, static images (5–10 images)
are recorded. These images (except the reference image) are dupli-
cated and renamed. The copy-rename process is repeated until the
desired number of images. The number of static images should be
equal to the number of real images in the test, as is discussed in the
following paragraphs. The last step is to analyze all these images
with the DIC software: since there is no real movement between
the images, the obtained displacements are the measurement
errors. Another way to obtain the desired number of static images
consists in recording images continuously until the desired num-
ber. These two solutions give the same results. The static approach
has numerous advantages: (i) This is easy in practice and can be
applicable for all material. (ii) We can multiple images as we wish
in order to investigate the performance of algorithms. In particular,
depending on the algorithm, the displacement errors can increase
with the number of images, and a loss of correlation can occur with
a sufficient large number of images, as is discussed in the following
paragraphs.
The performance of the static approach is illustrated through an
example with the DIC commercial software VIC-3D [26]. Here, we
analyse 1000 static images of a small-scale model of the masonry
structure described in Section 4. Each static image has more than
3  104 points, and the DIC analysis provides the coordinates, dis-
placements, strains, etc. Here, we focus on the horizontal and ver-
tical displacements, because the conclusions will be used directly
for the error analysis in the crack width development in the next
sections. Other information can be determined by the same
process.
Taking the measurement errors into account actually corre-
sponds to a determination of a probability distribution which
matches well with the DIC outputs. Distribution functions can vary
depending on the choice of the strategy for the DIC analysis, such
as itineration procedure or optimization options [27]. In particular,
the itineration procedure of images is the most important. For VIC-
3D, the itineration choice can be: (i) the default option in which
each image is compared to the reference image, or (ii) the
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15 3

incremental option in which each image is compared to the previ-
ous image. Fig. 2 represents, for a set of 100 static images, the dif-
ferent histograms of the horizontal and vertical displacements
according to different choices of itineration, the distribution being
fitted in each case with the normal density function. The results
show that the standard deviation values are much more significant
than the mean values, because the mean values are around zero
and can therefore be neglected.
The first histograms (Fig. 2a and d) correspond to the compari-
son of image 1 and the reference image 0 and can be considered as
the reference results. In the subsequent graphs in Fig. 2, a compar-
ison is made between the distributions of displacement obtained
for the image 100 depending on the itineration process. For the
default itineration, the outputs provide similar results to the image
1, i.e., the histograms of Fig. 2b and e are similar to those of Fig. 2a
and d. Besides, they can be approximated well using the normal
density function. On the other hand, data obtained using incremen-
tal itineration (100 steps) are sharp and not well captured by the
normal density function (Fig. 2c and f). However, the latter will still
be used to fit data in practice because it appears that the function is

Fig. 1. INERIS physical model for soil-structure interaction. (a) Description of phenomena. (b) 2D cross section. (c) Geometry configuration of stereo system. (d) Positions of
cameras. (e) Cameras monitoring the displacement field of soil. (f) Cameras monitoring the displacement field of masonry wall. (g) Small-scaled physical model of masonry
structure. (h) Position of jack.
4 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
Fig. 2. Histogram of (a–c) horizontal and (d–f) vertical displacements according to the default and incremental options. u,
the number of points.
larger than the experimental histogram (corresponding to an over-
estimation of the standard deviation of the measurement errors,
given that the mean values are almost identical).
Since the mean values are negligible, the difference between
two options is shown by the evolution of the standard deviators.
To simplify comprehension, we define a score as follows:
ri
g¼ ð1Þ
r1
Fig. 3. Standard deviators evolution of (a) horizontal and (b) vertical displacement.
v are horizontal and vertical displacements. np is
where ri is the deviator of image i. The first deviator r1 has the
same value in both the default and incremental options due to
the comparison of the same reference and successive images. Con-
sequently, the value of r1 is considered as a reference in Eq. (1). The
evolution of g (Fig. 3) shows that the standard deviators are station-
ary with the default option and increase in the case of the incre-
mental option. In practice, the number of static images used for
the estimation of measurement errors should be equivalent to that
of deformed images corresponding to the whole duration of the
mechanical test, and the correlation itineration should be identical
for both static and deformed images. For example, if we take 100
images during the loading part of the test and perform the analysis
using the incremental correlation option, we should choose 100 sta-
tic images with the incremental correlation option to determine the
characteristics of the measurement error distribution.
The default option clearly demonstrates better results with
g  1 in the case of static image analysis, but is difficult to apply
to the real analysis with deformed images. In fact, deformed
images have problems of loss of correlation in the vicinity of the
crack when the cracks appear [28], or when the deformation
between two images is greater than 20% [29]. The incremental
option avoids this drawback and provides better overall results.
However, this option reaches its limit after a certain number of
increments due to the increasing standard deviators of the mea-
surement error. In fact, the DIC technique is no more accurate
when the errors are getting close to the real displacements. With
this option, the standard deviators can be estimated with the
trends in Fig. 3. The standard deviator is multiplied approximately
by 2.5 after 100 images and by 20 after 1000 images. Otherwise,
the DIC outputs do not respect the law of large numbers due to
the cumulative errors when the incremental correlation is used.
3. Methodology
3.1. Reconstruction of masonry
From the point of view of ‘‘standard’’ DIC, an image is described
as a continuum. However, the masonry structure is usually
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
Fig. 4. Geometry reconstruction of blocks. (a) AOI in DIC software. (b) New coordinate system. (c) Blocks in detail.
considered as a discrete mechanical system. To make a connection
between the two descriptions, we can locally break down the dis-
placement fields obtained with continuum description into two
components: rotation and translation. Consequently, the individ-
ual blocks constituting the masonry can be described as discrete
elements and be independent of each other.
3.1.1. Geometry
Due to the loss of spatial resolution during DIC analysis, the full
geometry of blocks at their boundaries is not well known. The idea
is to create an equivalent system with masonry blocks having the
same coordinates as those provided by DIC software. To do this,
an equivalent masonry wall is firstly built, based on the designed
dimensions of the blocks (14  7 mm2) and the wall
(238  63 mm2). The key information is the coordinates of the
boundary of each block. Then, these coordinates are confronted
visually with those provided by DIC software via the calibration
of a translation operator using a marked point, see Fig. 4a and b
(during the installation of the structure, a particular attention
has been paid to the levelling of the structure, the ‘‘horizontal’’
direction is therefore well determined). In this way, interfaces
between blocks are also identified in the image and can be super-
imposed with the equivalent system in order to mesh the inner
part of each block. Real blocks have small differences in dimen-
sions with the theoretical blocks due to manufacturing tolerances.
In addition, even when the structure is built carefully, the initial
arrangement of blocks is not ‘‘perfect’’ and small gaps can exist
between them. It shall also not be forgotten that the translation
operator used for the marked point can be affected by an error
5
(even though it is small compared to the other sources). Therefore,
when points are sought inside each real block on the image to
define the reference mesh, a reduced area of the block is consid-
ered in order to optimize the accuracy of the results. For example
in this study, the blocks have dimensions of 14  7 mm2, but the
seeking area is 13.4  6.4 mm2 (corresponding to 87.5% the surface
of block). For the masonry wall in this study, the typical block has
about 150 measurement points (Fig. 4c). In Fig. 4c, the meshed area
covers from 85% to 90% of the total surface of each block.
3.1.2. Displacement in plane
Once the geometry of blocks is reconstructed, we transform dis-
placement fields obtained with a continuous description into a dis-
crete description, in which displacements are broken down into
two components: rotation and translation. For the continuous
description, we consider an arbitrary block X0 at the initial time
t0 = 0 and the current configuration Xt at the time t. The displace-
ment of a material point is described by the application
u : ðX0  RÞ ! Xt , which is the transformation of the point
X 2 X0 at time t0 into the point x ¼ uðX; tÞ 2 Xt at time t 2 ½0; T.
The displacement of the point X at time t is denoted by:
uðX; tÞ ¼ uðX; tÞ  X. In the Lagrangian description, the transfor-
mation gradient is a fourth-order tensor defined via F ¼ ru. Here,
we use: (i) the right Cauchy–Green tensor as the deformation
gradient C = FT  F, and (ii) the Green–Lagrange tensor for strain
E = (C  1)/2.
To connect the above information with the discrete description,
we present first the strain tensor in the spectral form: C = Rki2ni
ni, where ki are the eigenvalues of C and ni are the eigenvectors
6 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15

