Original Article

Structural Health Monitoring

2016, Vol. 15(1) 81–92

Multifractal analysis of crack patterns Ó The Author(s) 2016

Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
in reinforced concrete shear walls DOI: 10.1177/1475921715624502
shm.sagepub.com

Arvin Ebrahimkhanlou1, Alireza Farhidzadeh2 and Salvatore Salamone1

Abstract
Conventionally, the assessment of reinforced concrete shear walls relies on manual visual assessment which is time-
consuming and depends heavily on the skills of the inspectors. The development of automated assessment employing fly-
ing and crawling robots equipped with high-resolution cameras and wireless communications to acquire digital images
and advance image processing to extract crack patterns has paved the path toward implementing an automated system
which determines structural damage based on visual signals acquired from structures. Since there are few, if any, studies
to correlate crack patterns to structural integrity, this article proposes to analyze crack patterns using a multifractal anal-
ysis. The approach is initially tested on synthetic crack patterns, and then it is applied to a set of experimental data col-
lected during the testing of two large-scale reinforced concrete shear wall subjected to controlled reversed cyclic
loading. The structural response data available for each specimen are used to link the multifractal parameters with the
structural performance of the two specimens. A relationship between the multifractal parameters and the crack pat-
terns’ evolution and mechanism is noted. The results show that as the crack patterns extend and grow, multifractal para-
meters move toward higher values. The parameters jump as the mechanical response shows severe stiffness loss. In this
study, no attempt is made to automate the process of mapping cracks from images.

Keywords
Multifractal analysis, visual assessment, damage assessment, surface crack patterns, reinforced concrete structures

Introduction equipped with high-resolution camera and wireless

communications to acquire two-dimensional (2D) digi-
Reinforced concrete shear walls (RCSWs) are one of tal images; then image processing is used to automati-
the most commonly used seismic resisting systems in cally retrieve damage properties, such as crack length
conventional buildings and infrastructures. RCSWs are and width.8–12 Despite their increasing usage in the last
subjected to deterioration due to aging, increased load, few years, few automated VI methods, if any, other
and natural hazards. To secure the overall soundness than the authors’ recent publication exist13 to link the
of these structures, proper assessment is crucial. effect of damage to the overall soundness of the struc-
Commonly, RCSWs are assessed visually by tracking ture. Current systems only provide the information
defects such as corrosion and spalling and quantifying about crack length and width to engineers to determine
the length and the width of the existing cracks.1,2 the course of action: retrofit, replacement, or no action.
Concrete bridge decks are also commonly assessed The results presented in this article are part of a long-
based on crack’s density.3,4 Worldwide, various
condition-rating grades have been developed to quan-
1
tify damage conditions in terms of crack’s width and Smart Structures Research Laboratory (SSRL), Department of Civil,
length.5–7 Although manual visual inspection (VI) is a Architectural and Environmental Engineering, The University of Texas at
Austin, Austin, TX, USA
well-established method to inspect RCSWs, it is time- 2
Products & System Division, Mistras Group Inc., Princeton Junction, NJ,
consuming and depends heavily on the skills of the USA
inspectors. Moreover, it may take weeks to obtain
access to buildings hit by natural hazards, such as Corresponding author:
earthquake. In order to overcome these limitations, Salvatore Salamone, Smart Structures Research Laboratory (SSRL),
Department of Civil, Architectural and Environmental Engineering, The
automated-based VI is a field that has received signifi- University of Texas at Austin, 301 E Dean Keeton, C1748, Room ECJ
cant interest over the past few years. Automatic inspec- 4.710, Austin, TX 78712, USA.
tion techniques use flying and crawling robots Email: [email protected]
82 Structural Health Monitoring 15(1)

