Construction and Building Materials 101 (2015) 1016–1030

Contents lists available at ScienceDirect

Construction and Building Materials

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

Practical identification of favorable time windows for infrared

thermography for concrete bridge evaluation
Azusa Watase a,c, Recep Birgul a,b, Shuhei Hiasa a,c, Masato Matsumoto c, Koji Mitani c, F. Necati Catbas a,⇑
a
University of Central Florida, College of Engineering and Computer Science, Department of Civil, Environmental and Construction Engineering, 12800 Pegasus Drive, Suite
211, Orlando, FL 32816-2450, USA
b
Mugla Sitki Kocman University, College of Engineering, Department of Civil Engineering, Mugla 48000, Turkey
c
West Nippon Expressway Company Limited (NEXCO-West), 1015 18th St. NW Suite 504, Washington, DC 20036 USA

h i g h l i g h t s

Identification of infrared imaging weather conditions for concrete surfaces inspections.

Practical approach to identify the most beneficial time window to conduct infrared imaging.

Experimental studies using sensors on concrete surfaces and data from weather services.

Multiple regression analyses to generate prediction equations based on sensor data as well as data from weather services.

Approach to eliminate preliminary site work and data collection prior to the actual field tests.

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

Article history: Infrared Thermography (IRT) is one of the nondestructive inspection techniques to detect delaminations in
Received 14 November 2014 concrete bridge decks. These defects are identified by capturing the temperature gradient of concrete sur-
Received in revised form 22 October 2015 faces. In order for this technique to be effective in damage detection, IRT inspections should be conducted
Accepted 24 October 2015
at certain time windows with favorable temperature conditions to get clear temperature gradients on
inspected surfaces. This study is an experimental work examining the effects of ambient environmental
conditions at different times of a day to locate subsurface delaminations and voids at a shallow depth,
Keywords:
which is an additional influencing factor. This study also attempts to figure out a relationship between
Infrared thermography
Delamination
ambient environmental conditions and the temperature values of concrete surfaces to estimate the best
Bridge decks time window with appropriate environmental conditions for IRT inspections. To this end, specially
Nondestructive evaluation designed reusable concrete test plates with different thicknesses were manufactured to collect thermocou-
Thermocouple readings ple sensor readings. Multiple regression analyses were employed to generate prediction models that seek a
Defect detection relationship between environmental conditions and temperature gradients on the test plates attached to a
Concrete test plates target bridge. Regression models also utilized sensor data collected at another location different than the
target bridge location. It was found out that the most important aspect of sensor data collection was to
accomplish a perfect contact of test plates with concrete bridge deck surfaces to get discernible tempera-
ture gradients. When this condition is not met, data analyses yield spurious results leading to futile con-
clusions. On the other hand, it was also observed that prediction models generated by regression
analyses followed the same pattern as that of sensor readings. This makes it possible to have prediction
equations based on sensor readings to determine suitable time window for conducting IRT inspections.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction the capacity, serviceability and efficiency of transportation net-

works and adversely impact local, regional and national economic
Bridges show signs of deterioration, damage and other deleteri- growth. Therefore, highway agencies have become more sensitive
ous processes as they age. The bridges with poor condition impact to maintenance, repair and replacement issues of bridges [1]. Other
institutions, such as the American Society of Civil Engineers (ASCE),
⇑ Corresponding author. also express concerns and close interest in the current conditions
E-mail addresses: [email protected] (R. Birgul), m.matsumoto@w-nexco-usa. of infrastructures. As a specific example, 2013 Report Card for
com (M. Matsumoto), [email protected] (F.N. Catbas).

http://dx.doi.org/10.1016/j.conbuildmat.2015.10.156
0950-0618/Ó 2015 Elsevier Ltd. All rights reserved.
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1017

America’s Infrastructures: Bridges rates the bridges in the US as C+, this research, a practical method will be sought for monitoring
meaning ‘‘mediocre” condition, due to the fact that one in nine and analyzing variations of concrete surface temperature in order
bridges are rated as structurally deficient [2]. to estimate an appropriate time window without the need for
The lack of sufficient funding compels authorities to search for a bridge site visits.
remedy to overcome the problem of structurally deficient bridges.
As an alternative, rehabilitation is a lot cheaper than replacement if 2. Brief review of past work
damage seems unimportant. However, the proposition of insignif-
icant damage condition and maintenance prioritization must be Clark et al applied IRT to the nondestructive testing of concrete
substantiated by acceptable and scientific means. It appears that and masonry bridges [9]. The focus of their investigation was to
the use of nondestructive evaluation (NDE) methods may play an detect delaminations using IRT as opposed to coin tap testing. They
important role to inspect the current condition of existing bridges carried out an IRT survey on a bridge with nine spans and known
at a reasonable cost. Hence, NDE promises a way to improve the areas of delaminations. The duration of survey was 2 days to see
allocation of funding by improving the information these decisions if temperature conditions were more favorable on either day. They
are based on, by better assessing existing bridge conditions [3,4]. confirmed the delaminations that were previously identified by
Various NDE technologies are being utilized all over the world. coin tap test and also located new areas of delaminations that were
They range from traditional hammer sounding tests to advanced not known before. They reported that the IRT survey could not
technologies such as Ground Penetrating Radar [5], Digital Image detect only one delamination since it was close to the edge of
Correlation [6], Impact Echo [7], Ultrasound [8] and Infrared the slab and subject to local sunshine. The difference in the tem-
Thermography [9,10]. Highway bridge conditions have been tradi- perature of the delaminated areas was approximately 0.2–0.3 °C.
tionally monitored by means of different inspections such as ham- The sensitivity of the thermal camera used in their study was
mer sounding and chain dragging tests carried out by qualified 0.08 °C.
engineers and inspectors. Nonetheless, these techniques are A more recent review paper that deals with IRT for condition
time-consuming and can present potential danger to the inspector monitoring [15] surveys not only civil engineering applications
because the inspector needs to be physically very close to the but also IRT applications for condition monitoring in other disci-
inspected bridge element; this necessity may require lane closure plines. However, this study focuses on IRT inspections specifically
which is not favorable to the highway agencies. In addition, a for concrete bridge deck condition monitoring. Thus, literature
survey study [11] showed that bridge condition ratings are dealing with concrete bridge deck monitoring by infrared inspec-
assigned with significant variability since visual inspection is a tion is of importance. One such study was reported by Washer
highly subjective NDE technique. et al [16]. Motivated by objections to the subjective limits for ‘‘sun-
Concrete bridge decks are especially important to maintain in shine,” stated in the ASTM D4788-03 [14], Washer et al. [16] tai-
good condition; the SHRP2 Renewal Research report [1] lists 14 lored an experimental study using IRT imaging in concrete; their
candidate methods to check for deterioration in bridge decks. objective was to determine the effects of solar loading on the abil-
Infrared Thermography (IRT) is among these candidate techniques ity to detect subsurface defects. These defects were embedded by
because it has been in use since 1980s to detect concrete defects design in a large concrete test block; although they placed four
such as delaminations and concrete disintegration in roadways defects at different depths, the one with a depth of 51 mm was
or bridges. IRT monitors electromagnetic wave surface radiations analyzed because it resembled corrosion induced delamination in
related to temperature variations in the infrared wavelength to concrete bridge decks. Since there had not been a study reported
detect subsurface defects. This method is mostly utilized to locate examining environmental conditions over a long time period, they
voids and delaminations in concrete. It was stated in the report decided to take measurements at 10-min intervals for 24 h a day
that IRT along with five other technologies showed a good poten- between November, 2007 and January, 2008, for a total of 89 days.
tial for delamination detection and characterization. However, it Their evaluations were based on the quantitative measurements of
is not free from drawbacks, one of which is the lack of information the thermal contrast that appeared in thermal images obtained
about the depth of the defect. Additionally, deep flaws might easily from the test block. In their analyses, a threshold of 1 °C was pre-
be missed; surface anomalies and boundary conditions also affect ferred to ascertain the thermal contrast occurred at the embedded
the measurements [12]. It has been reported [13] that the accuracy defects, simply because this threshold value is 10–20 times the
of damage detection using IRT is greatly affected by daily temper- sensitivity of the thermal camera which was 0.08 °C. In their study,
ature variations; therefore, it is important to estimate an appropri- they reported that there could be a substantial delay between solar
ate time window for IRT applications before inspection activities loading on a concrete structure and observation of subsurface
take place. defects and that intermittent cloud cover could lessen the thermal
ASTM D4788-03 [14] mentions of both daytime and nighttime contrast in IR images. The fact that this technology was not useful
testing for data collection; it requires at least 0.5 °C temperature every day and environmental factors must be taken into account
difference between the delaminated and the adjacent solid con- when conducting a test was a disadvantage of the thermal inspec-
crete for IRT to identify the flaws. This standard stipulates a mini- tion technique. On the other hand, it was found to be interesting to
mum of 3-hour direct sunshine to be able to create a temperature observe that the maximum contrast for the deepest defect
difference of 0.5 °C and states that weather conditions must appeared a little more than 9 h after sunrise. The authors con-
include sunshine. However, it does not specify a time window to cluded that a late afternoon inspection would be essential to offer
meet these requirements. It also precludes testing when the wind optimized conditions for deeper defects in a concrete structure. It
velocity exceeds 50 km/h (30 mph). In some of the commercial should be noted that such a conclusion would be site specific
applications [13], the way to estimate an appropriate time window and would depend on environmental conditions during the inspec-
to comply with ASTM requirements is to continuously monitor the tion period.
temperature of concrete surfaces for a predefined time period. Yehia et. al. [7] experimentally investigated and compared abil-
Since the temperature varies depending on the area or region, ities of IRT, impact echo and ground penetrating radar techniques
there is a need to visit the inspection site before the actual test to capture common defects in concrete bridge decks. Six deck spec-
and monitor the temperature variation. This requires additional imens were fabricated with varying deck thicknesses and different
visits to a bridge site and also access to the bridge for measuring flaws. The IR camera had a thermal sensitivity of 0.08 °C (0.15 °F).
concrete temperature. Sometimes, this can be a practical issue. In IR measurements were performed at different times on each day
1018 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

