Computers and Structures 190 (2017) 205–218

Contents lists available at ScienceDirect

Computers and Structures

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

A data processing methodology for infrared thermography images of

concrete bridges
Shuhei Hiasa a,b, Recep Birgul c,1, F. Necati Catbas a,⇑
a
Department of Civil, Environmental, and Construction Engineering, University of Central Florida, 12800 Pegasus Drive, Suite 211, Orlando, FL, USA
b
West Nippon Expressway Company Limited (NEXCO-West), Dojima Avanza 19F, 1-6-20 Dojima, Kita-ku, Osaka 530-0003, Japan
c
Mugla Sitki Kocman University, Turkey

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

Article history: This study presents a methodology to improve the usability and efficiency of infrared thermography (IRT)
Received 9 November 2016 for subsurface damage detection in concrete structures. A practical and more objective approach to
Accepted 28 May 2017 obtain a threshold for IRT data processing was developed by incorporating finite element (FE) model sim-
Available online 16 June 2017
ulations. Regarding the temperature thresholds of sound and delaminated areas, the temperature of the
sound part was obtained from the IR image, and the temperature of the delaminated area was defined by
Keywords: FE model simulation. With this methodology, delaminated areas of concrete slabs at 1.27 cm and 2.54 cm
Infrared thermography
depths could be detected more objectively than by visually judging the color contrast of IR images.
Non-destructive evaluation
Bridge deck inspection
However, it was also found that the boundary condition affects the accuracy of the method, and the effect
IRT data processing varies depending on the data collection time. On the other hand, it can be assumed that the influential
Finite element modeling area of the boundary condition is much smaller than the area of a bridge deck in real structures; thus,
it might be ignorable on real concrete bridge decks. Even though there are some limitations, this method-
ology performed successfully paving the way towards automated IRT data analysis for concrete bridge
deck inspections.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction inspection of every highway bridge in the National Highway Sys-

Degradation of reinforced concrete bridge decks is a widespread
problem in the United States. Concrete bridge decks deteriorate
faster than other bridge components due to direct exposure to traf-
fic. Moreover, the Federal Highway Administration’s (FHWA’s)
Long Term Bridge Performance (LTBP) Program identified that
the most important bridge performance issue was the performance
of concrete bridge decks [1]. Cracking, spalling, and delaminations
were common defects requiring maintenance and rehabilitation.
Most State Departments of Transportation (DOTs) noted that
repair, rehabilitation and replacement of concrete bridge decks
account for 50–80% of their budgets for maintenance of bridges.
These State DOTs are seeking methods to detect defects and quan-
tify the extent and severity of bridge decks early, accurately, and
rapidly with minimal traffic impact, ideally, without lane closure
for inspections [1,2]. Furthermore, FHWA [3] requires biennial
⇑ Corresponding author.
E-mail addresses: hiasa615@Knights.ucf.edu, s.hiasa.aa@w-nexco.co.jp
(S. Hiasa), rbirgul@gmail.com (R. Birgul), catbas@ucf.edu (F. Necati Catbas).
1 NDE methodology regarding data collection among the other
Formerly, Visiting Scholar at the University of Central Florida, 12800 Pegasus
Drive, Suite 211, Orlando, FL, USA.
tem (NHS). Therefore, Non-Destructive Evaluation (NDE) methods
are desired to inspect bridge decks efficiently and effectively. Con-
sequently, NDE techniques are being developed to examine and
monitor deteriorating structures rapidly and effectively [4]. Most
NDE methods aim to achieve the highest quality of visual imaging
of the relevant internal features of structures [5,6]. However, there
is still no international standard NDE methods for concrete bridges,
although significant progress has been made towards an interna-
tionally common approach to NDE inspection [6]. As Gucunski
et al. [1] reports, one of the limitations of NDE methods for bridge
inspection is the speed of data collection.
Infrared thermography (IRT) is one of the NDE methods and has
been developed to detect invisible deteriorations including delam-
inations and voids in concrete structures with reasonable accu-
racy; it also helps avoid the time and expense of gaining
immediate access to the concrete surface to conduct traditional
sounding tests. IRT is a suitable approach for inspection of civil
infrastructures since it is a non-contact method and infrared (IR)
images can instantly portray a wide range of concrete structures
at one time [7–10]. Therefore, IRT can be the fastest and easiest
NDE methods, even though there are some limitations and