term project that aims at developing an automated- filling properties of the object as an indication of the
based inspection technology for the nondestructive eva- object’s complexity.14 In this approach, a virtual grid
luation (NDE) of RC structures. The technology is of squared boxes overlaps the object, and the number
based on the fractal analysis of 2D images taken in the of boxes of a given size (r) necessary to cover it is
visible spectrum, to retrieve surface defect patterns that counted. As r ! 0, the total area covered by the area
can provide a quantitative measure of damage. Fractal elements converges to the measure of FD.22 Based on
analysis is a relatively novel mathematical tool that has this method, FD is defined as
been used successfully in many fields including biology
and physical sciences.14–17 However, its application in log N (r)
D = lim ð1Þ
the NDE/structural health monitoring (SHM) commu- r!0 log (r)
nity has been modest.18–21 Recently, the authors have
used a monofractal analysis to discriminate different where N (r) is the total number of non-empty boxes of
damage grades (DGs) in RCSWs.22 They have shown linear size r.
that, in general, the fractal dimension (FD) increases as Many natural phenomena, however, have a more
the structural damage increases. Although these prelim- complex scaling relation, and their statistics cannot be
inary results using the monofractal analysis have been described by just a single scalar (i.e. D). For instance,
encouraging, it is important to recognize that using a consider a 2D signal such as a gray scale image. For
single scalar, the FD may be insufficient as a summary describing an object of the image, the box-counting
statistic for the overall crack patterns. In this article, a method may not be appropriate since it gives only a
more general approach is presented (i.e. multifractal relation between the non-empty boxes and the box
analysis) based on point-wise scaling properties (i.e. size;27 therefore, no weighting is done to the count
local FDs) of the image. As cracks develop, their width, according to the signal levels into the boxes.
length, density, and FD grow (local change). Besides, Multifractal analysis has shown the potential to pro-
as cracking mechanism changes, multifractal para- vide more insights into the scaling properties of these
meters form separate clusters which indicates the objects.28,29 The multifractal analysis is a generaliza-
change in mechanism (global change). So the idea is to tion, in which the FD, instead of being a global para-
decompose 2D images into many subsets characterized meter which only gives information on the support of a
by different local FDs, which quantify the local singu- given measure,30 is a local parameter (called singularity
lar behavior and as a result relate to the local scaling of strength a) that may change from box to box and pro-
the image.23 Thus, a 2D image will require a set of FD vides more insight into the way the measure is distribu-
to fully characterize its scaling properties. This article is ted over its support.28,31 In multifractal analysis, the
organized as follows. First, a brief overview on the scale-invariant properties of the object can be charac-
multifractal analysis is provided, followed by applica- terized on the basis of its generalized32 dimensions Dq,
tion of the method on synthetic cracks. Then, the defined as
experimental results are presented with application to
NP
(r)
the damage assessment of two large-scale RCSWs log Pi q (r)
under lateral cyclic loading. 1 i=1
Dq = lim ð2Þ
q  1 r!0 log (r)

Multifractal analysis where q is a real value moment order and Pi (r) is the
measure (or weight) associated with the ith box. For a
Background binary image, Pi (r) can be seen as the probability of
The term ‘‘fractal’’ was first introduced by existence of black pixels in that box, that is33
Mandelbrot24 to indicate self-similar objects whose
complex geometry cannot be characterized by an inte- Ni (r)
Pi (r) = ð3Þ
ger dimension. Since then, the term has been exten- P
M(r)
sively used to study natural and experimental physical Ni (r)
i=1
phenomena.25 A common way to characterize fractal
objects is by their FD which, very roughly speaking, is where Ni (r) is the number of black pixels inside the ith
related to their degree of complexity and the extent to box of size r, and M(r) is the total number of boxes con-
which features at different scales are related. More spe- taining at least one black pixel. The most popular gen-
cifically, FD is defined as a ratio comparing how detail eralized dimensions are as follows: the capacity
in a pattern changes with the scale at which it is mea- dimension (D0), the information dimension (D1),
sured.26 The most popular algorithm to calculate the and the correlation dimension (D2). In general, Dq is
FD is the box-counting, which considers the space a monotone decreasing function of q. If Dq is a
Ebrahimkhanlou et al. 83

single-valued function, equal to the D defined in (1), the any value of q. Then, the two functions a(q) and f (q)
object is called monofractal.14 Overall, the parameter q are evaluated
serves as a ‘‘microscope’’ that analyzes the object at dif-
ferent scales.14 In multifractal analysis, one can also P
N(r)
mi (q, r) logðPi (r)Þ
determine the number of boxes having similar local
scaling, that is same a, and define f (a) as the FD of the a(q) = lim i = 1 ð5Þ
r!0 log (r)
set of boxes with singularities a. The curve f (a), also
NP
(r)
called multifractal spectrum, is a convex function whose mi (q, r) logðmi (q, r)Þ
maximum corresponds to D0. As q ! ‘, a ! amax , i=1
f (q) = lim ð6Þ
and as q ! + ‘, a ! amin . In general, when f (a) and r!0 log (r)
Dq are smooth functions of a and q, a Legendre trans-
formation can be used to derive f (a) from Dq.34,35 For each q, the values ofP a(q) and f (q) are obtained
N (r)
However, f (a) can also be directly calculated, without from the slope of plots
PN (r) of i = 1 mi (q, r) log (Pi (r)) ver-