between 10:00 a.m. to midnight to be able to observe the effects of an influential parameter since, as the Stephan–Boltzmann law
testing time on the responses. It was concluded that the ability of indicates, the greater the value of emissivity, the higher the radiant
IR to detect deep defects depended on defect size and ambient flux emitted.
environmental conditions. Selecting an appropriate IRT device for a specific monitoring
Ahlborn et al [17] conducted a study, supported by United States task can be accomplished by the help of another Physics law
Department of Transportation Research and Innovative Technology known as Wien’s displacement law.
Administration, to compare the effectiveness of remote sensors for
km ¼ b=T
bridge condition assessment. They reviewed ten different remote
sensing technologies for use in different parts of bridges. The where
thermal cameras used in their research were FLIR i7 and the FLIR km = wavelength of maximum radiation (lm)
ThermaCAM SC640. They have concluded that IRT was one of the b = Wien’s displacement constant (2897 lm/K)
five remote sensing technologies for identifying bridge health indi- T = absolute temperature of the target (K)
cators on deck surfaces; however, they have also reported that
infrared thermography was not always a possible option due to This equation relates the surface temperature of the target to
its limitations and effects of the environmental variables. the maximum wavelength of the emitted energy. As an example,
for the 300 K ambient room temperature value, the peak wave-
3. Problem statement length of IR radiation would be 2897/300  10 lm. This informa-
tion along with expected temperature range is important when
Since damage identification accuracy using infrared images selecting the right IRT equipment for proper measurements.
depends on temperature gradient of objects’ surfaces which is Appropriate infrared sensors are able to detect those invisible
greatly affected by ambient temperature variation, favorable tem- wavelengths and allow us to determine surface temperature of
perature condition to enable defect detection is essential. Yet, the an object [18].
ambient temperature condition cannot be controlled; therefore, IRT is based on the premise that subsurface flaws have an effect
this paper explores a relationship between ambient environmental on the rate of heat flow inducing surface temperature differentials.
conditions and the temperature values of concrete surfaces. For Measurements of these surface temperature differentials lead to
this purpose, specially designed concrete plates with different the detection of the subsurface anomalies. However, there are
thicknesses are used; regression analysis is employed to establish important environmental factors affecting the ability of IRT devices
a relationship between ambient environmental conditions and to detect subsurface defects in concrete; these factors are solar
the temperature values of concrete surfaces. The goal of this study loading, wind speed, ambient temperature changes [21] and rela-
is to estimate the best time window efficiently with appropriate tive humidity [22]. Cited references give clear explanations on
environmental conditions for IRT inspections. how these factors affect the thermal readings on the surface of a
concrete deck. If one is concerned with an accurate temperature
reading of a specific point on an object, then the temperature cal-
4. IRT principals and their relations to present study culation process of an IRT device must take all these factors into
account precisely. On the other hand, if the goal is to capture the
Thermal properties of an object, namely specific heat, mass den- temperature differentials of adjacent points on the surface of an
sity, heat capacity, thermal conductivity and thermal diffusivity, object in the form of thermograms, then it would be sufficient to
determine how heat transfers from and into that object. The three determine an appropriate time window such that environmental
modes of heat transfer are conduction, convection and radiation. factors are at their optimal levels for the object in question to emit
Any heat transfer happens by one or more of these three modes. infrared radiation. As far as flaw detection is concerned, selecting a
IRT is based on measuring the radiative heat flow, so it is related favorable time window to capture temperature differentials on the
to the radiation mode of heat transfer [18]. Any object above surface of an object is vital to be able to depict subsurface anoma-
absolute zero temperature radiate energy in the infrared (IR) zone lies; the actual temperature values for specific points on the sur-
of the electromagnetic spectrum. IR radiation from a target surface face are immaterial. Therefore, this study focuses on determining
can be measured by infrared scanning devices and these measure- the best time window with favorable environmental conditions
ments can be converted to temperature values by the Stephan– to conduct IRT inspections for flaw detections in concrete bridge
Boltzmann law [19]. decks. The thermocouples attached to the surface of the experi-
W ¼ erT 4 mental specimens designed for IRT tests are exposed to the same
ambient environmental conditions so that the only variable to
Where monitor is the occurrence of an appropriate time window to detect
W = radiant flux emitted per unit area (watts/m2) subsurface damages by measuring the temperature differences
r = Stephan–Boltzmann constant (5.67  108 watts/m2/K4) between two thermocouple readings.
e = unitless emissivity value of the target (from available tables
and charts) 4.1. Experimental program
T = absolute temperature of the target (K)
The process IRT detects internal defects depends heavily on
According to this law, the radiant energy, W, the target surface readings of measurements of temperature differentials on the sur-
emits is equal to two factors multiplied by the fourth power of the face of an object. These readings are then translated into colors that
absolute temperature, T, of the target. An IRT device measures W make it easy to keep track of temperature changes on the surface of
and then calculates T easily since the Stephan–Boltzmann constant the object in question. The changing speed of surface temperature
is already a constant and the emissivity value (e) of the target is a of a body varies with the materials composing the body and its
surface characteristic which is also a constant for a material in thermal properties such as specific heat and heat conductivity. In
question over a given temperature range and under specific the case of concrete structures, peeling or exfoliation near the con-
measurement conditions. Emissivity value (e) is the ratio of crete surface contains air with low heat conductivity. Thus, differ-
actual emission from an object to that from a hypothetical ences in the surface temperature between these defective portions
source called a ‘‘black body” at the same temperature [20]; it is and sound sections may occur due to external means such as
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1019