(i = 1, 2 for the case of in-plane displacement). Then, the tensor F is

presented in the polar decomposition F = R  U, where R denotes
the pure rotation tensor and U is the pure deformation tensor.
The eigenvalues of U are the square roots of C, i.e., U = C1/2, and
the tensor R is orthogonal, i.e., R  RT = RT  R = 1. As a result, the
pure rotation tensor is determined as follows: R = F  U1. In the
case of rigid block, the pure deformation tensor is equal to the unit
tensor, i.e., U = 1. Consequently, the rotation tensor is determined
as follows:
R¼F ð2Þ
The translation of point X of a block is expressed by the follow-
ing representation:
c ¼ x  R  X  G0 ð3Þ
where G0 is the center of rotation. The delicate point is the calcula-
tion of the translation term in Eq. (3). Theoretically, this can be
identified when two rotations of block are known. If the loading is
slowing or varying during the test, we can consider that centers
of rotation are identical at times t and t + 1 for the purpose of deter-
mining the translation term. However, data usually have noise, so
centers of rotation are no longer accurate. This can be overcome
by considering G0 as the center of gravity. Afterward, a cost function
can be introduced in order to optimize displacements.
3.1.3. Numerical implementation
The calculation is based on the standard finite element method
with the use of finite element triangulation. According to this
method, the displacement gradient can be described by: ru = B  q,
where B is the matrix shape functions and q is the vector of node
displacements. The associated transformation gradient is pre-
sented in the form of F = 1 + ru. Next, the rotation and translation
terms can be determined by Eqs. (2) and (3), in which these values
are in the center of gravity of triangle.
3.1.4. Cost function
Associated with the rotation and the translation of block, we
introduce the new variable:
d :¼ fa; cg ð4Þ
where a is the angular rotation and c is the translation vector of the
block. The solution to minimization problem of the cost function is
determined by:
d ¼ inf d2W JðdÞ ð5Þ
3
where W ¼ fdjd ¼ ða; cÞ 2 R g.
A simple form of the cost function can be chosen by:
X 2
JðdÞ ¼ ðu  uexp Þ ð6Þ
Box 1. Numerical scheme for the discrete element model.
Initialization: The data obtained by DIC analysis: coordi-
nates (x, y) and displacements u = (u, v) of points are
known for each block.
Process:
1. Reconstruction of geometry
2. Spatial discretization
ru = B  q
3. Compute transformation gradient, right Cauchy–
Lagrange deformation, and pure deformation.
F = 1 + ru
C = FT  F
C = Rka2nana
U = C1/2
4. Compute rotation & translation
R = F (rigid block)
R = F  U1 (deformable block)
c = x  R  X  G0
G0 refers to the coordinates of center of blocks
5. Minimization of cost function
P
d ¼ inf d2W JðdÞ e.g. JðdÞ ¼ ðu  uexp Þ2
Using a Levenberg–MarguardtPalgorithm P with the
a c
initial values: d :¼ fa; cg; a0 ¼ n i ; c0 ¼ n i
For rigid blocks, the values for rotation and translation are com-
mon to all finite elements of the block being considered in the dis-
cretization step. Besides, the procedure can be applied to
deformable blocks in considering that the finite elements are inde-
pendent. In this case, the minimization of the cost function will be
more complicated, because the rotation and the translation will be
different for the finite elements of the block being considered.
3.2. Introduction of the damage indicators
3.2.1. Damage criteria
Table 1 presents the damage classification based on crack width
as suggested by Burland [12]. Three first classes D0, D1, and D2,
ranging from ‘‘negligible’’ to ‘‘slight’’ damage classes, correspond
to aesthetic damage of the buildings. Class D3 and D4, ranging from
‘‘moderate’’ to ‘‘severe’’ damage, relate to functional damage and
affect serviceability. Class D5 involves structural damage that
affects the stability of the structure due to the development of
large and numerous cracks. For the sake of convenience, we com-
bine class D4 and D5 into a single class D4&5 corresponding to lim-
iting tensile strain greater than 0.3%.