http://dx.doi.org/10.1016/j.compstruc.2017.05.011
0045-7949/Ó 2017 Elsevier Ltd. All rights reserved.
206 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
uncertainties in using IRT for bridge inspections such as data col-
lection time, size of delamination, camera specifications, data col-
lection speed and data interpretation.
The objective of this study is to develop a methodology of how
to objectively interpret and detect delaminations from IR images
since it becomes very subjective judging whether or not the color
contrast of the image is a damage indication. As Washer et al.
[11] argued, if the temperature span for IR images is setup too high
or too low, it appears in the IR image as if there is no anomaly even
though there are some defects. Therefore, they recommended
adjusting the temperature span of IR images continuously through-
out inspections. However, it might require a lot of work during or
after the bridge inspection. This study consequently explores a
more objective method than just comparing IR images to assess
IRT data. Kee et al. [12] and Oh et al. [13] processed IRT data math-
ematically by using MATLAB with certain thresholds defined by
iterative trials until the operator obtained the clearest contrast
between the sound and delaminated regions within each IR image.
However, this procedure is very subjective since the operator has
to determine whether the contrast depicts damaged or sound
regions, even though these regions are usually unknown areas in
terms of existing defects. Processing IRT data mathematically is
more objective than judging the data from the color contrast since
it does not require a temperature span setting as mentioned above.
However, how to determine the thresholds, in other words, how to
obtain the information of temperature difference between sound
and delaminated areas becomes a challenge for their methodology
to process mathematically without subjective trials.
Hiasa et al. [14] proposed a data processing method; however, it
was also discussed how obtaining the information of temperature
difference between sound and delaminated areas becomes a chal-
lenge. In this study, an easier and more objective method to obtain
the threshold for IRT data processing is explored by incorporating
finite element (FE) model simulations. The use of FE model simula-
tion has been increasing recently to simulate the temperature dis-
tribution of the object’s surface [7,15–19]. In the past study, Hiasa
et al. [20] employed the FE model simulation to explore sensitive
parameters for effective utilization of IRT without a large number
of experiments, which require extremely time-consuming work.
This study uses the FE model developed in the past study. The
aim of this study is twofold: to obtain information regarding the
temperature difference between sound and delaminated areas
from FE modeling, and to process IRT data in order to objectively
detect invisible subsurface delaminations. In this study, the IRT
data obtained from a field laboratory experiment under passive
IRT conditions [21] are used to develop a more objective data pro-
cessing method with the combination of FE model simulation.
2. Current practice and future potential of infrared
thermography for bridge inspection
Through literature reviews, several factors that might affect the
performance of IRT can be excerpted such as data collection time,
size of delamination, data collection speed and IR camera specifica-
tions. In terms of data collection time, Washer et al. [22,23] recom-
mended daytime measurements of 5–9 h after sunrise to detect
subsurface delaminations for the solar loading part. On the other
hand, Gucunski et al. [24] mentioned that a thermal image
recorded 40 min after sunrise yielded a much clearer image than
another one recorded around noon. Additionally, Yehia et al. [25]
found that the response of delaminations were described as
weaker in IR images as the time approached 3 PM. Moreover, Kee
et al. [12] also concluded that no indication was found from the
IR images taken 3 h and 45 min after sunrise (with the shallowest
delamination located at 6.35 cm depth) while the best results were
achieved using the cooling cycle in which even 15.24 cm deep
delaminations could be detected. Furthermore, Watase et al. [26]
proposed a favorable time for inspection depending on the parts
of the bridge; noon time for the deck top, and midnight for the
deck soffit. Through a field laboratory experiment and FE modeling,
Hiasa [27] concluded in his Ph.D. dissertation that the preferable
time period to apply IRT for concrete bridge deck inspection is dur-
ing the nighttime cooling effect in order to reduce the possibility of
misdetection due to sunlight, even though the delaminated areas
were also observed clearly during the daytime heating period in
the study. Hiasa [27] also found that there are interchange periods
between the nighttime cooling effect and the daytime heating
effect about 1–2 h in the morning and evening under the given
conditions, and IRT cannot detect delaminations during these peri-
ods. This can be considered as the cause that several researchers
concluded different results; the reason why Kee et al. [12] could
not detect delaminations from the IR images taken 3 h and
45 min after sunrise can be due to that time was the interchange
period from cooling to heating cycles. Regarding Yehia et al. [25],
3 PM might be close to the interchange period from heating to
cooling cycles under the experimental condition. It can also be con-
sidered that Gucunski et al. [24] captured clear IR images during
the cooling cycle effect in the early morning while Washer et al.
[22,23] took IR images during the heating cycle effect.
Regarding the effect of delamination size, some researchers
indicate that the size of delamination affects the detectable depth
of the delamination [25,28–30]; however, different size of artificial
defects have been utilized in their studies and they reported differ-
ent detectable depths using IRT for inspections. While Cheng et al.
[29] could not detect even 3 cm depth of delamination
(5
5
7 cm in size) in their experiment, Kee et al. [12] detected
15.24 cm depth of delamination (61
61
0.1 cm in size) in their
studies. The past studies on IRT were conducted with limited
experimental setups and limited conditions which make a differ-
ence in delamination detection due to the difficulty of making a
large number of test specimens since those specimens become rel-
atively huge to simulate concrete bridges. In order to overcome the
limitation, Hiasa et al. [20] utilized FE model simulation to explore
sensitive parameters for effective utilization of IRT. Through the FE
model simulations, it was found that the most critical factor is the
area of delamination; subsequently, the thickness affects the tem-
perature differences of the surface between sound and delami-
nated areas. The volume of delamination is not a significant
parameter for interior damage detection using IRT. In addition,
the FEM analysis also shows that as the area is getting larger, the
impact of the thickness is also increasing. Therefore, differences
in delamination size can be considered as the cause that past
research concluded different detectable depths by IRT; smaller
delaminations were detected at only shallower depths while larger
delaminations were detected at deeper locations. As it was clarified
by FE model simulation, the detectability is highly dependent on
the size of delamination. However, this can also be considered that
as a defect increases the severity by widening the area, the easier it
is to be detected by IRT. Usually, bridge administrators make main-
tenance plans in order of severity of bridge conditions, and even if
they find a tiny/light defect, they might leave it as it is and keep
monitoring it for several years until it becomes a severe defect.
Therefore, even if IRT cannot detect small and/or deep defects
which can be taken into account as minor damage at that time,
the limitation is not a serious problem since those defects do not
require immediate repair work [27].
In regards to the effect of data collection speed and camera
specifications, Hiasa et al. [21,31] conducted comparative studies