changes in the atmospheric temperature. It must be noted that it is 4.2. IR measurements

impossible to detect defects in an object using IRT if it is in an
environment with a constant temperature since there is no tem- Section 4 of this paper discusses the factors affecting the ability
perature gradient. Temperature changes can be initiated actively of IRT devices to detect subsurface defects in concrete. In addition
by heating and cooling the object to be inspected by IRT. to these factors, sensitivity of an IRT device is also an important
In this research, the other option known as passive type is constituent to get a quality image. IRT devices are manufactured
employed in which IRT captures naturally occurring temperature to yield accurate temperature readings with accuracies, for com-
gradients due to changes in the atmospheric temperature. mercial use, mostly in the order of ±2 °C or ±2% of measurements,
Internally defective parts of concrete structure present different whichever is higher, gathered from objects in question. As an
temperature distributions on its surface than that of sound parts example, for a measurement range between 0 and 50 °C, a temper-
as a result of natural temperature gradients due to changes in ature reading, say 35 °C, must be reported as 35 ± 2 °C; on the
atmospheric temperature as shown in Fig. 1 [13]. These tempera- other hand, for a measurement range between 100 and 500 °C, a
ture distributions are captured by an infrared camera, and the temperature reading value, e.g., 200 °C, from an object must be
infrared images so generated are displayed in an appropriate tem- reported as 200 ± 4 °C (200 ± 0.02  200). This accuracy is associ-
perature range so that disparate sections can easily be identified. ated with the calibration quality of the detectors [20]; the better
Then, quantitative analyses of infrared images are used to gather the calibration, the smaller the measurement error. It must be
information to help decision making process for the corrective noted that this parameter should not be confused with another
actions if deemed necessary. important IR camera parameter known as ‘thermal sensitivity’ or
In IRT applications, the ingenuity and the difficulty thereof, is in ‘temperature resolution’. The more common name in literature
figuring out the most favorable time window to obtain infrared for this parameter is ‘‘Noise Equivalent Temperature Difference”
images. Physical and operational constraints of any inspection pro- (NETD). It is basically a threshold level above which detectors
gram must also be considered to decide as to what time window is can distinguish temperature differences between two spots. Com-
most beneficial to conduct inspection works. This research mercially available IR cameras have sensitivities ranging from
attempts to identify the most beneficial time window to make 0.150 °C to 0.020 °C. The research works cited above, and many
the most out of the detector resolution of an infrared camera to others, use a budget-wise available IR camera to investigate the
be utilized in IRT inspections by means of an efficient approach. versatility of infrared technology by taking measurements on
The two constituents of the experimental program, infrared experimental specimens with predefined deficiencies. The outputs
camera and concrete test plates, are elaborated in the subsequent and the conclusions drawn from the research works are directly
sections of this paper. affected by the available camera’s thermal sensitivity level since

Fig. 1. Diurnal temperature flow in a concrete structure with delamination [13].

1020 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

all the reported types and depths of defects are measured by their target bridge is at a considerable distance from a maintenance
respective camera. office. Moreover, IRT inspections are desirable due to their non-
The IR camera used in this study is FLIR SC5600M; full details of contact nature; however, the task of attaching test plates onto a
this camera can be found in manufacturer’s website (www.flir.- structure to figure out the best time window for IRT inspections
com) [23]. Its 640  512 pixels format InSb (Indium Antimonide, requires access to the structure. Therefore, a different way needs
photonic type detector) focal plane array delivers a programmable to be explored in order to identify an appropriate time window
rate from 1 Hz to 100 Hz. Spectral sensitivity of the camera is spec- for IRT inspections without making additional visits to target
ified as 3–5 lm. Its NETD value is less than 25 mK, 20 mK being bridges. This study is tailored toward addressing this necessity
typical. These features and the others that are not listed here make with a clear objective to offer an efficient solution.
this camera a powerful tool in demanding scientific research. Concrete test plates with three different thicknesses of 1 cm,
2 cm and 3 cm as shown in Fig. 3a were the essential elements of
4.3. Concrete specimens designed for IRT tests the experimental phase of this study. Two sets, a total of six plates,
were manufactured for the research. A silicone material with
Prior to the details of concrete specimens, it will be informative 10 cm width and 0.5 mm thickness was glued, Fig. 3b, on one face
to discuss the need and utilization of these specimens. Fig. 2 of each plate to mimic a delamination defect. The center part of the
depicts some of the commercial practices of infrared imaging on plate with an area of 10 cm2, shown in Fig. 3b, constitutes the
an existing in-service structure. Concrete plates with predefined delamination portion while remaining plate area represents the
delaminations and thicknesses are fixated onto the structure that sound portion of the concrete test plate.
will be inspected via infrared imaging. A temperature measure- The process to attach these plates to a concrete surface, as
ment sensor with two channels is attached to the concrete plate illustrated in Fig. 4 [24], starts by fixing the anchor metals to the
for monitoring the surface temperature of the delaminated area surface with adhesive tapes; then, the concrete plate is placed
and the surface temperature of the sound area at the same time. between the anchor metals, and finally, bolts are used to secure
For the IR camera available in this research, there needs to be at the whole assembly. When the processed face is attached to a con-
least ±0.2 °C temperature differential between the readings coming crete structure, a 10 cm2 delamination area with 0.5 mm thickness
from the delaminated portion and sound portion of the concrete will have been formed. The distance between the delamination and
surface so that the IR images reveal the defects underneath. The the outer surface of concrete test plate will be 1 cm, 2 cm or 3 cm,
0.2 °C value is chosen as a rule of thumb and it is about ten times depending on the thickness of the plate. The final assembly of the
the thermal sensitivity of the available IR camera. Ambient tem- plates on a concrete structure is already shown in Fig. 2. A temper-
perature readings are also collected with the same type of temper- ature measurement sensor attached to the center of outer surface
ature sensors in close proximity of the concrete plates to keep track of the plate records the temperature of the concrete surface below
of the accuracy of the readings. It is obviously necessary to collect which delamination exists; any measurement outside the center
readings from concrete plates for a period of at least twenty-four portion of the plate comprises the surface temperature of sound
hours so that one-day temperature cycle and differentials of the concrete portion.
concrete surface are obtained to guide the decision making regard-
ing the most productive inspection period in a typical day for the 4.4. Locations of data collections
site where the structure is located.
However, the procedure depicted in Fig. 2 requires extra time The method developed in this study is to collect data at both
and additional visits to bridge sites before IRT inspection works Location A (reference site) and Location B (bridge site); then, com-
take place; sometimes this can pose difficulties, especially if a bine and analyze these data sets to identify the most effective time

Readings from
“Delaminaon”
Porons

Temperature
Data Loggers

“Ambient Readings from

Temperature” “Sound”
Readings Porons

Fig. 2. A practice to figure out the most productive inspection period.