The problem can be solved by using conventional schemes such 3.2.2. Total length of cracks
as the Newtonian methods. In this project, we used the Levenberg– The above reconstruction step provides the full information of
Marguardt algorithm with the initial values as follows: blocks such as coordinates and displacements. In this application,
we propose a simple algorithm for quantifying the total length of
P P
ai ci
a0 ¼ ; c0 ¼ ð7Þ
n n
Table 1
where n is the number of triangles in the block being considered. Damage classification system [1].

The initial values a0 and c0 are average values of rotation and trans- Id Damage class Crack width (mm)
lation in the block. D0 Negligible 0–0.1 mm
The criterion for stopping the cost function in Eq. (5) is chosen D1 Very slight 0.1–1 mm
in order to satisfy an at least 95% correspondence between model D2 Slight 1–5 mm
and experimental displacements. Box 1 summarizes the numerical D3 Moderate 5–15 mm or a number of cracks >3 mm
D4 Severe 15–25 mm, but also depends on number of cracks
scheme for the discrete element model dedicated to the masonry
D5 Very severe >25 mm, but depends on numbers of cracks
structure using the DIC outputs.
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15 7

Fig. 5. (a) Interface between two blocks and (b) definition of crack length for each point on the interface. h is the size of grid.

Fig. 6. (a) Histogram and (b) probability of ‘‘fictitious’’ crack widths.