knowing Dq, using the method proposed by Chhabra sus log (r), and i = 1 mi (q, r) log (mi (q, r)) versus log (r)
and colleagues34,35 The first step of this approach con- over the entire range of box size values under consider-
ation. Finally, these two data sets of a(q) and f (q) are
sists of defining a family of normalized measures
plotted with respect to each other to construct the sin-
mi (q, r) defined as
gularity spectrum (i.e. plot of a vs f (a)).
Pi (r)q
mi (q, r) = ð4Þ
NP
(r) Synthetic data analysis
Pi (r)q
i=1 The analysis of synthetic data sets presented in this sec-
tion is intended to illustrate how the multifractal para-
For each box i, the normalized measure mi (q, r) meters are calculated and shows their application for
depends on the order of the statistical moment, and on crack patterns’ quantification in RCSWs. As shown in
the box size and it takes values in the range [0, 1] for Figure 1, the data sets consisted of four binary images

Figure 1. Synthetic crack patterns: (a) CASE 0, (b) CASE 1, (c) CASE 2, and (d) CASE 3.
84 Structural Health Monitoring 15(1)

Figure 2. Probability and measure functions of synthetic cracks at q = 21, q = 0, and q = 1; all plots are normalized.

(size 5003400 pixels), representing typical crack pat- the boxes is lost. Therefore, for q = 0, a traditional
terns in a shear wall (size of the wall was assumed to be monofractal analysis is recovered.
500034000 mm). Figure 1(a) is assumed as a base pat- As it was described in section ‘‘Background,’’ a(q),
tern (CASE 0), while the other three cases were gener- f (q), and Dq were calculated using regression lines over
ated by simple manipulations of CASE 0. Specifically, the entire range of box sizes. Figure 3 illustrates some
for CASE 1 (Figure 1(b)), minor changes were applied of these regression lines for CASE 0 (Figure 1(a)). In
on CASE 0, while the overall shape was preserved. this figure, regression line plots for only three values of
CASE 2 (Figure 1(c)) was generated by flipping one of q, that is, q 2 f1, 0, 1g, are shown in column-wise
the two cracks of the base pattern. Hence, the crack order.
details are exactly the same, but the general shape was As illustrated in Figure 4, where f (a) and Dq curves
changed. Finally, CASE 3 is representative of crack are shown for the four synthetic crack patterns shown
patterns’ growth. in Figure 1, multifractal analysis could provide a sensi-
In order to perform the multifractal analysis of these tive tool for the discrimination of crack patterns. As
images, the Chhabra method described in section expected, the curves f (a) are convex with a single maxi-
‘‘Background’’ was used to directly evaluate the singu- mum at q = 0 while Dq are monotonically decreasing.
larity spectrum f (a) and the generalized dimension Dq. The relation between a and the parameter q is the fol-
The analysis was carried out by dividing each image lowing: the right branch corresponds to the negative
with rectangular boxes of aspect ratio similar to the values of q and the left one to the positive values of q.
original image. The number of division ranged from 1 By comparing CASE 0 and CASE 1, it can be observed
to 40 (i.e. the smallest box size was equal to that f (a) and Dq show negligible difference for the posi-
12:5310 pixels), while the range of q was selected within tive values of q, whereas more significant difference can
the range between 10 and 210 (i.e. q 2 ½10, 10). be observed for the negative values of q. The reason
Figure 2 shows Pi (r) and mi (q, r) for the data set behind this different behavior is that q acts as a scan-
shown in Figure 1(a). Two different box sizes were con- ning tool, that is, for q.0 regions with high probability
sidered, that is, 100380 and 50340 pixels. As expected, are amplified, while for q\0 regions with low probabil-
the higher value of Pi (r) was obtained at the boxes with ity are magnified, making the analysis more sensitive to
higher concentration of black pixels. In order to illus- the local variations; therefore, for q.0, the analysis
trate the effect of the parameter q, mi (q, r) was calcu- focuses more on the general shape rather than the local
lated for three different values of q, that is, features of the input patterns. For CASE 2, since the
q 2 f1, 0, 1g, as shown in Figure 2. It can be observed overall shape of the pattern changed, while the local
that the negative values of qs (q =  1) are dominated features were kept the same, both f (a) and Dq show the
by the boxes of low probability densities (i.e. rare change just in the negative range of q. For CASE 3, a
events), while the positive values of qs (q = + 1) are qualitative difference in the f (a) and Dq curves can be
dominated by boxes of high probability densities (i.e. clearly recognized, caused by differences in crack pat-
smooth events). For q = 0, each box is given equal terns. Therefore, it is conceivable that with the help of
weight regardless of the number of points in the boxes. the multifractal analysis, an automatic ‘‘early warning’’
Thus, information concerning distribution of points in of the damage progression could be implemented if the
Ebrahimkhanlou et al. 85