 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1021

Delaminaon
Sound

a) Prismatic concrete plates b) Simulated delamination in plates

Fig. 3. 30  30 cm concrete plates with three different thicknesses.

process was repeated for bridge deck soffit and its corresponding
spot, the ceiling of a staircase in a parking structure. These mea-
surements were taken between April 8, 2013 and April 16, 2013.
Data gathering durations were required to keep at a reasonable
level in order not to restrain the bridge traffic service intolerably.
The second source of data was a website, http://www.wunder-
ground.com/ [25], providing weather forecast information. A
weather website is used as this would be a practical method for
routine applications to collect data from a bridge site to be corre-
lated with a reference site for predictive analysis. This is obviously
Fig. 4. Attaching the concrete plates on a concrete structure [24]. a simple approach for the sake of practicality. However, it should
be noted that it is unquestionably preferable to have location-
specific data collected by localized weather stations on other struc-
window for IRT inspection work at Location B without making tures, if available, for weather forecast information in the vicinity
additional visits to Location B. Location A was chosen to be the of the bridge of interest. This would make the subsequent predic-
campus of University of Central Florida (UCF) due to convenience. tive analysis more refined for the structure that is in close proxim-
Florida Department of Transportation (FDOT) District 5 provided a ity to the structures that provide sensor weather data.
bridge at I-75over SR44 as Location B for the experiment. IRT Washer et al [21] reported that a linear relationship exists
inspections on concrete bridges mainly involve bridge decks and between solar loading and the development of thermal contrast;
bridge soffits; thus, it is reasonable to attach the concrete test they also found out that the solar loading over the course of a
plates on deck top and deck soffit surfaces of a bridge. Fig. 5 shows day may be more important than the intensity of sunlight at any
the data acquisition spots on the test bridge and their correspon- particular time. They have also considered the effects of wind
dence on UCF campus. The bridge deck top equivalence on UCF speeds on the thermal contrast and decided to use average values
campus was the parking area with concrete pavement in front of of wind speeds in four quarters of 12:00 midnight to 6:00 a.m.,
Engineering Building whereas bridge deck soffit was associated 6:00 a.m. till 12:00 noon, 12:00 noon till 6:00 p.m., and 6:00 p.m.
with the ceiling of a staircase in one of the parking structures on till 12:00 midnight. A similar approach was followed in the present
campus. study as well. The weather data that were extracted from the web-
The acquired data came from two different sources. The first site were temperature, dew point, humidity, pressure, wind speed,
source was the data obtained by temperature measurement sen- wind direction, visibility, rainfall and clouds. These data were
sors attached to the test plates. Each plate yielded two temperature downloaded four times a day; namely, midnight, 6:00 a.m., noon,
readings, one from the center portion of the plate below which the and 6:00 p.m., for both Location A and Location B. However, the
simulated delamination was positioned, and the other from the data that were collected by the sensor measurements were
sound portion of the plate. The calculated difference between these obtained at 5-min intervals. The inconsistency in terms of data
two readings provided the temperature gradient on the surface of recording intervals between the two sources was rectified by uti-
the concrete plates. Three concrete test plates on bridge deck top lizing the recordings of the sensor measurements that exactly
surface produced six readings, three of which constituted the read- matched with the weather data instances. While conducted out
ings for the delaminated portions, and the remaining three from of necessity for prediction analysis, this course of action unfortu-
the sound portions. Again, temperature gradients were calculated nately led to a substantial reduction of the temporal frequency of
by taking the difference between values of the delaminated and sensor data.
sound portions of the plates. Similar readings from the second
set of three concrete test plates were also recorded from the corre-
sponding spot, parking area on UCF campus, simultaneously 5. Data analysis methodology
between March 20, 2013 and April 02, 2013. Ambient temperature
readings were also recorded at both locations. All the values that Although recorded during data acquisition, some of the weather
were obtained by temperature measurement sensors were website data were excluded in the analysis phase. These were vis-
recorded at 5-minute intervals. The whole sensor measurement ibility and rainfall since they did not differ from day to day, as well
1022 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

Locatio n A Lo cation B

a) Equivalence to deck top b) Bridge deck top

c) Equivalence to deck soffit d) Bridge deck soffit

Fig. 5. Data acquisition spots on the test bridge and their counterparts on UCF campus.

as wind direction and clouds since they were unquantifiable in the predictor tests the null hypothesis, H0, that the coefficient is equal
same sense as that of included data. Therefore, five pieces of infor- to zero, which means the associated predictor has no effect on the
mation, temperature, dew point, humidity, pressure and wind response variable. Therefore, p-values lower than a = 0.05, corre-
speed values formed the weather data. The expected outcome of sponding to 95% confidence interval, suggest that the predictor is
data analysis was to determine the most beneficial time window contributing to the predictive relationship.
to conduct IRT inspections at Location B (any target bridge site) The statistical analyses of collected data followed a certain pro-
by utilizing the sensor and weather data collected at Location A cedure. First, the regression model was fit to the data collected at
(any maintenance office or any DOT facility to be used as a Location A to distinguish the significant predictors; then the
reference site), and weather data collected at Location B. regression model was modified accordingly to include only the sig-
Regression models are some of the most important applications nificant predictors. The same analyses were carried out for the data
of statistics to estimate the mean value of a response variable, y, collected at Location B. Finally, a relationship was sought between
based on knowledge of a set of seemingly related independent the temperature value of a spot on a concrete test plate at Location
variables, x1, x2, . . ., xk. In this study, a first-order multiple regres- B (the target bridge) and the significant predictors of both Location
sion analysis method was used to relate the dependent variable, A and Location B. It should be remembered that the main objective
i.e., temperature value of a spot on a concrete test plate, to the of this study was to eliminate the need of extra visits to Location B
independent variables, namely, sensor data and weather data. and the need of physical access to attach concrete test pieces.
The proposed first-order linear model had the following very well Therefore, the final stage of analysis contained only weather data
known form. at Location B. This restriction does not apply to Location A; con-
crete test plates will always be a part of data collection at Location
EðyÞ ¼ b0 þ b1 x1 þ b2 x2 þ b3 x3 þ    þ bk xk þ e A, which might be a maintenance office or a DOT facility as a refer-
ence site where data collection activities do not create a problem.
Where b’s are regression coefficients to be determined and e is the
error term. The software used for regression analysis is MATLAB.
Regression results indicate the statistical significance of the 5.1. Implementation of the methodology
relationship between the predictors (sensor and weather data)
and the response variable (temperature value of a spot on a con- The above methodology was applied for the delaminated por-
crete test plate). The observed significance level, p-value, for each tion, and was repeated for the sound portion of a concrete test
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1023