cracks. In the first step, we determine differential normal displace-
ments for points on both sides of an interface (joint between
blocks) as follows:
D u ¼ u 1  n 1 þ u 2  n2
where u1, u2 are displacement vectors of two corresponding nodes
on both sides of an interface and n1, n2 are their normal vectors,
respectively (Fig. 5a). A negative value for Du corresponds to the
opening of the interface and an increase in width. Each value for
Du, associated with a point on the interface, is distributed in the
classes defined in Table 1.
In the second step, the total crack length for each category of
damage is considered as the damage indicator. For each damage
class Di, a non-dimensional definition of this indicator is given as
follows:
P number of images to take is limited up to 100 images (Fig. 6e),
lDi
LDi ¼
L0
P
where lDi is the total crack length of class Di and L0 is the total
length of joints (both vertical and horizontal). The total crack length
is also proportional to the number of points. Fig. 5b defines the dis-
crete length for an interior point and for a point on the boundary (h
is the size of grid).
Considering the precision of measurement with DIC, and
according to the static approach in Section 2, the static images of
specimen at the beginning of the test are a way to evaluate mea-
surement ‘‘noise.’’ Hereafter, we will continue analysing the limits
of the incremental option for the crack width measurement with
respect to the damage classification in Table 1. Noise is responsible
for the distribution of ‘‘fictitious’’ crack width and increases as a
function of the number of images analyzed in an incremental
option.
ð8Þ
In this investigation, we use the small-scale masonry structure
with a 1/40 factor on geometry and take 100 deformed images for
the tests. The distributions of Du on the prototype scale, i.e., the
displacements obtained from DIC multiplied by 40, are presented
in Fig. 6a–d, and their cumulative probability is presented in
Fig. 6e. We observe that for more than 95% of cases, Du is less than
1 mm (see Fig. 6e) when the DIC is applied incrementally between
image 1 and 100. This means that classes D0 and D1 cannot be dis-
tinguished accurately, so we combine the two first classes into a
common class, denoted as class D0&1. In this paper, we accept the
DIC errors are up to 1 mm in the crack width. Consequently, the
ð9Þ and the crack analysis in class D0&1 should not be taken into
account.
3.3. Error simulation
With the purpose of studying errors of total crack lengths in Eq.
(9), Monte Carlo simulation is used according to the variation of
the ‘‘fictitious’’ crack width Du. The first solution would be to use
the histogram in Fig. 6, but it is difficult in practice because its form
is arbitrary. This can be overcome by directly introducing displace-
ment errors. To do that, in the first step, we analyse images accord-
ing to a static approach in order to determine the horizontal and
8 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
vertical displacement errors. The error value is denoted by dd, and
the corresponding histogram is illustrated in Fig. 3. In particular,
we can consider that each point of the image has the same distri-
bution law with common values for the mean and the standard
deviator. Also, this distribution law is applied for each point on
the joint levels. According to the determination of the crack width
for each point on the interface in Eq. (8), the error of Du should
take the displacements of two blocks u1 and u2 into account, which
have identical variations. Therefore, the error of Du in Eq. (8) is cal-
culated as follows:
pffiffiffi
dDu ¼ 2dd ð10Þ
The distribution law of Du is considered as a normal distribution
andpsimilar
ffiffiffi to that of dd, but the standard deviator is multiplied
by 2. In particular, the distribution law of Du has a slight differ-
ence between the horizontal and vertical joints.
In the second step, we use the distribution law of Du in order
to simulate the total lengths of cracks given in Eq. (9). In order to
simplify the calculation and reduce the calculation time, the error
simulation is taken with the maximum values of the static
images, i.e., the static image having the maximum errors is
applied for all deformed images. In addition, the horizontal and
vertical displacements can be considered independently despite
the existence of an intrinsic correlation therebetween. Otherwise,
the application of the parallel algorithm is aptly suggested for
quicker calculation.
Fig. 7 illustrates the results after 106 simulations for an image
of approximately 3  103 points on interface. The first three
damage classes can be fitted with the normal distribution func-
tion. Due to the small value for L⁄D, this is not possible for the
last class D4&5. The difference between the maximum and mini-
mum values of L⁄D is approximately 3%. However, the respective
relative values are significantly different for the various damage
classes.
Fig. 7. Histogram of damage indicators for a structure in the hogging zone in the
final state. (a and b) Aesthetic damage. (c and d) Functional or structural damage.
4. Application
4.1. Damage assessment using a small-scale physical model
4.1.1. Physical model
The use of physical modelling in research combined with DIC
technique is becoming widespread. Thanks to the non-destructive
DIC method, physical models are becoming more and more attrac-
tive. The main advantage of this approach lies in enabling model
complex systems to yield accurate results. The present physical
model has been developed to study the soil-structure interaction
due to underground excavations or soil subsidence [30–32].
Fig. 1b and c presents the configuration of the physical model with
a 1/40 scale factor. This model is equivalent to the prototype of a
house with dimensions of 10  10 m2. The choice of materials is
as follows:
(i) The analogical soil is Fontainebleau sand (SiO2 > 98%), well-
known to researchers in physical modelling in geotechnical
engineering [33]. This sand is very smooth, and a category
NE 34 sand was chosen (see Appendix A).
(ii) The analogical structure is a small model made of wooden
pieces (see Appendix B). A comparison of different models
(polycarbonate slab, silicon slab, sugar blocks, wooden
blocks) is presented in a recent paper [32]. The masonry
walls have neither windows nor mortar and constitute six
horizontal layers of blocks with a corner connection
obtained by the odd superposition of layers.
The cameras set-up is illustrated in Fig. 1c–f. The displacements
of the soil surface and of the facade wall are monitored by a system
of 4 cameras in stereo configuration (Fig. 1c and d). We recall that a
stereo configuration needs two cameras disposed in space with
restraint geometry, see Fig. 1c. This geometry condition is
expressed by 3 parameters: Angle (A), Distance (D), and Base (B).
The angle A is the stereo angle. This angle should be neither too
small to avoid significant errors in the out-of-plan displacements
nor too large to avoid an important distortion of images [27]. The
recommended value of the angle A is between 20° and 30°
[26,27]. The distance D is the distance from the cameras to the
specimen. The value of D should be neither too small to have a
sufficient depth of the field of cameras (zone of sharpness of
camera) nor too large to avoid an important size for a pixel in
mm, i.e., the image of the specimen becomes small when the dis-
tance D increases. For this surface of the wall of dimensions
250  103 mm2, the recommended value of D is between 80 and
120 cm. The Base B, which is the distance between two cameras,
is the last value to be determined. According to Fig. 1c, the value
of B is calculated through the stereo angle A and the distance D
such as: B = 2D tan(A/2). For A = 20–30° and D = 80–120 cm, the
distance between the two cameras, B, varies from 28 to 64 cm.
The retained values for two following examples (two cameras
monitoring the wall) are: the stereo angle A = 22°, the distance
between the cameras and the model D = 91.1 cm, and the distance
between two cameras B = 35.4 cm. With this configuration, the
typical magnification factor is 0.35 mm/pixel.
4.1.2. Test procedure
The test procedure can be summarized in three main steps: (i)
The platform is first filled with a homogeneous layer of Fontaine-
bleau sand (a specific procedure has been defined in order to
obtain a uniform density over the 0.30 m thick layer). (ii) The sub-
sidence is reproduced using a mechanical–electrical jack placed at
the lower boundary of the tank at a sufficiently low speed
(0.15 mm/s). The displacements of the ground surface and of the
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15 9

structure are captured by four rapid high-resolution cameras. (iii)

Images are analyzed using the software VIC-3D in order to deter-
mine the displacement fields.
In this study, the image digitization uses 8-bit encoding sensors,
meaning that a pixel can take any value of 28 gray levels (from 0 to
255). According to [34], the correlation algorithm will not detect
displacements smaller than 28 pixel. The cameras in this study
use sensors 7.4 lm/pixel, dimensions 2048  2048 pixel2, there-
fore, the minimal value detectable of displacements is 0.03 lm.
A particular attention is paid to the choice of parameters in the
DIC software. To study the displacement field solely, two parame-
ters are essential in the DIC method: the subset size or Zone Of
Interest (ZOI), and the step or the separation [25,34]. These param-
eters are chosen independently. Concerning the subset size choice,
the user objective is to obtain the maximum number of indepen-
dent displacement measurement, the spatial resolution of dis-
placement should thus be minimum, consequently, the subset
size should be the lowest [34]. However, the displacement errors
increase when the subset size decreases [25]. Therefore, the choice
of the subset size is the compromise between the spatial resolution
and the measurement uncertainty. The number of steps is chosen
with respect to the time of calculation, and has no influence on
the displacement measurements [25]. For information, the recom-
mended number of points to analyze should be inferior to
5  104 points.