Figure 3. Regression lines on the synthetic crack pattern CASE 0’s data. Note: r is in millimeters.

Figure 4. (a) Singularity spectrum and (b) generalized dimension of the synthetic crack patterns.

actual shape of f (a) and Dq curves deviates consider- structures. Figure 5 shows an overall view of the speci-
ably from critical curves representing ‘‘normal’’ operat- mens, and Table 1 summarizes their main characteris-
ing conditions. tics. More details regarding specimens’ design can be
found in Rocks.38 To simulate earthquake loading
effects, each specimen was subjected to a quasi-static
Experiments cyclic loading. Cyclic loads were applied by two
Experimental tests were carried out on two large-scale hydraulic actuators to the top of the specimens while
their foundations were fully connected to the strong
RC concrete shear walls, namely, SW1 and SW2 with
floor laboratory with 14 post-tensioned, 1.5 in nominal
height-to-width ratios of 0.94 and 0.54, respectively. diameter Dywidag bars. Load transfer between the
The walls were designed in compliance with ACI 318- actuators and the walls was provided with brackets and
08 including its chapter 21, earthquake resistance plates at the sides of the wall. The plates were post-
86 Structural Health Monitoring 15(1)

consisted of two load cycles, except load step zero

which had three cycles. This load step was intended to
verify functionality of the experimental setup and its
results are not considered. Figure 6 shows the test load-
ing protocols. Loading rate was set around 6 and
1.3 mm/s, respectively, for SW1 and SW2. Crack map-
ping was carried out at each peak and zero displace-
ment. The maximum crack widths were measured using
transparent scales, and pictures were taken at the end
of each load step using a high-definition camera. Due
to severe crushing and spalling of SW1 at the end of
experiment, and tilting of SW2 at LS10, crack monitor-
ing was not considered at the end of the last load step.

Experimental results
Mechanical behavior
Figure 7 shows the force–displacement hysteresis loops
for the two specimens. A trilinear backbone curve is
superimposed on the graphs. This curve is suggested by
ASCE standard 41-13, for nonlinear analysis of low
aspect ratio walls.39 The trilinear backbone curves (I,
II, and III) identified in Figure 7 correspond to concrete
cracking (I), yielding of the reinforcement (II), and ulti-
mate strengths (III), observed at load steps LS2, LS6,
and LS9, respectively for SW1, while at LS1, LS7, and
LS9 for SW2.