with three different types of IR cameras at a normal driving speed,
from 48 to 64 km/h (30–40 mph). It was found that if shorter inte-
gration time devices, which cooled type cameras are typically
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
equipped with, are used for high-speed scanning of concrete bridge
decks, they are not affected by data collection speeds. Furthermore,
high-speed scanning by IRT with a cooled type IR camera per-
formed at a similar or better level of accuracy compared to other
NDE methods such as impact echo (IE), ground-penetrating radar
(GPR) and chain drag [31]. Therefore, it is possible to collect reli-
able data rapidly without the need for lane closures by using
appropriate IR cameras attached to vehicles, thereby eliminating
the need for sounding inspection or other NDE techniques includ-
ing robotics-assisted system, which mostly requires lane closures.
Thus, IRT has a potential to become a more practical and faster
method of conducting bridge deck inspections than other NDE
technologies to detect subsurface defects. As for the data collection
speed of robotics-assisted bridge inspection system, an average
data collection speed by this system is estimated to be 350 m2 of
bridge deck area per hour [1]. On the other hand, the data collec-
tion speed of high-speed scanning using IRT depends on the driv-
ing speeds; if the vehicle is driving at 80 km/h, IRT can scan
about 350 m2 of bridge deck area in only 4.5 s, and the productivity
is 800 times higher than that of a robotics-assisted system. In this
case, IRT can capture more than one lane width, about 3.5 m, at one
time, such that the scanning speed can be calculated from the driv-
ing speed. This is much faster than the robotics-assisted system
and traditional sounding inspection, and would be feasible for
biennial inspection intervals since the total number of bridges in
the USA is 611,840 and the total deck area is 369 km2 (concrete
deck area: 343 km2) as of 2015, and the number is increasing year
by year [31].
It can be construed that IRT is not a perfect method for damage
detection of concrete bridge decks because sometimes it might fail
to detect some defects depending on several conditions which
might affect the detectability of IRT; e.g. small delaminations at
deep locations, unsuitable data collection time, inadequate IR cam-
era specifications and misinterpretation of IRT data. However, the
data collection speed is very attractive for periodic bridge inspec-
tions; hence, there is a great possibility to reduce inspection time,
labor and budget drastically if high-speed bridge deck scanning by
IRT becomes a standard bridge deck inspection method. In order to
complement the imperfections of IRT, other NDE methods such as
chain drag, hammer sounding, IE and GPR should also be utilized
for a more comprehensive and detailed diagnosis of concrete
bridge decks with longer intervals than the current requirement
of two-year intervals. For example, if those complementary but
time-taking and expensive technologies are utilized at 4–6 year
intervals, or when critical defects are found by high-speed IRT
scanning, bridge deck inspection would become more effective,
efficient, economical and practical than current practice.
3. Summary of past studies
The present study aims to develop an IRT data processing
methodology to objectively detect delaminations using FE model
simulation. For that purpose, this study uses past experimental
data and FE models developed in the previous study. This section
summarizes these past studies.
3.1. Field laboratory experiment under passive IRT conditions
Previously Hiasa et al. [21] conducted a field test with manufac-
tured concrete blocks which had artificial delaminations at differ-
ent depths from the surface, 1.27 cm (0.5 in.), 2.54 cm (1 in.),
5.08 cm (2 in.) and 7.62 cm (3 in.). The dimensions of each delam-
ination were 10.2 cm (4 in.)
10.2 cm
approximately 0.3 cm
(1/8 in.). All four concrete blocks had the same dimensions as
91.4 cm (3 ft.)
91.4 cm
20.3 cm (8 in.), and the thickness was
207
designed to simulate a typical bridge deck in the USA. In this exper-
iment, three different types of IR cameras, SC5600 (cooled detector
type camera), T650 and T420 (uncooled detector type camera)
manufactured by FLIR, were utilized to explore the effects of cam-
era specifications along with data collection speed on IRT from a
stopped and moving vehicle. The results showed that IRT with
SC5600 can detect up to 2.54 cm depth of delamination at any time
period of a day, 9 AM, 3 PM, 8 PM and 12 AM, under both static and
dynamic data collection situations. The reason why 5.08 cm and
7.62 cm depths of delaminations were not detected can be consid-
ered due to the small size of delamination, 10.2
10.2
0.3 cm, as
mentioned in Section 2.
Fig. 1 shows raw IR images of each concrete slab taken by
SC5600 at 3 PM and 12 AM under static situation in the field test
[21]. In this study, only IRT data taken by SC5600 is used for data
processing since it provided better results than the other two cam-
eras. The temperature ranges of the IR images are set up to 10 K for
the images taken at 3 PM and 5 K for the images taken at midnight
due to the difference of temperature distribution. Two types of
delamination were embedded at the center and the corner of the
block (upper right side for 1.27 cm depth, lower right side for
others). In this study, when the concrete blocks were casted, carry-
ing handles were set up at the two corners on a diagonal line to a
reversed position of the corner delamination by reinforcing bars
(upper left and lower right corner for 1.27 cm depth specimen,
upper right and lower left corner for others). Furthermore, thermo-
couples were attached on the surface of concrete slabs with
1.27 cm and 7.62 cm depth of delaminations. Since they have a dif-
ferent temperature distribution from the concrete surface due to
different materials and colors, they were not considered as delam-
inations even though IR images indicated different colors. More-
over, since the temperature distribution of the concrete surface
was not entirely homogeneous due to boundary conditions and
the effect of sunlight, some temperature differences were judged
as noise or errors, namely, they were evaluated as sound areas.
However, this method becomes relatively subjective because the
locations of the delaminated parts were known beforehand. There-
fore, this study aims to obtain defect detection results based on a
more objective judgment methodology.
3.2. FE model of the concrete block
Since the results of the field experiment regarding detectable
depth of subsurface delaminations contradict with the results of
other research publications due to fact that delaminations deeper
than even 5.08 cm were reported as being detected by IRT
[12,23,25], Hiasa et al. [20] used FE modeling to explore the cause
of contradicting conclusions in regard to detectable depth of
delaminations by IRT. Surface temperatures of the concrete test
specimens were modeled and simulated using Heat Transfer Mod-
ule of COMSOL Multiphysics software as shown in Fig. 2. In this
model, the same size of concrete model, 91.4
91.4
20.3 cm,
was placed on larger ground, 6
6
1 m, and two pieces of styro-
foam, 10.2
10.2
0.3 cm, were installed inside the concrete as
described in Fig. 3. The depths from the surface were the same as
the experiment, 1.27 cm, 2.54 cm, 5.08 cm and 7.62 cm. The orien-
tation of the concrete block was also set up in almost the same way
as the experiment shown in Fig. 3. The picture on the left shows the
real concrete blocks used in the experiment, and the images on the
right depict the numerically modeled concrete block using FEM. In
the experiment, concrete blocks were placed on wooden stands to
make space through which wind blows, and those were set up on
wooden pallets as shown in the picture. However, in the model, a
concrete block was placed on concrete stands, which have the
same height (20 cm) as wooden stands and the pallet for simpler
modeling and simulation. In terms of the mesh size, the ‘‘Finer”
208 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218