plate. Therefore, two sets of data analyses were performed for each Regression output shown in Table 1 can be scrutinized in light
test plate. Three test plates with thicknesses of 1 cm, 2 cm, and of above explanations. Since the observed significance level of
3 cm were attached to the corresponding spots at Location A; this the model, that is p-value in the last row, is much lower than
led to 6 sets of the application of aforementioned procedure for a = 0.05, the model can be deemed adequate. R2 value of 95.9%
statistical data analyses (3 for sound areas, 3 for delaminated indicates that the model fits the data very well. A small value of
areas). Since the same data were collected simultaneously at Loca- S also supports this observation. Residual, e, is the difference
tion B, 12 sets of data analyses would be performed for the bridge between an observed value and its corresponding fitted value.
deck top; another 12 sets of data analyses for bridge soffit led to 24 Any regression analysis has the following four assumptions for e:
times application of data analyses procedure altogether. It is obvi- (1) mean of e is zero; (2) variance of e is constant for all predictors;
ously not practical to present all statistical analyses; thus, the first (3) distribution of e is normal; (4) values of e are independent [26].
analysis, the delaminated portion of 1 cm-plate at Location A, will Fig. 6 shows the residual plots for the model shown in Table 1. The
be presented in detail. The final outcome of all other analyses will points are generally expected to form a straight line if the residuals
be presented in graphs. are normally distributed. Histogram of the residuals is an explora-
Table 1 shows the results of regression analysis of the collected tory tool to show general characteristics of the residuals. The plot
data for the temperature of delaminated portion of 1 cm-thick of residuals versus fitted values shows a random pattern of resid-
plate at Location A, in this case, UCF in Orlando. This corresponds uals on both sides of 0. Any recognizable pattern in this plot should
to the first stage of the procedure explained above; i.e., all the pre- lead to model modification. Residuals versus observation order of
dictors are included in the analysis. The first-order regression data is a plot of all residuals in the order that the data was collected
model was fit to the data, with the following results. and can be used to find non-random error, especially of time-
In the table, the coefficients (Coef) are the numbers by which related effects.
the predictors in the regression equation are multiplied. The value Since the assumptions on the random error e are checked and
of each coefficient determines the contribution of the associated verified as shown in Fig. 6, the next step is to determine the signif-
predictor, given that all other predictors are held constant. Stan- icance level of each predictor so that insignificant predictors can be
dard errors of coefficients (SE Coef) are standard deviations of removed from the model. Of the six predictors in the model, only
the estimates of regression coefficients. They measure how pre- two of them have p-values lower than a = 0.05. Even the tempera-
cisely the data can estimate the coefficient’s unknown value. It is ture value obtained from the weather website is not a significant
always a positive value, and smaller values indicate a more precise predictor since its p-value is 0.228, which is much higher than
estimate. Test statistic to assess the adequacy of a predictor might a = 0.05. At first glance, it may seem odd to get a p-value of almost
be Student’s T statistics; hence, T values are listed for all predictors. zero for ambient air temperature measurements obtained by ther-
There are additional steps to calculate an acceptance range for T mocouple sensors, which makes it a significant predictor, whereas
values. However, the same conclusion can also be easily reached a p-value of 0.228 for temperature readings obtained from the
by the observed significance level, p-value. The standard error of website, which makes it an insignificant predictor, leading to its
regression (S) provides an overall measure of how well the model removal from the model. However, the explanation lies in one of
fits the data. S represents the average distance that the observed the several pitfalls of multiple regression analysis, known as
values fall from the regression line. In other words, it presents multicollinearity. Multicollinearity exists when two or more inde-
how wrong the regression model is on average in the unit of the pendent variables in regression are correlated; in other words, they
response variable. Smaller values are desirable due to the indica- will contribute redundant information. Multicollinearity can also
tion that the observed data points are closer to the fitted line. R2 affect the sign of the parameter estimates; a coefficient value
is a sample statistic; it represents the percentage of the sample may have the opposite sign from what is expected. The negative
variation of the response variable that is attributable to the regres- value of x2, temperature from website, contradicts the expectation
sion model. Its value ranges from 0, a total lack of fit of the model that the temperature readings will positively affect the response.
to the data, to 1 if the model passes through each and every data There are several alternatives available to correct the problem of
point, a perfect fit. In general, the higher the R2 value, the better multicollinearity [26], one of which is to include only one of the
the model fits the data. Analysis of Variance, ANOVA, yields the correlated predictors in the final model. However, if the model will
p-value shown in the last row of Table 1. It indicates the signifi- be used for estimation and prediction, as opposed to establishing
cance of the regression model as a whole; it, too, needs to be lower cause and effect, then all the variables including the correlated
than a = 0.05, corresponding to 95% confidence interval, to show ones can be kept in the model [26]. In this study, both alternatives
overall adequacy of multiple regression model. will be tried to see any improvements on the overall adequacy of
the final model before making a decision.
Table 1 First, all predictors with p-values higher than 0.05 will be
Regression results of 1 cm-thick plate at Location A with all available predictors. dropped from the final model; the results for this alternative are
The regression equation is
given in Table 2. Then, temperature value obtained from the website
y = 364 + 1.41x1  0.296x2 + 0.53x3 + 0.048x4 + 12.1x5 + 0.0263x6 will be included in the model even though it has a high p-value. The
Predictors Coef SE Coef T p
results for this particular alternative are given in Table 3.
From the statistical point of view, removing the temperature
x1 Ambient air temperature at UCF 1.40805 0.08347 16.87 0.000
data from the website may seem to be a better alternative since
x2 Temperature in Orlando from 0.2957 0.2420 1.22 0.228
website all p-values are zero as shown in Table 2. However, including the
x3 Humidity in Orlando from 0.529 7.777 0.07 0.946 temperature data from the website does not damage the overall
website adequacy of the alternative final model as shown in Table 3. S,
x4 Dew point in Orlando from 0.0481 0.2526 0.19 0.850
R2, and p-values for both cases are almost the same. Even the
website
x5 Pressure in Orlando from website 12.068 2.349 5.14 0.000
respective p-value of temperature data from website is now lower
x6 Wind speed in Orlando from 0.02631 0.07321 0.36 0.721 than 0.05. Since the data for Location B in its final form will include
website only the data obtained from website, removing the temperature
S = 2.05024
data from the independent variables set will have deleterious
R2 = 95.9% effects on the results. Therefore, it was decided to keep the temper-
p = 0.000 ature data from the website in the regression model. Fig. 7 shows
1024 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

0.99 5
0.98
0.95
0.90

Probability
0.75

Residual
0.50 0
0.25
0.10
0.05
0.02
0.01
-5
-5 0 5 10 20 30 40
Residual Fitted Value
12
5
10

8
Frequency

Residual
6 0

0 -5
-10 -5 0 5 10 10 20 30 40 50
Residual Observation Order

Fig. 6. Residual analysis to verify the assumptions for e.

Table 2
Regression results of 1 cm-thick plate at Location A with only significant predictors. Results of regression analysis of the data obtained at the bridge
The regression equation is site, Location B, resemble those of the data at UCF, Location A. Thus,
y = 382.8 + 1.23x1 + 12.65x5 the observations and inferences are the same for both data sets;
Predictor Coef SE Coef T p temperature data obtained by sensor measurements and pressure
x1 Ambient air temperature at 1.23322 0.039053 31.5781 0.000
data from the website are the significant predictors. Once again,
UCF even though it had previously a high p-value as shown in Table 4,
x5 Pressure in Orlando from 12.6536 1.65800 7.63186 0.000 temperature data from the website is still included in the modified
website version of the regression model as shown in Table 5.
S = 2.11768
R2 = 95.2% 5.2. Regression model for bridge temperature predictions
p = 0.000
A regression model can be constructed by utilizing only the sig-
Table 3 nificant predictors out of the available data of both locations. For
Regression results of 1 cm-thick plate at Location A with significant predictors the sake of brevity, the rest of the analyses will be presented in
including temperature data from website.
graphs. Since the main objective requires the inclusion of only
The regression equation is website data for a bridge site, pressure and temperature data
y = 316 + 1.39x1  0.259x2 + 10.5x5 obtained from the website are used as the significant predictors
Predictor Coef SE Coef T p of bridge site data. The regression model include the following
x1 Ambient air temperature at UCF 1.39015 0.07531 18.46 0.000 UCF data: sensor data gathered from the delaminated portion of
x2 Temperature in Orlando from 0.2588 0.1079 2.40 0.020 concrete test plate, sensor measurements of ambient temperature,
website and pressure and temperature data from the website. The depen-
x5 Pressure in Orlando from website 10.475 1.826 5.74 0.000 dent variable, y, is the temperature value measured by thermocou-
S = 2.02364 ples at the delaminated portion of the concrete test plate attached
R2 = 95.7% to the bridge deck top. The first-order model was fit to the data,
p = 0.000 with the following result.
ydelaminated ¼ 335  0:284x1 þ 1:11x2 þ 0:246x3 þ 10:8x4
the residual plots for the model shown in Table 3 to verify the
assumptions for e. þ 0:259x5 þ 0:24x6
The same type of data was collected at Location B, the target where
bridge near Wildwood; therefore, the same multiple regression x1: sensor measurement of delaminated portion of 1 cm plate at
analyses were repeated for the data obtained at Location B. The UCF,
results of the analyses for the data that includes all the variables x2: sensor measurement of ambient air temperature at UCF,
are given in Table 4, and for the data that includes only the signif- x3: temperature in Orlando from website,
icant predictors in Table 5. Although the respective figures are not x4: pressure in Orlando from website,
included here, the assumptions on the random error e are checked x5: temperature in Wildwood (bridge location) from website,
and verified before proceeding to the next step. x6: pressure in Wildwood (bridge location) from website.
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1025

0.99 5
0.98
0.95
0.90

Probability
0.75

Residual
0.50 0
0.25
0.10
0.05
0.02
0.01
-5
-4 -2 0 2 4 6 10 20 30 40
Residual Fitted Value

15
5

10
Frequency

Residual
0
5

0 -5
-10 -5 0 5 10 10 20 30 40 50
Residual Observation Order

Fig. 7. Residual analysis for the model in Table 3.