VI
Figure 5. Experimental setup: (a) SW1 and (b) SW2.
At the end of each cycle, the formation of new cracks
was observed while the existing cracks were expanded.
Table 1. Specimens’ details.36,37 Therefore, the length or density of the cracks increased
monotonically. Crack width measurements were per-
Characteristics SW1 SW2 formed at peak displacements, and zero displacements
Height (mm) 3300 2100 (i.e. residual cracks), for each load step. In order to cate-
Width (mm) 3050 3050 gorize the severity of damage, condition-rating grades
Thickness (mm) 200 200 suggested by the International Atomic Energy Agency
Reinforcement ratio (%) 0.67 1 (IAEA) guidebook5 were adopted. Specifically, three
Wall f 0c (28th day) (MPa) 21 35
DGs were defined (different DGs are detonated with
Wall f 0c (test day) (MPa) 25 48.2
alphabets in this article while roman letters are used in
Rebar yield strength (MPa) 464 434
Rebar ultimate strength (MPa) 708 600 the IAEA guidebook): grade A in which the maximum
crack width was less than 0.2 mm, grade B (i.e. moder-
ate damage) in which the largest crack width was com-
tensioned connected to each other at 450 mm below prised between 0.2 and 1 mm, and grade C (i.e. critical
the top of each wall. In order to ensure synchronized damage) for cracks larger than 1 mm. In general, no
loading, the actuators were actively controlled with repair is needed for grade A, whereas appropriate reha-
master-slave method. They were also sloped downward bilitation strategy is necessary for grades B and C.
about 9° from horizon to prevent out-of-plane beha- Figure 8 compares the crack widths at peak and zero
vior. Due to typical low axial stress in low aspect ratio displacements during testing. The DG thresholds sug-
RCSW, axial load was not applied in these gested by IAEA guidebook are also shown on the back-
experiments. ground of the graphs. For the sake of clarity, cracks at
It was proposed to apply 10 load steps on each wall; peak displacements are shown in the middle of each
however, failure occurred during LS10. Each load step load step, while residual cracks are shown at the end of
Ebrahimkhanlou et al. 87

Figure 6. Load protocol:37 (a) SW1 and (b) SW2.

Figure 7. Force–displacement hysteresis loops:37 (a) SW1 and (b) SW2.

Figure 8. Crack width at peak and zero displacement, damage grades suggested by IAEA are shown with background colors:22 (a)
SW1 and (b) SW2.
88 Structural Health Monitoring 15(1)

Figure 9. SW1 residual cracks after LS4: (a) original image and (b) crack patterns.22

Figure 10. Probability and measure functions of SW1-LS4 cracks at different scales and box sizes.

each load step. It is worth noting that damage classifi- Multifractal analysis results
cation based on residual crack measurements may lead
The Chhabra method was used to calculate the multi-
to underestimation of the actual severity of damage.
fractal singularity spectrum and generalized dimension
For instance, according to the DGs assigned to SW1
for the residual cracks mapped at the end of each load
based on its residual crack width, critical damage (grade
step. Images were treated as binary images, composed
C) was reached at LS8 (Figure 8(a)), while the mechani-
cal behavior of the wall identified severe damage at of pixels which are either black or white. Each image
LS7. Similarly, DG C was identified in LS9 of SW2 was partitioned into a grid of rectangular boxes of the
(Figure 8(b)), while its backbone curves show severe same width-to-length ratio of that specimen under
damage in LS8. In addition, as the structure returns to investigation, and the box sizes ranging from the wall
its rest position, cracks close; therefore, different lateral size to 40 times smaller ones. Binary images containing
displacements of the wall result in different crack width crack patterns were, respectively, of size 5523443 and
measurements for the same actual level of damage. 7303322 pixels for SW1 and SW2. Therefore, the smal-
Crack patterns images were also collected at the end lest boxes have short edges of at least 11 and 8 pixels,
of each load step, when the wall reached the zero dis- respectively, for SW1 and SW2.
placement (residual cracks). In order to extract binary Figure 10 illustrates Pi (r) and mi (q, r) at scales
images from the original ones, crack patterns were q 2 f1, 0, 1g, and for two box sizes, that is,
drawn manually on the original images using Adobe 203:33163:2 mm and 101:7381:6 mm, corresponding
Photoshop, as shown in Figure 9. A complete sequence to partitioning the original image into 15 and 30 boxes,
of residual crack patterns for both SW1 and SW2 can respectively (note: a portion of each wall was under the
be found in Farhidzadeh et al.22 braces and not visible). As box sizes decrease, the
Ebrahimkhanlou et al. 89

Figure 11. Regression lines for SW1-LS4 crack patterns multifractal analysis. Note: r is in millimeters.