Fig. 1. IR images of each depth of delamination taken at 3 PM (1st row), and 12 AM (2nd row).

Fig. 2. Simulation results of each depth of delamination for 3 PM (1st row), and 12 AM (2nd row).

element size was selected for the concrete block in the COMSOL
software among the several predefined mesh sizes for more accu-
rate characterization. Rumbayan and Washer [15] concluded that
there is a balance between computational economy and accuracy
in solution, which is true for any FE modeling application. As for
the styrofoam, the stands of the concrete block and the ground, a
‘‘Coarser” element size was selected for economy since it can be
considered that those materials are not sensitive for concrete sur-
face temperature. The material properties were set up as shown in
Table 1.
In this model, the primary heat source was solar radiation, and
the direction of the sunlight (zenith angle and the solar elevation)
over the simulation time was automatically computed from the
longitude, latitude, time zone, date, and time by the software
[32]. In this simulation, the information was set up as follows: lon-
gitude; 81.190250, latitude; 28.591463, time zone; 5 h. The
location is the test field of the previous experiment and the loca-
tion information was referred to Google Maps [33]. It was assumed
that there was no cloud cover through the simulation period. The
value of solar irradiance was set up by sensitivity analysis. Further-
more, there were three kinds of ambient temperature conditions in
this model. (1) The ambient air temperature was set up based on
the measured temperature during the experiment (9 AM to 12
AM) and weather history of the climatological substation near
the test site (12 PM to 8 AM), derived from a weather database
on January 21, 2015 [34]. (2) Convective heat transfer coefficient
was also defined by sensitivity analysis based on wind speed data
from the climatological substation near the test site. (3) The
ground at 1 m below the surface was assumed to be at a constant
temperature of 22.5 °C throughout the day, corresponding to the
mean annual earth temperature around the central Florida area
[35]. Through sensitivity analysis, the following conditions were
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
Fig. 3. Concrete blocks and the FE simulation model.
Table 1
Material properties of the simulation model.
Material properties Unit
Thermal conductivity W/(m K)
Heat capacity at constant pressure J/(kg K)
Density kg/m3
defined in the simulation: Solar irradiance is 1300 W/m2; Heat
transfer coefficient is 20 W/m2 K [20].
Fig. 4 shows the comparison between IRT results obtained by
three IR cameras and results of temperature differences obtained
by FE simulation. In this comparison, the center part of the delam-
ination was used to determine the temperature of the delaminated
part, while a point 5 cm away from the edge of the delamination
was used to determine the temperature of the sound part in the
simulation results. As to the temperatures of IRT results, three dif-
ferent points in what can be considered the center part of the
delamination and in another area 5 cm away from the edge of
the delamination were used to determine the average tempera-
tures since the exact location of the delamination cannot be seen
from the IR images and each pixel of the IR image has slightly dif-
ferent temperatures. Through the FE model simulation, the numer-
ical model simulated the temperature differences between
delaminated and sound parts very accurately, less than 1 °C of
accuracy as shown in Fig. 4. However, temperatures of each point
at delaminated and sound areas have some different values
between simulation and infrared results, even though those differ-
ences of each IRT data and simulation results are not big (at most
only a few degrees Celsius). Actually, even the IRT data results var-
ied depending on which IR camera was used. Therefore, it is diffi-
cult to determine exact temperatures of sound and delaminated
areas from only FE model simulation, yet it is possible to obtain
the reliable value of temperature difference between sound and
delaminated areas from the simulation.
4. Processing IRT data
4.1. Application of FE model simulation to define thresholds for data
processing
The present study utilizes FE model simulations to obtain
thresholds for IRT data processing more easily and objectively than
the method Kee et al. [12] and Oh et al. [13] conducted with iter-
ative trials. In this study, the same FE model and conditions
209
Concrete Styrofoam Ground
1.6 0.024 0.6
880 1130 800
2300 25 1500
explained in Section 3 were used to obtain temperature differences
between sound and delaminated areas. Fig. 5 shows some simula-
tion results at 9 AM, 3 PM, 8 PM and 12 AM (midnight). Since the
surface temperatures are not homogeneous due to effects of sun-
light, delaminations and boundary conditions as shown in Fig. 5,
temperatures of several points of a simulated concrete surface
were compared to examine how temperatures distribute on the
concrete slab, and to decide the reference point of the temperature
at the sound area in the FE model simulation.
Figs. 6 and 7 describe the results of the comparisons at several
points. Eight points with 1.4 cm apart from the center of the
delamination up to 9.8 cm away (4.7 cm away from the edge of
the delamination) were plotted and each point was numbered
from 1a to 8a from the center point as shown in Fig. 6(a). Fig. 6
(b) depicts temperature differences between each adjacent point.
It can be assumed that a slightly larger/smaller area than the real
delamination size shows the temperature difference on the con-
crete surface, and other sound areas have almost the same temper-
ature as shown in Fig. 8. Therefore, several adjacent points of
temperatures were compared to verify the assumption. From this
comparison, it was observed that temperature differences were
increased as the point became farther from the center in the
delaminated area while the difference became maximum when
the temperature difference was compared between sound (4a)
and delaminated (5a) parts, ‘‘4a–5a” in the graph, and those were
decreased as the point became farther from the center in the sound
area. This result indicates that the highest (lowest) temperature
during daytime (nighttime) is the center part of the delamination,
and the temperature is decreasing (increasing) gradually as the
point being away from the center even if it is a delaminated area.
Furthermore, the last three points, 6a, 7a and 8a, indicated almost
the same temperature; 0.09 °C at most between 6a and 7a, and
0.01 °C at most between 7a and 8a. Since the point of 5a, which
is located at 0.5 cm away from the edge of the delamination, shows
a different temperature from the other sound areas (6a is located
1.9 cm away from the edge), it can be assumed that a slightly wider
area, about 1 cm from the edge in this case, than the real delami-
210 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218

Fig. 4. Comparison of temperature differences between FE modeling and IRT results.

Fig. 5. Surface temperature distributions at each time obtained from FE model simulation.

nation size shows the temperature difference on the concrete sur-
face, and other sound areas have almost the same temperature as
drawn in Fig. 8. In this model, the period of daytime heating effect
was about 10.5 h, starting from between 7 AM and 7:30 AM until
between 5:30 PM and 6 PM, and night time cooling effect was
about 13.5 h from midnight to morning and evening to midnight.
In addition, Fig. 7(b) shows the temperature difference between
the center part of the delamination, 1b, and 5 cm away from the
edge of the delamination, 2b, and differences among other points
located in the sound area with intervals of 5 cm from 2b as
depicted in Fig. 7(a). In this case, the sound area up to 20 cm away
from the edge of the delamination, from 2b to 5b, shows an almost
homogeneous temperature distribution, the difference is 0.06 °C at
most; however, the last three points, 6b to 8b, showed different
temperatures from these four points, 2b to 5b. Those differences
increased as the point became father from the center, and those
maximum differences of ‘‘5b–6b”, ‘‘6b–7b” and ‘‘7b–8b” were
0.10 °C, 0.18 °C and 0.32 °C respectively. It can be considered that
these last three points were affected by the boundary conditions.
Therefore, it can be concluded that the boundary condition affects
about 15 cm from the edge of the concrete block under given con-
ditions in this model simulation.
Through the comparison of temperatures at several points of
the concrete surface, the temperature of the sound area can be
regarded as identical temperature except for the areas surrounding
the delamination and outside of the concrete block about 15 cm
from the edge; thus, the point of 23 cm from the center, the mid-
point of an imaginary line that connects the center of the delami-
nation and the edge of the concrete slab, was chosen as the tem-
perature of the sound area, and the temperature of the center of
delamination was selected as the temperature of the delaminated
area. In this way, temperature differences between sound and
delaminated areas were obtained by the FE model simulation.
4.2. Methodology of data processing
In this study, MATLAB is used to analyze IRT data mathemati-
cally for processing and judging data more objectively than by just
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218 211

Fig. 6. Surface temperature differences (from center to 9.8 cm away with intervals of 1.4 cm).