4.2. Experimental results

We will now demonstrate the performance of the proposed

approach using two examples. The damage indicator is outlined
in Eq. (9) based on the reconstruction of blocks as discrete dis-
placements in Box 1. The two following illustrations represent
the response of the masonry structure in the hogging zone, where
the slope of the settlement trough is at its maximum (Fig. 8a), and
the sagging zone, where the vertical displacement is at its maxi-
mum (Fig. 8b). The structural positions are captured by two cam-
eras oriented in a vertical direction (perpendicular to the soil
surface). These cameras only enable analysis of soil displacements
at the ground surface. These results are not discussed in this paper.
Two additional cameras are installed in the horizontal direction in
order to monitor the displacement fields of one of the structure
walls. Unfortunately, a lack of equipment and the limited dimen-
sions of the platform did not allow for the observance of all walls.
The current model of the masonry structure cannot be built
with perfect contact conditions between all the blocks due to the
shape of the wooden pieces and the absence of mortar. Small dif-
ferences in size and slight deviations from theoretically perpendic-
ular faces of a block lead to an initial structure where joints can be
opened from the outset. In addition, what we see is only the appar-
ent face of the wall, and there may have variable defects of contact
in the direction of the depth of the wall for which no information is
accessible by the cameras. Consequently, the development of the
subsidence trough can lead to a partial closure of the initially
opened joints. We can eliminate this defect by taking only the posi-
tive value for Du in Eq. (9) into account.
The proposed damage indicator related to the total length of
cracks is clearly not enough to conclude on the damage level of
the whole structure, because this depends also on the crack posi-
tions. Therefore, the damage assessment should be based on both
the damage quantification (crack length) and the orientation of
cracks (failure mode). In particular, the failure modes can be based
on two principal approaches. The first approach, or global point of
view, relates to failure modes of an equivalent beam, in which two
failure modes can be occured by flexion, and by shear [15]. The sec-
ond approach, or local point of view, concerns the failure of the I
type (failure by traction) or II type (failure by shear) [7].
Fig. 8. Configuration of physical model in the hogging zone (a) and the sagging zone
(b).
4.2.1. Structure in the hogging zone
The configuration of the structure in the hogging zone corre-
sponds to the physical model in Fig. 8a. As mentioned above, two
cameras in the horizontal direction (cameras 3 and 4) capture
the displacements of the selected façade. Therefore, the Area Of
Interest (AOI) for the DIC analysis is equal to the size of the
observed wall, i.e., 238  63 mm (see Figs. 1b and 4a). In VIC-3D,
the best subset size for minimizing the loss of correlation is 17 pix-
els. For the purpose of reducing calculation time but retaining the
maximum possible amount of information, the number of points
for each image should not exceed 5  104 points. Here, the number
of steps is equal to 2, corresponding to 30,488 points per image. In
order to determine the damage indicator, a regular grid is gener-
ated; its dimensions are 238  63 mm with size h = 1 mm, equal
to 15,296 points, but only the points on the interfaces are of inter-
est, because it is those that will be used to determine the crack
width. Here, the total length of the joints is L0 = 2940 mm, contain-
ing up to approximately 3  103 points. In keeping with the above
discussion of error measurement, we took 100 static images and
used the incremental correlation, which produces 95% fictitious
cracks with an opening smaller than 1 mm (acceptable threshold
in this study). The results show that the horizontal displacement
error is du = (0.6 ± 6.2)  103 mm, and the vertical displacement
error is dv = (0.2 ± 5.6)  103 mm.
We first focus on crack location. Fig. 9c presents the Lagrangian
strain distribution in the horizontal direction, which is provided by
the VIC-3D software. As a result, the bias error of strain value is less
10 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15

than 1.5  102% inside of the blocks. The calculated strain cannot
be represented in blocks because of its very high rigidity from
16  103 to 19  103 GPa (Appendix B), while the load is very
low during the test (in which the stress on the blocks simply
results in a deflection of the small model under the ground move-
ments). To overcome this drawback, the reconstruction of blocks is
required in Box 1. Because blocks are considered as rigid bodies,
cracks appear only at joint levels as shown in Fig. 9d. The location
of damage classes is shown in Fig. 9e. In this figure, cracks of dam-
age classes D3 and D4&5 are majors and are not affected by the
noises. Therefore, we observe only the cracks of these two classes.
From a mechanical point of view, cracks focus on the soil-structure
contact area. In addition, major cracks are straight indicating
that the flexion is predominating. At the local scale, the failure
mode can be considered as Mode I (failure by traction). This
observation is absolutely consistent with [15,16]. The final state
of the structure corresponds to a cantilever beam configuration.
We also observed that cracks propagate from right to left and from

Fig. 9. Final state of physical model in the hogging zone. (a) Horizontal displacement field. (b) Vertical displacement field. (c) Horizontal Lagrangian strain provided by the
VIC-3D software. (d) Location of cracks (reconstruction of a discrete model of the masonry). (e) Location of damage classes for the maximum displacement applied by the jack
(with reference to the initial position of the blocks).
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15 11

Fig. 11. Definition of deflexion and slope (a and b) and deformation (c).