Table 4 R2 = 93.2%, S = 2.45 °C and overall p-value = 0.000 are the result-
Regression results of 1 cm-thick plate at Location B with all available predictors. ing statistics for this model.
The regression equation is The model obtained by this procedure estimates the tempera-
y = 96.5 + 1.16x1  0.173x2  2.01x3 + 0.041x4 + 3.30x5 + 0.0361x6 ture value of the delaminated portion of the concrete test plate
Predictor Coef SE Coef T p attached to the target bridge based on the data collected at the
x1 Ambient air temperature at 1.16094 0.04861 23.89 0.000
bridge site and UCF. In order to get temperature differentials, the
Bridge whole procedure must be repeated to arrive at another model that
x2 Temperature in Wildwood from 0.1732 0.1271 1.36 0.179 estimates the temperature value of the sound portion of the con-
website crete test plate. Then, the difference will produce temperature dif-
x3 Humidity in Wildwood from 2.015 3.472 0.58 0.565
ferential of the surface of 1 cm-thick concrete plate.
website
x4 Dew point in Wildwood from 0.0406 0.1164 0.35 0.729 The model that estimates the temperature value of the sound
website portion of the concrete test plate is given below without the tabu-
x5 Pressure in Wildwood from 3.303 1.233 2.68 0.010 lated results of regression analyses; all the details of these analyses
website and the ones regarding 2 cm- and 3 cm-thick plates for bridge deck
x6 Wind speed in Wildwood from 0.03608 0.03683 0.98 0.332
website
top and deck soffit can be found in a research report [27].

S = 1.03123
R2 = 98.8%
ysound ¼ 249  0:120x1 þ 0:911x2 þ 0:148x3 þ 6:30x4 þ 0:286x5
p = 0.000 þ 1:90x6

R2 = 91.6%, S = 2.63 °C and overall p-value = 0.000 are the result-

ing statistics for this model.
It will be more meaningful to graph both equations to observe
Table 5
the temperature differentials of the surface of 1 cm-thick concrete
Regression results of 1 cm-thick plate at Location B with significant predictors
including temperature data from website. plate attached to the bridge deck top. Fig. 8 depicts the above pre-
diction models whereas Fig. 9 shows the actual temperature vari-
The regression equation is
ation history at the matching hours of the days.
y = 107 + 1.19x1  0.138x2 + 3.59x5
Fig. 8 shows that regression models predict the temperature
Predictor Coef SE Coef T p
variations of the test plate in a fairly good accuracy as verified by
x1 Ambient air temperature at 1.19147 0.04325 27.55 0.000 Fig. 9, the actual temperature variations. This form of presentation,
Bridge
although being informative in terms of emphasizing the similarity
x2 Temperature in Wildwood from 0.13797 0.05752 2.40 0.020
website
between predicted and actual daily temperature variations, does
x5 Pressure in Wildwood from 3.5928 0.9403 3.82 0.000 not provide direct answers toward the main goal of this study. Since
website the objective is to identify the time window that is best for IRT mea-
S = 1.03617
surements, the time intervals where the temperature differentials
R2 = 98.7% between the delaminated and the sound portions are greater or less
p = 0.000 than ±0.2 °C need to be determined. Therefore, the same results for
1026 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

Fig. 8. Estimated temperature variation of 1 cm-thick concrete plate.

the predicted temperature values are presented in a different for- According to the results of prediction models, 6 a.m. is the best
mat as shown in Fig. 10. In this figure, each data point represents time to conduct IRT inspection for capturing delaminations that are
the temperature difference between the delaminated and sound at a depth of 1 cm from the surface on a bridge deck top. On the
portions at the corresponding time in one day; so, each designated other hand, sensor measurements show that 79% (as opposed to
hour contains 13 data points which corresponds to 13 days, the time 100%) of data at 6 a.m. are outside the target temperature value,
duration for the experiment. The target temperature value, ±0.2 °C, ±0.2 °C; still, it is a very good percentage to follow and a good time
is specified as both positive and negative boundaries on the graph window to conduct IRT inspections on the target bridge. At the
with two straight lines, respectively. Any temperature differential other three times of the days, prediction models and sensor mea-
value that is in between these two boundaries is considered unfa- surements are in fairly good agreement and show that any time
vorable for detection purposes whereas any values outside these of the days is suitable to perform IRT inspections. Before delving
boundaries are considered favorable. In order to make an easier into further discussion, the same data analyses are carried out for
comparison between the predicted values and the actual measure- 2 cm and 3 cm thick concrete test plates and presented in Tables
ments, the same presentation format is also used for the actual mea- 7 and 8, respectively.
surement values obtained by sensors and shown in Fig. 11. The The results of prediction models and sensor measurements are
percentages of the values that are outside the boundaries are explic- in almost perfect agreement at noon and midnight; for the other
itly specified on both Fig. 10 and Fig. 11 to quantitatively compare two instances, prediction models yield much higher percentage
the results and identify the time window for best IRT measurement values than those of sensor measurements.
periods. Although the temperature differentials based on sensor It should be mentioned here that the data readings of 3 cm thick
measurements are graphed for every hour since the data was col- test plate after 6th day into the measurements are discarded due to
lected at 5-min intervals, the matching hours are emphasized by sensor malfunctioning and erroneous data readings. Therefore, the
boxed lines in Fig. 11 to point out the comparison times between results for 3 cm thick concrete test plate are based on a total of
the predicted and the actual temperature differentials. 6 day measurements as opposed to 13 day measurements with
Fig. 11 shows the analysis results of sensor readings attached to 1 cm and 2 cm thick plates.
1 cm-thick test plate to record the actual temperature differentials. Both prediction models and sensor measurements suggest that
It shows that negative differential points that are expected to occur any time of the days are suitable to conduct IRT inspections for
during the nighttime periods are more than the positive differen- capturing delaminations that are at a depth of 3 cm from the sur-
tial points that are expected to occur during the daytime periods. face on a bridge deck top.
One would expect to observe a symmetric distribution over the The results of data analysis presented so far involved sensor
course of a day [21]. A similar distribution is also seen in Fig. 10 measurements and prediction models obtained from the top sur-
because estimated temperature differential analysis involves sen- face of the bridge deck. Since the same sensor measurements were
sor readings as input values in regression models; therefore, the also carried out at the bottom surface of the bridge deck, the same
same situation seen in Fig. 11 is also observed in Fig. 10 as well. analyses were performed once more for the data obtained from the
An explanation for this kind of distribution to happen is that the bridge soffit and presented in Table 9, Tables 10 and 11 for concrete
physical contact between the test plates and the concrete surface test plates with 1 cm, 2 cm and 3 cm thicknesses, respectively.
is not satisfactory and air flows freely at some parts of the interface
between the test plate and the concrete surface. 6. Discussion on the predictive analysis results
Table 6 shows the same information presented in Figs. 10 and
11; i.e., the percentages obtained by using the prediction models The percentages of any temperature differential values that are
and sensor measurements. Each value in the table presents above or below ±0.2 °C are shown in Table 6, Tables 7 and 8 for the
percentage of the temperature differentials that are outside the test plates attached to the bridge deck top. As expected, the per-
boundaries. centages obtained by the prediction models for the bridge deck
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1027

Fig. 9. Actual temperature variation of 1 cm-thick concrete plate.

this case. However, the conditions worsen for delaminations that

Percentages are at 2 cm depth; Table 7 shows that only the noon time is suit-
0.5 able for IRT inspection for the delaminations at this depth. This is
Temperature Differential Values in C