Fig. 7. Surface temperature differences (5 cm intervals from the edge of the delamination).

comparing the color contrast of IR images. Since every IR image where T(x, y) is the experimental temperature value of each ele-
consists of a group of pixels, and each pixel has a numerical value, ment, T(del.) is the temperature of the delaminated part, T(sound)
IRT data can be manipulated mathematically. Theoretically, the is the temperature of the sound area and F(x, y) is the specified
concrete surface above the delamination becomes warmer than value of each element. However, how to determine these two
the surroundings during the daytime, and the area over the delam- thresholds has been a challenge in the past studies [12–14]. The
ination appears cooler during the nighttime [36]. Therefore, two methodology of how to obtain these two values will be discussed
types of calculation are conducted in this method depending on in Section 4.3. Please note that ‘‘0” value is represented by a ‘‘black
the time of data collection in order to obtain the same output pixel” indicating damage.
despite of the daytime heating or nighttime cooling effect. Secondly, Eq. (2) is utilized for daytime heating effect in order to
Firstly, Eq. (1) is applied for the nighttime cooling effect. IRT display the delaminated area as black and the sound area as white,
data are specified into a range of gray levels, such that all values which is the same as the nighttime cooling effect.
lower than those for the delaminated area are depicted black (0 8
> 0 ðif Tðx; yÞ P Tðdel:ÞÞ
in MATLAB) and all values greater than the sound area are drawn <
as white (1 in MATLAB) and all intermediate values are scaled Fðx; yÞ ¼ 1 ðif Tðx; yÞ 6 TðsoundÞÞ ð2Þ
>
: Tðx;yÞTðsoundÞ
within the range as given in Eq. (1). ðif Tðdel:Þ > Tðx; yÞ > TðSoundÞÞ
Tðdel:ÞTðsoundÞ
8
> 0 ðif Tðx; yÞ 6 Tðdel:ÞÞ Furthermore, binary images are yielded for noise reduction by
<
Fðx; yÞ ¼ 1 ðif Tðx; yÞ P TðsoundÞÞ ð1Þ displaying only values under certain thresholds of the specified
>
: Tðx;yÞTðdel:Þ range as black (0 in MATLAB) and the others are drawn as white
TðsoundÞTðdel:Þ
ðif Tðdel:Þ < Tðx; yÞ < TðsoundÞÞ
(1 in MATLAB) as shown in Eq. (3).
212 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218

Fig. 8. Image of surface temperature of delaminated and sound areas (left: daytime, right: nighttime).

Table 2
Simulation results at each point and the difference (Temperature: °C).

Time 1.27 cm (0.5 in.) 2.54 cm (1 in.) 5.08 cm (2 in.) 7.62 cm (3 in.)
Defect Sound Difference Defect Sound Difference Defect Sound Difference Defect Sound Difference
9 AM 16.02 15.16 0.85 15.61 15.16 0.45 15.31 15.16 0.15 15.22 15.16 0.06
3 PM 27.02 25.60 1.43 26.45 25.60 0.85 25.93 25.60 0.32 25.74 25.62 0.12
8 PM 17.67 18.47 0.80 18.11 18.46 0.35 18.43 18.47 0.04 18.53 18.48 0.05
12 AM 14.13 14.78 0.66 14.47 14.78 0.31 14.76 14.78 0.02 14.86 14.78 0.08

Fig. 9. Areas where temperature readings were picked randomly.

Table 3
IRT readings at each point and the difference (Temperature: °C).

Time 1.27 cm (0.5 in.) 2.54 cm (1 in.) 5.08 cm (2 in.) 7.62 cm (3 in.)
Point 1 Point 2 Point 3 Point 4 Point 1 Point 2 Point 3 Point 4 Point 1 Point 2 Point 3 Point 4 Point 1 Point 2 Point 3 Point 4
9 AM 11.52 11.57 11.46 11.56 11.61 11.71 11.76 11.66 11.45 11.46 – – 11.76 11.82 11.84 11.61
3 PM 26.14 26.24 26.22 26.10 25.44 25.31 25.77 25.66 25.46 25.23 25.52 25.33 25.73 25.62 25.89 25.78
8 PM 15.45 15.74 15.84 15.72 15.18 15.44 15.64 15.40 15.18 15.54 15.63 15.47 15.29 15.53 15.67 15.42
12 AM 11.71 11.94 12.19 11.91 11.79 12.08 12.16 11.85 11.95 12.19 12.35 11.96 11.68 11.91 11.95 11.79


0 ðifFðx; yÞ 6 0:1 to 0:6 reduction procedure is relatively subjective, yet this is more
Gðx; yÞ ¼ ð3Þ
1 ðothersÞ objective to judge a delaminated area than to decide from raw
IR images since the indication is displayed as black while the
where G(x, y) is the value of each element of binarized image. sound area is shown as white. In this study, these thresholds of
Then, from those images, an image which has relatively less specified temperature are determined from the FE model
noise and clearer delamination indication is chosen. This noise simulations.
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218 213

Table 4 delamination, 1.27 cm in this model, causes the highest tempera-

Established temperature span at each time (Temperature: °C). ture difference and the others are within the range, the specified
Time T(sound) T(del.) temperature span is set up by the difference of 1.27 cm deep
9 AM 11.63 12.48 delamination model. Based on these values of temperature differ-
3 PM 25.72 27.14 ences at each time, which is obtained from FE model simulation,
8 PM 15.51 14.71 the IRT data processing method is applied to detect delaminations.
12 AM 11.96 11.31 In terms of the reference temperature of the sound or delaminated
point, the sound part of temperature is obtained from the IR image.
Since the location of the delaminated area is usually unknown and
4.3. Temperature span setting
this method is expecting to detect the delaminated part, the tem-
perature of the sound area should be chosen for the standard value.
Table 2 summarizes simulation results of both temperatures of
Therefore, the specified temperature span DT and temperature of
defective and sound areas at each time for each depth of delamina-
the sound area TðSoundÞ are set up as follows;
tion. The differences between those two temperatures are also
listed in the table. It is obvious that the surface temperatures of
the sound part for each model are the same at the same time in (a) DT ¼ TðsoundÞ  Tðdel:Þor DT ¼ Tðdel:Þ  TðsoundÞÞ: From the
spite of the depth of delamination, and the differences are FE model simulation
decreasing as the depth becomes deeper. Since the shallowest (b) TðSoundÞ: From experimental IRT data