overall area covered by the boxes decreases; as a result, crack pattern changes in a spectrum of scales from local
the shape of the filled boxes will approach the original to global scales. Since cracks are monotonically increas-
shape of the crack patterns. Supposing that the mesh is ing, a monotonic increase in local scaling is expected
large enough that all the boxes are filled by at least one while global changes are expected only after changes in
crack, and the entire special domain is fully covered by cracking mechanism which results in generating differ-
filled boxes. As a finer mesh covers the domain, regions ent patterns.
having less density of cracks are more likely to remain In order to correlate the multifractal parameters, with
empty. Therefore, the boxes with lower density are the mechanical behavior of the specimens, the curves
more likely to lose part of their area when covered by a were clustered according to the trilinear backbone curve
finer mesh. These boxes are called rare events. Lower shown in Figure 7. The three clusters (I, II, and III) are
density in these boxes results in lower probability. superimposed in Figure 13. The following observation
Negative q magnifies these boxes, and eventually pro- can be made for SW1. First, significant changes for the
vides higher mi (q, r). On the other hand, the boxes with negative values of q can be observed between load steps 2
higher density (e.g. where two cracks cross each other) (LS2) and 3 (LS3), where the initial yielding of the wall
are called smooth events because they are less probable occurred. This may also indicate the occurrence of some
to loss area as box size reduces. minor localized cracks, similar to CASE 1 for the syn-
Figure 11 shows regression lines fitted to data points thetic data. Then, significant changes can be seen for the
obtained from SW1 at LS4 using different box sizes. entire range of q (i.e. positive and negative), in the subse-
Each column in this figure corresponds to a different quent load steps in which significant yielding and stiffness
value for q, and rows correspond to regressions for dif- degradation occurred, similar to the results predicted for
ferent parameters. Figures 12 and 13 show f (a) and Dq CASE 3. For SW2, the changes in the multifractal para-
curves for SW1 and SW2, respectively. It can be seen meters occurred in the entire range of q, with the most
that as the crack patterns extend and grow, f (a) and Dq significant changes corresponding to the initial concrete
curves move toward higher values, which was expected cracking (I), yielding of the reinforcement (II), and ulti-
from the preliminary analysis carried out on the syn- mate strengths (III), observed at load steps at LS1, LS7,
thetic data (see CASE 3). Multifractal analysis tracks and LS9, respectively.
90 Structural Health Monitoring 15(1)

Figure 12. Multifractal analysis of SW1 at different load steps; clusters are indicated with circles and guidelines: (a) singularity
spectrums and (b) generalized dimension.

Figure 13. Multifractal analysis of SW2 at different load steps; clusters are indicated with circles and guidelines: (a) singularity
spectrums and (b) generalized dimension.

Figure 14. Capacity and information fractal dimensions at the end of each load step; multifractal clusters are indicated with
background colors: (a) SW1 and (b) SW2.

Figure 14 shows information, and correlation jump in the curve could indicate severe changes in
dimensions at various load steps, and compares them the mechanical behavior of the structure. In addi-
with the clusters. It can be observed that both infor- tion, the difference between information and capac-
mation and capacity dimensions of crack patterns ity dimension in region I is notably less than other
increase monotonically with load steps, and each regions.
Ebrahimkhanlou et al. 91