100 86 85 85 obviously contrary to common sense; it is expected that if inspec-

0.2 tions can be conducted for deeper delaminations, then it should be
0 no trouble to detect delaminations at shallower depths as well. The
same inconsistency can be observed for the sensor measurements
-0.2
collected at the bridge deck soffit. Table 11 suggests that midnight
is the more suitable time to detect delaminations that are at 3 cm
-0.5
depth from the concrete surface; this is to be expected, however,
Tables 9 and 10 show that almost no time window is any good
for inspections of defects at shallower depths. This can only hap-
-1
pen when the sensors measuring the temperature values at the
surface of the concrete test plates are registering very close values
to each other at both the sound part and the delaminated part so
-1.5 that no discernible temperature differentials occur for use in IRT
inspections. The only logical explanation lies in the speculation
that certain imperfections must have occurred during the attach-
-2 ment of the test plates to the concrete surface as explained in
6 12 18 24
Diurnal Hours Fig. 4. In order not to have distinct temperature differentials, the
sensors either measure both parts as sound or as delaminated. If
Fig. 10. Estimated temperature differentials of 1 cm-thick concrete plate. the test plates were perfectly attached to the concrete surface with
good sealing, the sensor readings for the delaminated portion
would have differed from the sound portion. Therefore, it is reason-
top do not exactly match the percentages obtained by the sensor able to assume that the contact between the test plates and the
measurements collected at the bridge deck top although they are concrete surface is not satisfactory and air flows freely at some
in good agreement for the most part except the instances shown parts of the interface between the test plate and the concrete sur-
in Table 7. Investigating the reasons for this discrepancy starts with face. When this happens, both sensors on the test plate register
scrutinizing the sensor data itself, the last columns of all the tables, very close values since now the sensor trying to measure the tem-
since prediction models also use sensor measurement data col- perature of the supposedly sound part is actually measuring
lected at UCF campus in addition to the website temperature and another but unintentional delaminated part on the plate.
pressure data. Sensor measurements shown in Table 6 suggest that The main objective of this study was to develop a methodology
almost any time of the day is suitable to detect delaminations that to overcome the need of extra visits to bridge sites or the need of
are at 1 cm depth from the bridge deck surface. The results of attaching test plates when the working conditions on a bridge
prediction models in Table 6 also suggest the same conclusion. are very difficult to determine a suitable time window for IRT
According to the sensor measurements, the conditions are still inspections on the bridge. If one looks on the bright side of all this
favorable for inspections to detect delaminations that are at 3 cm study, the results of the prediction models are not too far from the
depth from the surface as shown in Table 8. Again, the prediction results of sensor measurements except one instance in Table 11,
models in Table 8 also point to the same conclusion; sensor mea- the last row. This suggests that when the sensor data at a reference
surements and prediction models are in very good agreement for site is gathered consistently well in quality, then the prediction
1028 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

1.5
Percentages

1 71 86 86 79 93 86 87 93 100 86 86 86
Temperature Differential Values in C
79 71 79 86 93 93 87 80 100 79 86 93
0.5

0.2
0
-0.2

-0.5

-1

-1.5

-2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Diurnal Hours

Fig. 11. Actual temperature differentials of 1 cm-thick concrete plate.

Table 6 Table 9
Percentages of temperature differentials outside the boundaries for 1 cm plate Percentages of temperature differentials outside the boundaries for 1 cm plate
attached to deck top. attached to deck soffit.

Time Prediction models (%) Sensor measurements (%) Time Prediction models (%) Sensor measurements (%)
6:00 100 79 6:00 75 67
12:00 86 93 12:00 0 0
18:00 85 100 18:00 13 36
24:00 85 93 24:00 38 42

Table 7 Table 10
Percentages of temperature differentials outside the boundaries for 2 cm plate Percentages of temperature differentials outside the boundaries for 2 cm plate
attached to deck top. attached to deck soffit.
Time Prediction models (%) Sensor measurements (%) Time Prediction models (%) Sensor measurements (%)
6:00 77 43 6:00 38 42
12:00 93 93 12:00 0 27
18:00 62 43 18:00 0 45
24:00 69 64 24:00 13 25

Table 11
Table 8 Percentages of temperature differentials outside the boundaries for 3 cm plate
Percentages of temperature differentials outside the boundaries for 3 cm plate attached to deck soffit.
attached to deck top.
Time Prediction models (%) Sensor measurements (%)
Time Prediction models (%) Sensor measurements (%)
6:00 75 58
6:00 100 100 12:00 43 55
12:00 83 93 18:00 75 73
18:00 100 93 24:00 25 83
24:00 100 100
reusable small test plates to simulate delaminations is to make
sure a perfect and sealed contact between the plate and the con-
models will show the most suitable time windows to conduct IRT crete surface. A procedure should be developed first to check this
inspections. condition before moving onto data collection using the test plates.
The studies reviewed [7,10,16,17] in literature did not report Without a doubt, the current practice of identifying the best
similar problems since they used large test specimens containing time window using sensor measurements on a target bridge as
predefined and permanent defects as opposed to the portable depicted in Fig. 2 is still the most viable choice. However, if local
and reusable test plates utilized in this study. Needless to say, conditions such as a bridge crossing a river prove to be difficult
the results are not comparable due to the experimental errors to attach test plates on the bridge, then estimating the time win-
encountered during data collection of this study. Nevertheless, it dow by prediction models might be an alternative. Utilizing pre-
has become clear that the single most important aspect of utilizing diction models to estimate the best time window before an
 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030 1029