Fig. 10. IR images of each depth of delamination taken at (1) 9 AM, (2) 3 PM, (3) 8 PM, (4) 12 AM.
214 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
Fig. 11. Specified images by the established temperature spans.
Then, temperature of delaminated area can be obtained as
defined in Eq. (4)

Tðdel:Þ ¼ TðSoundÞ þ DT ðdaytimeÞ
ð4Þ
Tðdel:Þ ¼ TðSoundÞ  DT ðnighttimeÞ
Regarding the temperature readings from IR images, tempera-
tures of four points at four separate areas as enclosed by dotted
lines in the left image of Fig. 9 were picked up from an IR image.
Since each IR image shows the effects of boundary conditions at
the edge parts, the areas close to the center part were chosen for
reference temperatures of sound parts. Even though 5.08 and
7.62 cm deep delaminations were not detected from thermal
images, four similar locations which should be sound areas were
selected randomly because the size and location of delaminations
are known. Since this procedure is relatively subjective, tempera-
tures were read from four different areas as shown in the figure,
and the average value was used for the data processing to reduce
the subjectivity. However, within the IRT data, the image of
5.08 cm deep delamination taken at 9 AM indicates a tree shadow
at the right lower side as shown in right figure of Fig. 9. Therefore,
only for this image, temperatures of only two areas, which are not
affected by the shadow were selected for data input as shown in
the right figure of Fig. 9. Table 3 lists temperatures from IR images
of each concrete blocks which have different depths of delamina-
tions at each time. Since these four test specimens were closely
aligned on a roadside as shown in Fig. 3, every point indicates very
close temperature at the same time irrespective of the test speci-
mens, which have different depths of delaminations. Therefore,
TðSoundÞ was also obtained from the average temperature of every
test specimen at each time, and both TðSoundÞ and Tðdel:Þ are sum-
marized in Table 4.
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218 215

Fig. 12. Processed images.

5. Results of the IRT data processing methodology 6. Discussion

Based on the temperature spans defined in Table 4, each IRT
data was specified into the temperature range. In this study, every
image was smoothened by Gaussian filter as described in Eq. (5) to
reduce noise [37].
 2  hand, processed images converted into binary images indicate only
1 x þ y2
Fðx; yÞ ¼ pffiffiffiffiffiffiffi exp  ð5Þ
2pr2 2r 2
In this study, 9
9 matrix with r = 1.5 was used for the filter.
Fig. 10 shows IR raw images and Fig. 11 displays those specified
and filtered images. Furthermore, these images were converted
into binary images as described in Eq. (3) to reduce noise, and
Fig. 12 shows those converted images. In each figure of Figs. 10–
12, the locations of delaminations are enclosed by a dotted line.
As can be seen in Fig. 11, there are a lot of gray level indications,
making it is ambiguous to judge whether those are sound or
delaminated areas. A different evaluation might be obtained
depending on the person who judges those images. On the other
two colors, white or black as shown in Fig. 12; thus, the judgment
becomes objective regardless of the person who evaluates pro-
cessed images.
In terms of the damage detection capability of this method,
every delamination at the middle part of the concrete block at
1.27 cm and 2.54 cm deep can be successfully detected by this data
processing methodology, while delaminations at the corner of the
specimens at 1.27 cm and 2.54 cm deep can be detected clearly
216 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218

only from the processed images taken at 9 AM as displayed in conditions can be assumed approximately within 40 cm distance
Fig. 12. Images taken at 3 PM and 12 AM indicate a lot of black from the edge at any time. Since the areas of the bridge deck are
areas, meaning those areas are considered as delaminated regions much larger (practically infinite) than the potentially influential
even though those are sound areas. This can be considered that areas of the boundary condition, it can be assumed that the bound-
temperatures outside of the concrete slabs were strongly affected ary condition might be ignored for bridge deck inspections. There-
by the boundary condition, and those areas were heated up or fore, this data processing methodology with FE model simulation
cooled down faster than the other areas. At 3 PM, each image dis- has a potential to bring great improvements towards automated
plays a large black area on the left side, which means the left side analysis of large IR images for bridge deck inspections.
temperature is higher than the established temperature range. On
the other hand, the right side temperature of the specimens at 12
7. Advantage of FE model simulation
AM is lower than the temperature span.
In the past experiment [21], these concrete blocks were set up
Even though there are some limitations regarding boundary
with the left side facing south; thus, the left side of the concrete
conditions in this method, combined use of FE model simulation
was heated up faster than the right side due to the sun loading dur-
with IRT showed that the combination can improve efficiency of
ing the daytime, and the right side was cooled faster than the left
IRT for concrete bridge inspections to detect subsurface delamina-
side during the nighttime. This can be attributed to so-called ‘tem-
tions. Fig. 13 displays scaled original IRT data image at 8 PM for
perature leakage’ at the boundaries of the test specimen. Therefore,
1.27 cm deep delamination (left) and specified range image of
this processing method has a limitation due to the fact that the
the same condition with data smoothing (right). Obviously, the
edge areas of concrete structures are affected by the boundary con-
specified range image shows much clearer indications of delamina-
ditions. It should be noted that when this method is employed at
tions than the original scaled image (the bottom indication is the
the edge part of the structure, the processed images might indicate
thermocouple attached on the surface). This is because the temper-
those areas as delaminated areas. However, the size of the test
ature difference between the sound and delaminated area is much
specimens used in this study is only 91.4
91.4 cm, and the mid-
smaller than the concrete surface temperature as shown in the left
dle part of the delaminations were detected successfully even at 3
graph of Fig. 14. This graph shows temperatures of each element at
PM and 12 AM. Therefore, the influential parts of the boundary
the middle part of the original scaled image as a drawn dotted line

Fig. 13. Original scaled image and specified range image.