Fig. 10. Damage indicator for the structure in hogging zone. (a and b) Aesthetic
sagging zone (Fig. 11a and b). And the maximum deformation is
damage. (c) Functional damage. (d) Structural damage. linked to the extension length of the structure (Fig. 11c).
The adaptation of conventional indicators is presented in
Fig. 12. The positive value of deflexion in Fig. 12b indicates that
the structure is in the hogging case. In addition, the two deforma-
top to bottom with increasing vertical displacement during the
tions on the top and bottom of the wall are positive (Fig. 12c). This
test.
means that the neutral axis is on the silicon foundation part.
The second damage analysis is concerned with crack character-
Burland’s method considers the building as a simple elastic
ization. Fig. 10 presents the evolution of damage classes with a
beam and establishes damage levels based on the relationship
confidence interval of 95%, i.e., twice the standard deviator.
between the maximum deflexion and deformation. The original
Because aesthetic damage classes D0&1 and D2 do not affect the
model assumes that the building has equivalent dimensions
safety, we focus on two last classes, D3 and D4&5. The evolution
(height H and length L) with a specific ratio L/H = 1, and that the
of the damage classes can be divided into 2 components: a linear
maximum deflexion is applied in the middle of the beam. Saeidi
part for a jack displacement less than 12 mm, and non-linear part
[35] studied the sensitivity of the maximum deformation to the
for the rest. The threshold of the linear part is 10% for class D3 and
variation of the ratio L/H and E/G. The ratio L/H varies from 0 to
2% for class D4&5. Nevertheless, a small variation in the relative
crack length in the latter case leads to substantial structural dam-
age and potential collapse.
According to the crack position analysis and the total length of
cracks, the conclusion of the test for the structure in the hogging
zone would be that the structure is exposed to very severe damage.
Conventional criteria of damage assessment can be based on the
average structure slope, the critical horizontal deformation, and
the combination of the deflexion and the deformation of the struc-
ture. Firstly, according to the typical values for maximum building
slope and settlement proposed for damage risk assessment by
CIRIA PR30, 1996 [11], damage is considered ‘‘negligible’’ when
the maximum slope ranges between 0% and 0.2%, ‘‘slight’’ damage
corresponds to values between 0.2% and 0.5%, moderate damage
falls between 0.5% and 2%, and for high damage, the slope is greater
than 2%. Secondly, for the criterion based on horizontal deforma-
tion (see [10]), the damage levels depend on the limiting tensile
strain such as the negligible damage (0–0.05%), very slight damage
(0.05–0.075%), slight damage (0.075–0.15%), moderate damage
(0.15–0.3%), and severe damage (>0.3%). Thirdly, the damage
assessment can be based on the method of Burland, in which the
damage levels are estimated in terms of the relationship between
the deflexion and the maximum deformation of the structure
[12,13]. Fig. 11 gives an illustration of the average slope, the rela-
tive maximum deflexion, and the maximum deformation of the
structure which are adapted for our model. The average slope b
is the gradient of the vertical displacements, which are calculated Fig. 12. Conventional indicators for the structure in the hogging zone. (a) Average
from the two extremities of the foundation (Fig. 11a). The relative slope. (b) Relative deflexion. (c) Deformation. (d) Burland’s method for damage
maximum deflexion refers to the relative value of the maximum assessment: distance between two points in the test results is equal to 0.3 mm of
deflexion of the foundation divided by the length of the hogging/ the jack displacement djack.
12 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15

8, while the ratio E/G falls between 2.6 and 12.5. The ratio E/G can
be greater than 2.6 because of the existence of the openings (win-
dows, doors, etc.) and also due to the anisotropy of the masonry. It
appeared that the deformation due to the shear force is critical
when the ratio L/H is less than 1.2 for E/G = 2.6, and L/H is less than
6.3 for E/G = 12.5. Consequently, the ratio E/G has a significant
influence on the failure mode of the beam. It also shows that the
neutral axis is important for determining the critical thresholds
of damage. In fact, considering the neutral axis at the bottom of
the structure is more conservative than assuming that it is located
on the median axis of the beam. In addition, the neutral axis can be
in the bottom of the beam in flexion because the maximum tensile
stress could be at the bottom and the minimum value could be at
the top [10,36].
The actual model corresponds to a beam subjected to a uniform
load with the ratio L/H > 1. According to the geometry of the model
in Fig. 1b, it has dimensions of 103  250 mm, i.e., L/H = 2.5, and
the neutral axis is on the silicon foundation part. The ratio between
the Young’s and shear modulus E/G is assumed to be equal to 2.6
with the Poisson’s ratio m = 0.3, which mostly corresponds to typi-
cal non reinforce masonry building [35]. Finally, the modified
threshold curves are obtained and presented Fig. 12d. It appears

Fig. 13. Final state of the physical model in the sagging zone. (a) Horizontal displacement field. (b) Vertical displacement field. (c) Horizontal Lagrangian strain. (d) Location of
cracks. (e) Location of damage classes.
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
Fig. 14. Damage indicator for the structure in the sagging zone. (a and b) Aesthetic
damage. (c) Functional damage. (d) Structural damage.
that the damage thresholds are less than the original curves of Bur-
land [12].
We observe a high correlation between the results in Figs. 10d
and 12. The first severe cracks (class D4&5) appear in Fig. 10d at
5 mm of jack displacement. At the same time, the value for relative
deflexion in Fig. 12b is 0.2% and a deformation of 0.3% is reached in
Fig. 12c. The injection of these values into a Burland curve
(Fig. 12d) indicates the severe & very severe category. Here, we
note the reliability of Burland’s method once again. However, this
method is unable to indicate the location of cracks. Consequently,
it also cannot be applied in case of local damage. In the next para-
graph, we demonstrate the performance of the proposed approach
in evaluating the structural damage in the sagging zone with local
cracks.
4.2.2. Structure in sagging zone
The geometric setup in the sagging test is shown in Fig. 8b. The
parameter selected for the DIC analysis and the generated grid are
the same as in the previous configuration. The average value of
measurement errors is du = (0.5 ± 6.3)  103 mm for the horizon-
Table 2
Comparison of different indicators.
13
Fig. 15. Conventional indicators for the structure in the sagging zone. (a) Average
slope. (b) Relative deflexion. (c) Deformation. (d) Burland’s method for damage
assessment (distance between two points in the test results is equal to 0.3 mm of
the jack displacement djack).
tal displacement and dv = (0.8 ± 5.3)  103 mm for the vertical
displacement.
The crack location is shown in Fig. 13. The main cracks are dis-
tributed on the two vertical sides of the wall. According to [15], the
deformed shape of the structure is expected to be a flexural beam
and cracks should appear at midspan. However, the initial setup of
our physical model has a small difference with the perfect interface
between blocks and foundation C1 in Fig. 1b. This configuration
restricts the development of deformation in the middle of the wall.
In addition, the high flexural rigidity of the foundation induces less
deformation in the structure: the foundation is 40 mm thick and
250 mm wide, and Young’s modulus is equal to 5 MPa, so the flex-
ural rigidity is 3.3 N m2 [37]. Fig. 13e illustrates the location of the
damage classes. In this case, the cracks present a local failure at the
extremities of the observed wall due to the influence of perpendic-
ular walls.
The damage indicator is presented in Fig. 14. In this case, the
relative value of the total length of cracks is smaller than for the
14 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15