Conclusion article: This work was supported by the National Science

Foundation under Grant No. CMMI-1333506.
RCSWs are one of the most commonly used seismic
resisting systems in conventional buildings. Commonly,
RCSWs are assessed visually by tracking defects such References
as corrosion and spalling and quantifying the length 1. Foudazi A, Ghasr MT and Donnell KM. Characteriza-
and the width of the existing cracks. Although manual tion of corroded reinforced steel bars by active micro-
VI is a well-established method to inspect RCSWs, it is wave thermography. IEEE T Instrum Meas 2015; 64(9):
time-consuming and depends heavily on the skills of 2583–2585.
the inspectors. This article presented an approach 2. Farhidzadeh A, Ebrahimkhanlou A and Salamone S.
Corrosion damage estimation in multi-wire steel strands
based on the multifractal analysis of 2D images taken
using guided ultrasonic waves. In: Proceedings of SPIE,
in the visible spectrum, to retrieve surface defect pat-
structural health monitoring and inspection of advanced
terns that can provide a quantitative measure of dam- materials, aerospace, and civil infrastructure (ed PJ Shull),
age. Traditional multifractal parameters, including San Diego, CA, 8–12 March 2015, vol. 9437, p. 94371F.
singularity spectrum and generalized dimension curves, Bellingham, WA: SPIE.
were first illustrated for synthetic crack patterns. Main 3. Schmitt TR and Darwin D. Bond of epoxy-coated wire to
differences between monofractal and multifractal anal- concrete. SL report 92-5, November 1992. Lawrence, KS:
ysis were emphasized. The approach was applied to a University of Kansas Center for Research, Inc.
set of experimental data collected during testing of two 4. Lindquist WD, Darwin D and Browning JP. Cracking
large-scale RCSW subjected to controlled reversed cyc- and chloride contents in reinforced concrete bridge decks.
lic loading in the plane of their web. In particular, the SM report 78, February 2005. Lawrence, KS: University
of Kansas Center for Research, Inc.
available data set included detailed records of crack
5. International Atomic Energy Agency (IAEA). Guidebook
images and structural performance for each test speci-
on non-destructive testing of concrete of structures. Vienna:
men. The structural response data available for each IAEA, 2002.
specimen were used to link the multifractal parameters 6. Federal Highway Administration (FHWA). Recording
with the structural performance of the two specimens. and coding guide for the structure inventory and appraisal
It was observed that as the crack patterns extend and of the nation’s bridges. Washington, DC: FHWA, 1995.
grow, f (a) and Dq curves move toward higher values 7. Office of the Federal Register (OFR). 23 C.F.R. §650C
and jump as cracking mechanism changes. This trend National Bridge Inspection Standards. Washington, DC:
was also predicted by the preliminary analysis on the U.S. Government Printing Office, 2015.
synthetic crack patterns. 8. Chen Z and Hutchinson TC. Image-based framework for
It is conceivable that with the help of the multifractal concrete surface crack monitoring and quantification.
analysis, an automatic alarming system could be imple- Adv Civ Eng 2010; 2010: 1–18.
9. Jahanshahi MR and Masri SF. A new methodology for
mented that will alert the appropriate engineers if the
non-contact accurate crack width measurement through
actual multifractal spectrum deviates considerably from photogrammetry for automated structural safety evalua-
a critical multifractal curve representing ‘‘normal’’ oper- tion. Smart Mater Struct 2013; 22: 035019.
ating in a structure. However, more tests and investiga- 10. Farhidzadeh A, Ebrahimkhanlou A and Salamone S. A
tions are needed to standardize the approach and define vision-based technique for damage assessment of rein-
a critical multifractal curve for each RCSW based on forced concrete structures. In: Proceedings of SPIE,
their specifications. health monitoring of structural and biological systems 2014
(ed T Kundu), San Diego, CA, 9–13 March 2014, vol.
Acknowledgements 9064, p. 90642H. Bellingham, WA: SPIE.
The authors thank Prof. Andrew Whittaker for sharing the 11. Torok MM, Golparvar-Fard M and Kochersberger KB.
experimental data (under Grant No. CMMI-0829978) and the Image-based automated 3D crack detection for post-
technical staff at the NEES Equipment Site at the University disaster building assessment. J Comput Civil Eng 2014;
at Buffalo. 28: A4014004.
12. Ellenberg A, Kontsos A, Bartoli I, et al. Masonry crack
Declaration of Conflicting Interests detection application of an unmanned aerial vehicle. In:
Proceedings of computing in civil and building engineering,
The author(s) declared no potential conflicts of interest with Orlando, FL, 23–25 June 2014, pp. 1788–1795. Reston,
respect to the research, authorship, and/or publication of this VA: American Society of Civil Engineers.
article. 13. Ebrahimkhanlou A, Farhidzadeh A and Salamone S.
Multifractal analysis of two-dimensional images for
Funding damage assessment of reinforced concrete structures. In:
The author(s) disclosed receipt of the following financial sup- Proceedings of SPIE, sensors and smart structures technol-
port for the research, authorship, and/or publication of this ogies for civil, mechanical, and aerospace systems (ed
92 Structural Health Monitoring 15(1)