Fig. 14. Original temperature values and specified range values at the middle part of each image (Horizontal axis shows the location of the image which is equally divided
into 430).
 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
in Fig. 13. As can be seen from the graph, the delaminated part
shows a slight dimple at the middle part; thus, delaminations of
the original scaled image are unclear.
On the other hand, the right graph of Fig. 14 depicts the delam-
inated area clearly at the middle part. This graph presents the ran-
ged values of each element at the middle part of the specified range
image as described by a dotted line part in Fig. 13. Therefore, by
specifying a certain temperature range with FE model simulation,
it becomes much easier to detect a delaminated area not only for
this data processing methodology, but also for other image pro-
cessing methodologies such as feature extraction, image segmen-
tation and edge detection techniques. FE modeling can provide
crucial information to decide the temperature range for specifying
the original data more objectively and accurately than judging data
from the color contrast of IR images. Since the temperature differ-
ence between sound and delaminated areas can be simulated very
accurately using FE modeling, IRT data can be manipulated math-
ematically with the temperature difference to detect delaminated
area without any subjective trials.
Through the FE simulation procedure described in this study, an
inspector can lay out a much more productive inspection program
ahead of time for a specific bridge. For instance, the inspector can
obtain the longitude, latitude, and the time zone information for
the specific bridge; this information would provide the necessary
input for the software to calculate the solar radiation as the heat
source radiating towards the bridge deck. Then, the inspector
would predetermine the target defects that are to be detected
inside the deck, in terms of the size and the depth from the surface.
Evaluating the results of FE simulation, the inspector would be able
to decide the most favorable time window to conduct the inspec-
tion for that specific bridge for the predetermined types of defects.
Since the IRT data is collected under a controlled
setup towards specific flaws, the MATLAB code can be readily
applicable to the IRT data obtained from the bridge deck; in turn,
the end results would be reported in an automated fashion. There-
fore, the utilization of FE model simulation has the potential to
facilitate automated image processing of large IRT data sets effec-
tively for bridge deck inspections to detect interior damage.
8. Conclusions
In this study, a methodology that combines numerical modeling
and IRT data is presented to improve the usability and efficiency of
data analysis, possibly leading to automated analysis and evalua-
tion. Kee et al. [12] and Oh et al. [13] processed IRT data mathemat-
ically by using MATLAB with certain thresholds defined by iterative
trials until the operator obtained the clearest contrast between the
sound and delaminated regions within each IR image. Processing
IRT data mathematically is more objective than judging data from
the color contrast since it does not require a subjective tempera-
ture span setting as Washer et al. [11] mentioned. However, how
to determine the thresholds; namely, how to obtain the informa-
tion of temperature differences between sound and delaminated
areas becomes a challenge for the previous studies.
In order to obtain thresholds for data processing, FE model sim-
ulation was utilized. This study used the FE model developed in the
previous research [20]; the IRT data obtained from a past study
[21] under passive IRT conditions were used to develop a more
objective data processing method with the combination of FE
model simulation. In terms of the temperature thresholds of sound
and delaminated areas, the temperature of the sound part was
obtained from the IR image, and the temperature of the delami-
nated area was defined by FE model simulation.
By using this method, delaminated areas of concrete slabs at
1.27 cm and 2.54 cm deep could be detected more objectively than
217
by judging the color contrast of IR images. However, it was also
found that the boundary condition affects the accuracy of the
method (due to so-called ‘temperature leakage’), and the effect
varies depending on the data collection time; the effect was the
least at 9 AM, and it was the largest at 3 PM and 12 AM in this
study. Therefore, it is important to find a suitable time for data col-
lection in order to reduce misdetection by this method. On the
other hand, it can be assumed that the influential area of the
boundary condition is much smaller than the area of a bridge deck
in real life situations. Thus, it might be ignorable on real concrete
bridge decks since those areas which would be affected by the
boundary conditions on bridge decks are very limited compared
to the total bridge deck area. Even though there are some limita-
tions regarding the boundary condition in this method, integrated
use of FE model simulation with experimental IRT showed that the
combination can improve the efficiency of IRT use for concrete
bridge deck inspections to detect subsurface delaminations.
Acknowledgments
This work was supported mainly by West Nippon Expressway
Company Limited (NEXCO-West) and also by the Scientific and
_
Technological Research Council of Turkey (TÜBITAK) and National
Science Foundation (NSF CMMI #1463493). The authors would like
to express their sincere gratitude to Dr. Koji Mitani, Mr. Masato
Matsumoto, Mr. Shinji Nagayasu and Mr. Kyle Ruske of NEXCO-
West USA for feedback and support throughout the studies pre-
sented in this manuscript. Without their contributions, this project
would have not been realized. The authors would also like to
express sincere appreciation to Mr. Juan Cruz and our other
research group members for their support with manufacturing test
specimens and setting up equipment. Furthermore, the authors are
also thankful to Andy Derewiany for editing the manuscript.
References
[1] Gucunski N, Kee S, La H, Basily B, Maher A. Delamination and concrete quality
assessment of concrete bridge decks using a fully autonomous RABIT platform.
Struct Monit Maint 2015;2:19–34.
[2] Brown MC, Gomez JP, Hammer ML, Hooks JM. Long-term bridge performance
high priority bridge performance issues. FHWA-HRT-1. McLean, VA; 2014.
[3] FHWA. Memorandum ‘‘Collection of Element Level Data for National Highway
System Bridges”; 2013 <http://www.fhwa.dot.gov/map21/guidance/
guideeldnhsb.cfm>.