inspection work at a bridge site seems to be a promising way bond between the test plates and the concrete surfaces of the
because, as can be observed Tables 6–11, the percentage values bridge. Since the results obtained utilizing test plates will be used
obtained by prediction models follow somewhat closely the per- to validate the effectiveness of IRT, it is clear, then, the next step is
centage values given by sensor measurements. Presumably, if reli- to search for alternative means for a secure and fast way to attach
able sensor readings are collected, then more accurate prediction test plates to concrete bridge surfaces. Another approach is to
models can be generated through regression analysis by using develop advanced multiphysics finite element models that also
the sensor measurements as well as website temperature and incorporate location information of the structures, with the pri-
pressure data. However, it must be emphasized that these predic- mary source of heat as solar radiation. Solar radiation can be sim-
tion models would be location specific models; the regression coef- ulated over a particular time window by using the longitude,
ficients would have to be recalculated for a bridge at a different latitude, and time zone, and time of the year in addition to material
location. properties such as thermal conductivity and solar flux at the sur-
face. Such studies, while require advanced engineering, may be
developed for characterizing structures that are in close proximity.
7. Conclusions
To better understand infrared phenomenon as applied to bridge
decks, larger test specimens may be developed and tested to min-
The main objective of this study was to explore another
imize the heat leakage at the ends of the test plates. These tests can
approach to be an adjunct to the current practice of identifying
be conducted using different cameras with different thermal sensi-
the best time window for IRT inspections. The same setup as that
tivity, infrared resolution and detector types as these have pro-
of current time window identification was utilized to be able to
found effects on the use of IRT technology. Finally, it should be
compare the results. The setup was comprised of portable concrete
noted that complementary technologies to IRT applications may
test plates with different thicknesses to collect sensor data out of
need to be employed. For example, research studies may incorpo-
certain spots on the surfaces of the plates. The present study addi-
rate other NDE methods such as Ground Penetrating Radar (GPR),
tionally involved multiple regression analysis to generate temper-
ultrasound and other technologies for detecting deeper flaws in
ature prediction equations based on sensor measurements as well
concrete objects. While IRT is extremely fast and efficient, integrat-
as temperature and pressure data obtained from a website. Evalu-
ing IRT with such other technologies would improve final decision
ation of the results obtained by the three test plates attached to the
on evaluation of structural condition of a structure.
bridge deck top reveals that noon is the favorable time to conduct
IRT on the deck top for this particular bridge because percentages
of temperature differentials are all high values for this time win- Acknowledgements
dow and consistent with each other as shown in Table 6–8. As
for the deck soffit, midnight is the favorable time to carry out IRT The authors would like to thank to the members of Florida
as suggested by Table 11. Department of Transportation for their cooperation and NEXCO-
Data analyses also revealed that an absolutely perfect contact West USA, Inc. for their financial support and technical help with
between the test plates and the concrete surfaces must be achieved the project in Florida. The authors also wish to acknowledge the
to simulate delaminations by portable test plates. If this condition financial support of Tübitak (The Scientific and Technological
is not realized, analyses on the collected sensor data result in spu- Research Council of Turkey) for covering the expenses of 2nd
rious outcomes with incomplete conclusions. On the other hand, it author while he was a visiting scholar at UCF. Several individuals
was also observed that regression models to predict the tempera- from the research group at UCF must be acknowledged as well
ture values on the surface of test plates follow the same pattern for their help in assisting the experimental works; their names in
as that of sensor readings. Therefore, depending on the reliable no particular order are: Evan Prado, Tung Khuc, Masoud Malekza-
sensor readings, it is expected to have regression equations to pre- deh, Travis Henriques, Martisa Washington, and Keke Peng. Juan
dict the temperature values on the test plates based on sensor Cruz was extremely instrumental with manufacturing the experi-
readings to reveal suitable time windows for conducting IRT mental specimens.
inspections. It should be noted that these equations are developed
for a certain location; the coefficients of the equations would need References
to be recalculated for another location.
[1] N. Gucunski, A. Imani, F. Romero, S. Nazarian, D. Yuan, H. Wiggenhauser, P.
Shokouhi, A. Taffe, D. Kutrubes, Nondestructive testing to identify concrete
Recommendation for research directions bridge deck deterioration. SHRP 2 Report, S2-R06A-RR-1, Transportation
Research Board, Washington, D.C., 2013.
[2] ASCE 2014, 2013 Report card for America’s infrastructure: Bridges. <http://
For quick and reliable NDE inspection of bridges, an objective www.infrastructurereportcard.org/factsheet/bridges> [retrieved on July, 15,
and consistent inspection and assessment methodology is doubt- 2014].
[3] D.M. McCann, M.C. Forde, Review of NDT methods in the assessment of
lessly desired. IRT is a promising NDE method to offer such an concrete and masonry structures, NDT&E Int. 134 (2001) 71–84.
inspection methodology to support better evaluation of structural [4] M.C. Forde, International practice using NDE for the inspection of concrete and
condition tasks for the engineers and field personnel engaged in masonry arch bridges, Bridge Struct 6 (1–2) (2010) 25–34, http://dx.doi.org/
10.3233/BRS-2010-004.
monitoring and inspection works. The fact remains that results [5] K. Maser, Integration of ground penetrating radar and infrared thermography
acquired by IRT measurement procedures have to be substantiated for bridge deck condition testing, Mater. Eval. 66 (11) (2008) 1130–1136.
by results that are obtained by existing procedures, preferably [6] T. Hutt, P. Cawley, Feasibility of digital image correlation for detection of cracks
at fastener holes, NDT&E Int 42 (2) (2009) 141–149.
standardized procedures. In determining the favorable time win- [7] S. Yehia, O. Abudayyeh, S. Nabulsi, I. Abdelqader, Detection of common defects
dow to carry out IRT measurements, data gathered by test plate in concrete bridge decks using nondestructive evaluation techniques, J. Bridge
attachment to bridge surfaces yield accurate and informative Eng. 12 (2007) 215–225.
[8] I.O. Yaman, N. Hearn, H.M. Aktan, Active and non-active porosity in concrete.
results. On the other hand, test plate attachment and data gather-
Part II: Evaluation of existing models, Mater. Struct. 35 (2002) 110–116.
ing process pose some difficulties. Obviously, utilizing IRT mea- [9] M.R. Clark, D.M. McCann, M.C. Forde, Application of infrared thermography to
surement results would be a definite improvement to make a the non-destructive testing of concrete and masonry bridges, NDT&E Int. 6
decision regarding the favorable time window for IRT monitoring. (2003) 265–275.
[10] G. Washer, R. Fenwick, N. Bolleni, J. Harper, Effects of environmental variables
The single most important issue featured in this experimental on infrared imaging of subsurface features of concrete bridges, Transport Res
study is the detrimental effect caused by the absence of a perfect Rec: J Transport Res Board 2108 (2009) 107–114.
1030 A. Watase et al. / Construction and Building Materials 101 (2015) 1016–1030

[11] B.A. Graybeal, B.M. Phares, D.D. Rolander, M. Moore, G. Washer, Visual [19] M.A. Starnes, N.J. Carino, E.A. Kausel, Preliminary thermography studies for
inspection of highway bridges, J. Nondestruct. Eval. 21 (3) (2002) 67–83. quality control of concrete structures strengthened with fiber-reinforced
[12] C. Meola, G.M. Carlomagno, L. Giorleo, Geometrical limitations to detection of polymer composites, J. Mater. Civ. Eng. 15 (3) (2003) 266–273.
defects in composites by means of infrared thermography, J. Nondestruct. Eval. [20] W. Minkina, S. Dudzik, Infrared Thermography: Errors and Uncertainties, John
23 (4) (2004) 125–132. Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19
[13] M. Matsumoto, K. Mitani, F.N. Çatbasß, Bridge assessment methods using image 8SQ, United Kingdom, 2009.
processing and infrared thermography technology: on-site pilot application in [21] G. Washer, R. Fenwick, N. Bolleni, Development of hand-held thermographic
Florida, in: Proceedings of Transportation Research Board; 92nd Annual inspection technologies. Organizational Results Research Report, September
Meeting, Washington, D.C., 2013. 2009, OR10-007, Submitted to MoDOT.
[14] ASTM D4788–03, Standard Test Method for Detecting Delaminations in Bridge [22] J. Zhang, A. Gupta, J. Baker, Effect of relative humidity on the prediction of
Decks Using Infrared Thermography, American Society of Testing Materials, natural convection heat transfer coefficients, Heat Transfer Eng. 28 (4) (2007)
2007. 335–342.
[15] S. Bagavathiappan, B.B. Lahiri, T. Saravanan, J. Philip, T. Jayakumar, Infrared [23] FLIR website: <www.flir.com>. (retrieved on August, 15, 2014).
thermography for condition monitoring – a review, Infrared Phys. Technol. 60 [24] Nexco USA: <http://www.w-nexco-usa.com/services.html> (retrieved on July,
(2013) 35–55. 15, 2014).
[16] G. Washer, R. Fenwick, N. Bolleni, Effects of solar loading on infrared imaging [25] Weather website: <http://www.wunderground.com/> (accessed when needed
of subsurface, J. Bridge Eng. 15 (4) (2010) 384–390. in March and April, 2013).
[17] T.M. Ahlborn, R.A. Shuchman, L.L. Sutter, D.K. Harris, C.N. Brooks, J.W. Burns, [26] W. Mendenhall, T. Sincich, Statistics for Engineering and the Sciences, fifth ed.,
Report: Bridge condition assessment using remote sensors. United States Prentice-Hall, 2007.
Department of Transportation Research and Innovative Technology [27] Research Project Report submitted to Nexco-USA: Implementation of high
Administration. 2013; Project No: DT0S59-10-H-00001. resolution digital image and infrared thermography technology for inspection
[18] H. Kaplan, Practical Applications of Infrared Thermal Sensing and Imaging and monitoring of civil infrastructure systems. June 30, 2013.
Equipment, third ed., 2007.