structure in the hogging zone, with a maximum of 4.1% for class D3 Appendix A. Fontainebleau NE 34 sand for the analogue soil
and 0.6% for class D4&5. According to the crack position analysis and
the quantity of the total length of cracks, the damage level of the
structure in the sagging zone should be in the moderate damage State Unit Dr Young Peak Residual
category. However, the initial condition on the interface between weight (%) modulus friction friction
the blocks and the silicon foundation can lead to defective struc- (kN/m3) E (MPa) angle (°) angle (°)
tural behavior in this position. Medium 15.42 44 – 30–36 24–33
The conventional indicators are shown in Fig. 15. The average
slope in Fig. 15a is almost zero in the first 15 mm of jack displace-
ment, then increases and reaches a maximum value of 1.4% in the Appendix B. Azobe wood for the analogue structure
final state. This explains why the structure is stable at the first time
and then unbalanced due to the dissymmetric movements of soil.
The deflexion in Fig. 15b is less significant than in the tensile test. Parameter Prototype Ideal model Azobe
The negative value indicates that the structure is in the sagging wood
case. What is more, the negative deformations in Fig. 15c indicate Dimensions 200  250  500 5  6.25  12.5 7  7  14
that the neutral axis is out of the beam. (mm)
Finally, discrepancies between the 4 indicators are highlighted
in Table 2 and indicate that the Burland’s method is not applicable Blocks
to the present model. The slope indicator coincides with the pro- Unit weight 8.9 8.9 10.3
posed approach for the two examples of the structure positions. (kN/m3)
However, this indicator yields different results with other positions Young 10,000 10,000 16,000–
of the structure, the results of which are presented in another modulus 19,000
paper. E (GPa)
Friction 20–35 20–35 30 ± 9
angle (°)
5. Conclusion
Mortar
In this paper, we proposed a new approach for crack identifi- Cohesion 0.5–1.2 0.0125–0.03 No used
cation using physical modelling that incorporates an experimen- (MPa)
tal damage criterion based on crack characterization: width,
length and orientation, with attention to the masonry structure.
Based on the DIC technique, a numerical scheme is presented in References
order to reconstruct the blocks in a discrete system description,
in which the displacement fields are broken down into rotation [1] Moës N, Dolbow J, Belytschko T. A finite element method for crack growth
without remeshing. Int J Numer Methods Eng 1999;46:131–50.
and translation components. Because the cracks appear near [2] Réthoré J, Hild F, Roux S. Extended digital image correlation with crack shape
joints, we can determine three important properties of cracks – optimization. Int J Numer Methods Eng 2008;73:248–72.
location, width, and length – and use them for an accurate dam- [3] Chen J, Zhang X, Zhan N, Hu X. Deformation measurement across crack using
two-step extended digital image correlation method. Opt Lasers Eng
age evaluation. 2010;48:1126–31.
Thus, a damage indicator has been developed for quantifying [4] Réthoré J, Hild F, Roux S. Shear-band capturing using a multiscale extended
damage in masonry structures that is associated with the total digital image correlation technique. Comput Methods Appl Mech Eng
2007;196:5016–30.
length of cracks. This indicator is based on the observations of [5] Mathieu F, Hild F, Roux S. Image-based identification procedure of a crack
crack width in the numerous real cases proposed by Burland. This propagation law. Eng Fract Mech 2013;103:48–59.
is then completed by using the total length of cracks exhibiting a [6] Nguyen TL, Hall SA, Vacher P, Viggiani G. Fracture mechanisms in soft rock:
identification and quantification of evolving displacement discontinuities by
certain crack width. As a result, the proposed indicator has numer-
extended digital image correlation. Tectonophysics 2011;503:117–28.
ous advantages compared to the conventional methods that use [7] Lourenço PB. Computational strategies for masonry structures. PhD thesis,
the limiting tensile strain of the structure. In particular, it makes Universidade do Porto, Portugal; 1996.
it possible to evaluate the local damage of the structure, which is [8] Tung S-H, Shih M-H, Sung W-P. Development of digital image correlation
method to analyse crack variations of masonry wall. Sadhana 2008;33:767–79.
not possible using conventional methods. [9] Mohammadipour AH, Willam KJ, Ayoub A. Experimental studies of brick and
However, the characterization of cracks involves some uncer- mortar composites using digital image analysis. In: Van Mier JGM, Ruiz G,
tainties due to experimental processes. The Monte Carlo method Andrade C, Yu RC, Zhang XX, editors. 8th International association of fracture
mechanics for concrete and concrete structures, FraMcoS-8, Toledo, Spain;
has been used to determine the effect of displacement measure- 2013.
ment errors on the value of the damage indicator. It appeared that, [10] Boscardin M, Cording E. Building response to excavation-induced settlement. J
in general, the total length of cracks is known within ±3% with a Geotech Eng 1989;115:1–21.