[4] Kashif Ur Rehman S, Ibrahim Z, Memon SA, Jameel M. Nondestructive test
methods for concrete bridges: a review. Constr Build Mater 2016;107:58–86.
http://dx.doi.org/10.1016/j.conbuildmat.2015.12.011.
[5] Forde MC. Differences in International Strategy for the NDT of Concrete. In:
Günesß O, Akkaya Y, editors. Nondestruct. Test. Mater. Struct., vol. 6, Dordrecht:
Springer, Netherlands; 2013, p. 1217–27. http://dx.doi.org/10.1007/978-94-
007-0723-8.
[6] Forde MC. International practice using NDE for the inspection of concrete and
masonry arch bridges. Bridge Struct 2010;6:25–34. http://dx.doi.org/10.3233/
BRS-2010-004.
[7] Khan F, Bolhassani M, Kontsos A, Hamid A, Bartoli I. Modeling and
experimental implementation of infrared thermography on concrete
masonry structures. Infrared Phys Technol 2015;69:228–37. http://dx.doi.
org/10.1016/j.infrared.2015.02.001.
[8] McCann DM, Forde MC. Review of NDT methods in the assessment of concrete
and masonry structures. NDT E Int 2001;34:71–84. http://dx.doi.org/10.1016/
S0963-8695(00)00032-3.
[9] Maldague XPV. Introduction to NDT by active infrared thermography. Mater
Eval 2002;60:1060–73.
[10] Waugh RC. Development of Infrared Techniques for Practical Defect
Identification in Bonded Joints. Dev. Infrared Tech. Pract. Defect Identif.
Bond. Joints. 1st ed., Springer International Publishing; 2016, p. 21–37. http://
dx.doi.org/10.1007/978-3-319-22982-9.
[11] Washer G, Fenwick R, Nelson S, Rumbayan R. Guidelines for the thermographic
inspection of concrete bridge components in shaded conditions. Transp Res
Rec J Transp Res Board 2013;2360:13–20. http://dx.doi.org/10.3141/2360-02.
[12] Kee S-H, Oh T, Popovics JS, Arndt RW, Zhu J. Nondestructive bridge deck testing
with air-coupled impact-echo and infrared thermography. J Bridge Eng
2012;17:928–39. http://dx.doi.org/10.1061/(ASCE)BE.1943-5592.0000350.
218 S. Hiasa et al. / Computers and Structures 190 (2017) 205–218
[13] Oh T, Kee S, Arndt RW, Popovics JS, Asce M, Zhu J. Comparison of NDT methods
for assessment of a concrete bridge deck. J Eng Mech 2013;139:305–14. http://
dx.doi.org/10.1061/(ASCE)EM.1943-7889.0000441.
[14] Hiasa S, Birgul R, Catbas FN. Infrared thermography for civil structural
assessment: demonstrations with laboratory and field studies. J Civ Struct Heal
Monit 2016;6:619–36. http://dx.doi.org/10.1007/s13349-016-0180-9.
[15] Rumbayan R, Washer GA. Modeling of environmental effects on thermal
detection of subsurface damage in concrete. Res Nondestruct Eval
2014;25:235–52. http://dx.doi.org/10.1080/09349847.2014.933993.
[16] Cannas B, Carcangiu S, Concu G, Trulli N. Modeling of active infrared
thermography for defect detection in concrete structures. COMSOL Conf
2012:7.
[17] Yuan M, Wu H, Tang Z, Kim H, Song S, Zhang J. Prediction of effect of defect
parameters on the thermal contrast evolution during flash thermography by
finite element method_Yuan. J Korean Soc Nondestruct Test 2014;34:10–7.
[18] Restrepo Girón AD, Loaiza Correa H. New 3D finite difference method for
thermal contrast enhancement in slabs pulsed thermography inspection. J
Nondestruct Eval 2013:62–73. http://dx.doi.org/10.1007/s10921-013-0203-y.
[19] Krishnapillai M, Jones R, Marshall IH, Bannister M, Rajic N. NDTE using pulse
thermography: numerical modeling of composite subsurface defects. Compos
Struct 2006;75:241–9. http://dx.doi.org/10.1016/j.compstruct.2006.04.079.
[20] Hiasa S, Birgul R, Catbas FN. Investigation of Effective Utilization of Infrared
Thermography (IRT) through Advanced Finite Element Modeling. Constr Build
Mater, in press.
[21] Hiasa S, Catbas FN, Matsumoto M, Mitani K. Monitoring concrete bridge decks
using infrared thermography with high speed vehicles. Struct Monit
Maintenance An Int J 2016;3:277–96. http://dx.doi.org/10.12989/
smm.2016.3.3.277.
[22] Washer G, Fenwick R, Bolleni N. Development of Hand-held Thermographic
Inspection Technologies. Report No. OR10-007: 2009.
[23] Washer G, Fenwick R, Bolleni N. Effects of solar loading on infrared imaging of
subsurface features in concrete. J Bridge Eng 2010;15:384–90. http://dx.doi.
org/10.1061/(ASCE)BE.1943-5592.0000117.
[24] Gucunski N, Nazarian S, Yuan D, Kutrubes D. Nondestructive Testing to
Identify Concrete Bridge Deck Deterioration. Washington, D.C., USA:
Transportation Research Board, SHRP 2 Report S2-R06A-RR-1; 2013.
[25] Yehia S, Abudayyeh O, Nabulsi S, Abdelqader I. Detection of common defects in
concrete bridge decks using nondestructive evaluation techniques. J Bridge
Eng 2007;12:215–25. http://dx.doi.org/10.1061/(ASCE)1084-0702(2007)12:2
(215).
[26] Watase A, Birgul R, Hiasa S, Matsumoto M, Mitani K, Catbas FN. Practical
identification of favorable time windows for infrared thermography for
concrete bridge evaluation. Constr Build Mater 2015;101:1016–30. http://dx.
doi.org/10.1016/j.conbuildmat.2015.10.156.
[27] Hiasa S. Investigation of Infrared Thermography for Subsurface Damage
Detection of Concrete Structures. Electronic Theses and Dissertations. Paper
5063. <http://stars.library.ucf.edu/etd/5063>; 2016.
[28] Maierhofer C, Brink A, Ro M, Wiggenhauser H. Quantitative impulse-
thermography as non-destructive testing method in civil engineering –
experimental results and numerical simulations 2005;19:731–7. http://
dx.doi.org/10.1016/j.conbuildmat.2005.06.002.
[29] Cheng C-C, Cheng T, Chiang C. Defect detection of concrete structures using
both infrared thermography and elastic waves. Autom Constr 2008;18:87–92.
http://dx.doi.org/10.1016/j.autcon.2008.05.004.
[30] Abdel-qader I, Yohali S, Abudayyeh O, Yehia S. Segmentation of thermal
images for non-destructive evaluation of bridge decks 2008;41:395–405.
http://dx.doi.org/10.1016/j.ndteint.2007.12.003.
[31] Hiasa S, Catbas FN, Matsumoto M, Mitani K. Considerations and issues in the
utilization of infrared thermography for concrete bridge inspection at normal
driving speeds. J Bridge Eng, in press.
[32] COMSOL. Heat Transfer Module User’s Guide Version 5.1. 2015.
[33] Google. Google Maps 2015 <https://www.google.com/maps/>.
[34] WeatherUnderground. Hourly Weather History & Observations 2015 <http://
www.wunderground.com/personal-weather-station/dashboard?ID=
KFLORLAN126#history/tdata/s20150121/e20150121/mdaily>.
[35] Morris A, Sheets A. Geothermal systems –System Types, Applicability and
Environmental Impacts- 2011 <https://repositories.lib.utexas.edu/bitstream/
handle/2152/13326/9-Morris_Sheets-Geothermal_Systems.pdf?sequence=2>.
[36] Holst GC. Common sense approach to thermal imaging. Winter Park, FL,
USA: JCD Publishing; 2000.
[37] Shih FY. Image processing and pattern recognition: fundamentals and
techniques. Hoboken, New Jersey, USA: Wiley-IEEE Press; 2010.