Lynch JP), San Diego, CA, 8–12 March 2015, vol. 9435, 27. Reljin I and Reljin B. Fractal geometry and multifractals
p. 94351A. Bellingham, WA: SPIE. in analyzing and processing medical data and images.
14. Lopes R and Betrouni N. Fractal and multifractal analy- Arch Oncol 2002; 10: 283–293.
sis: a review. Med Image Anal 2009; 13: 634–649. 28. Theiler J. Estimating fractal dimension. J Opt Soc Am A
15. Yadav RP, Dwivedi S, Mittal AK, et al. Analyzing the 1990; 7: 1055–1073.
LiF thin films deposited at different substrate tempera- 29. Voss RF and Voss RF. Fractals in nature: from charac-
tures using multifractal technique. Thin Solid Films 2014; terization to simulation. In: Peitgen H-O and Saupe D
562: 126–131. (eds) The science of fractal images. New York: Springer,
16. Li L, Hu W, Liu L, et al. Evaluation of breast cancer che- 1988, pp. 21–70.
motherapy efficacy with multifractal spectrum analysis of 30. Weiss J. Fracture and fragmentation of ice: a fractal
magnetic resonance image. Biomed Mater Eng 2014; 24: analysis of scale invariance. Eng Fract Mech 2001; 68:
163–171. 1975–2012.
17. Chakraborty B, Haris K, Latha G, et al. Multifractal 31. Harte D. Multifractals: theory and applications. London:
approach for seafloor characterization. IEEE Geosci CRC Press, 2001.
Remote S 2014; 11: 54–58. 32. Rényi A. On measures of information and entropy. In:
18. Sarmiento E, Uruchurtu J, Sarmiento O, et al. Fractal Proceedings of the fourth Berkeley symposium on mathe-
analysis of the corrosion inhibition of carbon steel in a matics, statistics and probability, Berkeley, CA, 20 June–
bromide solution by lithium chromate. Int J Electrochem 30 July 1960, pp. 547–561. Berkeley, CA: University of
Sc 2009; 4: 144–155. California Press.
19. Moustafa A and Salamone S. Fractal dimension-based 33. Cao M, Ren Q and Qiao P. Nondestructive assessment
Lamb wave tomography algorithm for damage detection of reinforced concrete structures based on fractal damage
in plate-like structures. J Intel Mat Syst Str 2012; 23: characteristic factors. J Eng Mech 2006; 132: 924–931.
1269–1276. 34. Chhabra A and Jensen R. Direct determination of the
20. Bai R, Ostachowicz W, Radzieński M, et al. Vibrational f(a) singularity spectrum. Phys Rev Lett 1989; 62:
damage detection using fractal surface singularities with 1327–1330.
noncontact laser measurement. J Vib Control. Epub 35. Chhabra A, Meneveau C, Jensen R, et al. Direct determi-
ahead of print 17 September 2014. DOI: 10.1177/ nation of the f(a) singularity spectrum and its application
1077546314548088. to fully developed turbulence. Phys Rev A 1989; 40:
21. Lee E-T and Eun H-C. Damage detection of steel beam 5284–5294.
using frequency response function measurement data and 36. Luna B, Rivera J, Rocks J, et al. University at Buffalo—
fractal dimension. J Vib Acoust 2015; 137: 034503. low aspect ratio rectangular reinforced concrete shear
22. Farhidzadeh A, Dehghan-Niri E, Moustafa A, et al. wall—specimen SW1, 2013, https://nees.org/warehouse/
Damage assessment of reinforced concrete structures experiment/3316/project/676
using fractal analysis of residual crack patterns. Exp 37. Luna B, Rivera J, Rocks J, et al. University at Buffalo–
Mech 2013; 53: 1607–1619. low aspect ratio rectangular reinforced concrete shear wall–
23. Ivanov PC, Amaral LA, Goldberger AL, et al. Multifrac- specimen SW2, 2013, https://nees.org/warehouse/experi-
tality in human heartbeat dynamics. Nature 1999; 399: ment/3315/project/676
461–465. 38. Rocks JF. Large scale testing of low aspect ratio reinforced
24. Mandelbrot B. The fractal geometry of nature. New York: concrete walls. M.Sc. Thesis, The State University of New
Henry Holt & Co., 1982. York at Buffalo, Buffalo, NY, 2012.
25. Kamer Y, Ouillon G and Sornette D. Barycentric fixed- 39. American Society of Civil Engineers (ASCE)/SEI. ASCE
mass method for multifractal analysis. Phys Rev E 2013; standard 41-13: seismic evaluation and retrofit of existing
88: 022922. buildings. 2013th ed. Reston, VA: ASCE, 2014.
26. Falconer K. Fractal geometry: mathematical foundations
and applications. 3rd ed. Chichester: John Wiley & Sons,
2013.