[11] Loganatban N. An innovative method for assessing tunnelling-induced risks to
95% confidence interval. For practical application of the proposed adjacent structures. One Penn Plaza ed. Parsons Brinckerhoff Inc.; 2011.
indicator, it is advised that a safety margin of +3% be added to [12] Burland JB. The assessment of the risk of damage to buildings due to tunnelling
the results of the severe and very severe damage classes. and excavations. Earthquake geotechnical engineering. Balkema: Editions
Ishihara; 1997. p. 1189–201.
Using our physical model, the investigation also addressed [13] Potts DM, Addenbrooke TI. A structure’s influence on tunnelling-induced
damage to masonry due to underground excavation. For this ground movements. Proc Inst Civil Eng-Geotech Eng 1997;125:109–25.
model, due to the fact that the foundation has elastic behavior [14] Deck O, Singh A. Analytical model for the prediction of building deflections
induced by ground movements. Int J Numer Anal Methods Geomech
and the masonry is considered as a brittle material, the foundation
2012;36:62–84.
is considered as an element that transfers damage into the [15] Boone SJ. Assessing construction and settlement-induced building damage: a
structure. return to fundamental principles. California: Golder Associates Inc.; 2002. p.
Research may be improved with more realistic models of 559–70.
[16] Giardina G, Marini A, Hendriks MAN, Rots JG, Rizzardini F, Giuriani E.
masonry structures, such as those which take the windows, the Experimental analysis of a masonry façade subject to tunnelling-induced
mortar, etc., into account. settlement. Eng Struct 2012;45:421–34.
 H.-L. Nghiem et al. / Engineering Structures 86 (2015) 1–15
[17] Laefer DF, Hong LT, Erkal A, Long JH, Cording EJ. Manufacturing, assembly, and
testing of scaled, historic masonry for one-gravity, pseudo-static, soil-
structure experiments. Constr Build Mater 2011;25:4362–73.
[18] MacKay DJ. Introduction to Monte Carlo methods. Learning in graphical
models. Springer; 1998. p. 175–204.
[19] Keating TJ, Wolf PR, Scarpace FL. An improved method of digital image
correlation. Am Soc Photogr Rem Sens 1975:41.
[20] Chu TC, Ranson WF, Sutton MA. Applications of digital-image-correlation
techniques to experimental mechanics. Exp Mech 1985;25:232–44.
[21] Grédiac M. The use of full-field measurement methods in composite material
characterization: interest and limitations. Compos A: Appl Sci Manuf
2004;35:751–61.
[22] Sutton M, McNeill S, Helm J, Chao Y. Advances in two-dimensional and three-
dimensional computer vision. In: Rastogi P, editor. Photomechanics. Berlin,
Heidelberg: Springer; 2000. p. 323–72.
[23] Orteu J-J. 3-D computer vision in experimental mechanics. Opt Lasers Eng
2009;47:282–91.
[24] Haddadi H, Belhabib S. Use of rigid-body motion for the investigation and
estimation of the measurement errors related to digital image correlation
technique. Opt Lasers Eng 2008;46:185–96.
[25] Robert L, Nazaret F, Cutard T, Orteu JJ. Use of 3-D digital image correlation to
characterize the mechanical behavior of a fiber reinforced refractory castable.
Exp Mech 2007;47:761–73.
[26] VIC-3D. Testing guide, projection error, bias and noise; 2010.
<www.correlatedsolution.com>.
[27] Fazzini M. Développement de méthodes d’intégration des mesures de champs.
PhD thesis, Université de Toulouse, Toulouse; 2009.
15
[28] Salmanpour A, Mojsilović N. Application of Digital Image Correlation for strain
measurements of large masonry walls. Singapore: APCOM & ISCM; 2013.
[29] Garcia D, Orteu JJ. 3D deformation measurement using stereo-correlation
applied to experimental mechanics; 2001. p. 50–60.
[30] Deck O. Etude des conséquences des affaissements miniers sur le bâti:
proposition pour une méthodologie d’évaluation de la vulnérabilité du bâti.
PhD thesis, Institut National Polytechnique de Lorraine; 2002.
[31] Hor B. Evaluation et réduction des conséquences des mouvements de terrains
sur le bâti: approches expérimentale et numérique. PhD thesis, Institut
National des Sciences Appliquées de Lyon; 2012.
[32] Al Heib M, Nghiem HL, Emeriault F. Understanding sinkhole consequences on
masonry structures using large small-scale physical modeling. In: 7th
International conference on case histories in geotechnical engineering. The
Westin Chicago North Shore, Wheeling, Illinois; 2013.
[33] Garnier J. Properties of soil samples used in centrifuge models. Physical
modelling in geotechnics. St John’s, Canada; 2002. p. 5–19.
[34] Triconnet K, Derrien K, Hild F, Baptiste D. Parameter choice for optimized
digital image correlation. Opt Lasers Eng 2009;47:728–37.
[35] Saeidi A. La vulnérabilité des ouvrages soumis aux aléas mouvements de
terrains; développement d’un simulateur de dommages. PhD thesis, Loraine:
Institut national polytechnique de Lorraine; 2010.
[36] Burland JB, Wroth CP. Settlement of buildings and associated damage. In:
Conference on settlement of structure. Cambridge, Pentech Press, London;
1974. p. 611–54.
[37] Al Heib M, Emeriault F, Caudron M, Nghiem L, Hor B. Large-scale soil-structure
physical model (1g)-assessment of structure damages. Int J Phys Model
Geotech 2013;13:138–52.