Submitted September 2004

Active Infrared Thermography for NDT of Concrete Structures Strengthened with Fiber Reinforced Polymer
by Monica A. Starnes* and Nicholas J. Carino

ABSTRACT
Numerical and experimental methods are used to investigate the effectiveness of infrared thermography to estimate the width of subsurface discontinuities in fiber reinforced polymer plates bonded to concrete. The study focuses on establishing the potential for quantitative infrared thermography capable not only of detecting but also characterizing subsurface discontinuities. Finite element analyses are used to examine surface temperature response due to simulated discontinuities at the interfaces of fiber reinforced polymer plates applied to a concrete substrate. In particular, the effects of discontinuity width and depth are studied. It is concluded that the minimum detectable discontinuity width depends on discontinuity depth and the thermal resolution of the infrared detection system. A procedure is proposed for estimating the discontinuity width based on the point of inflection in the surface temperature profile measured along a given line. Controlled discontinuity experiments are performed to verify the procedure for estimating the width of subsurface discontinuities. Good agreement is found between the estimated and actual discontinuity dimensions. Data smoothing is shown to be effective in removing noise from measured temperature profiles. Keywords: bonding, concrete structures, fiber reinforced polymer composites, discontinuities, numerical simulation, nondestructive testing, infrared thermography.

INTRODUCTION
Fiber reinforced polymer composites are being used to rehabilitate and strengthen existing concrete structures. For civil engineering structures, advances in the use of fiber reinforced polymer composites are being made by the bridge engineering community in an effort to strengthen existing bridges in need of rehabilitation. Fiber reinforced polymer composite layers or plates are bonded to existing reinforced concrete structures using adhesives such as epoxy resins. The major factor that contributes to the optimum performance of the composite system is the quality of bond between the fiber reinforced polymer and the concrete substrate. Due to their nature and installation method, fiber reinforced polymer composites bonded to concrete are prone to interlaminar discontinuities and debonding. While nondestructive testing (NDT) techniques for fiber reinforced polymer composites are widely researched and used in the aerospace industry, the quality of the manufacturing process, size of the structures, environmental conditions and available budget for testing aerospace structures are quite different from those of civil structures. In order to assess the quality of fiber reinforced polymer composites in civil engineering applications, rapid, low cost NDT techniques need to be developed or adapted. It has been shown that infrared thermography can be used successfully for discontinuity detection in fiber reinforced polymer
* Building and Fire Research Laboratory, National Institute of Standards and Technology, 100 Bureau Drive Stop 8611, Gaithersburg, MD 20899; (703) 695-1904; <viton@alum.mit.edu>. Building and Fire Research Laboratory, National Institute of Standards and Technology, 100 Bureau Drive Stop 8611, Gaithersburg, MD 20899; <nicholas.carino@nist.gov>.
56 Materials Evaluation/January 2005

plates and laminates bonded to concrete. Trial case studies performed by the Federal Highway Administration and the New York Department of Transportation (Alampalli et al., 2001), among others, confirmed that infrared thermography is a promising NDT technique for bridge assessment, considering testing speed and ability to detect discontinuities. Infrared thermography is nonobtrusive and allows localized and global testing of structures. Technological advancements in uncooled infrared detectors and data analysis software are making infrared thermography a more accurate and economical testing method than it was in the past. Tests using infrared thermography have focused mostly on qualitative assessment of the presence of discontinuities. The qualitative nature of past tests is due to the complex relationships among the variables affecting the thermal response of the bonded laminates or plates. There is a need to develop the scientific bases for using infrared thermography for quantitative assessment of detected discontinuities, that is, for estimating discontinuity size and depth. To enable widespread use of infrared thermography for quantitative assessment of fiber reinforced polymer materials applied to concrete and masonry structures, a standard test technique is needed. To develop such a standard, however, it is necessary to have an understanding of the factors affecting the thermal response due to discontinuities within fiber reinforced polymer composites bonded to concrete. The National Institute of Standards and Technology, in cooperation with the Massachusetts Institute of Technology, initiated research to gain this understanding. The initial research involved numerical simulations using the finite element method and experimental studies using controlled discontinuity specimens (Starnes et al., 2002). This paper presents key results dealing with estimation of the width of subsurface air discontinuities through infrared thermography testing. First, the principles involved in infrared thermography are summarized. Then, finite element modeling is described, along with the effects of the width of the discontinuity on the thermal response of the test object. A procedure for estimating discontinuity width is proposed and the procedure is used to estimate the width of an air void in a controlled discontinuity specimen. The width of the discontinuity was confirmed by destructive testing.

THEORETICAL PRINCIPLES
Infrared thermography, as a nondestructive tool for discontinuity detection, is based on the principle that heat transfer in any material is affected by the presence of subsurface discontinuities or any other change in material thermal properties. Variations in heat flow cause localized energy differences on the surface of the test object, which can be measured using an infrared detector or radiometer. Through data processing, the measured infrared radiation levels are transformed into their corresponding temperature distributions and recorded in the form of thermograms (isotherm plots). Variations in the thermogram can indicate the presence of subsurface discontinuities in the test object. Discontinuity detection and characterization in civil engineering structures require active thermography, which involves an external heat source and transient heat transfer phenomena. Moreover,

discontinuity characterization requires time resolved infrared thermography. Using this technique, the surface temperature of the test object is monitored and analyzed as a function of time, instead of being monitored statically at only one particular point in time.

ANALYTICAL MODELING
The differential equations that govern the temporal and spatial variations of temperature in an object under transient conditions are difficult to solve explicitly, except for simple boundary conditions. For realistic problems, however, it is possible to obtain approximate solutions using numerical tools such as the finite difference or finite element methods. In this study, a commercially available, multipurpose finite element program is used to obtain the temperature histories at discrete points of models representing concrete substrates with bonded fiber reinforced polymer layers. A series of parametric studies has been carried out to gain an understanding of the effects of various parameters on the thermal response of test objects with discontinuities (Starnes and Carino, 2003). Validation of the finite element models against experimental results was presented in Starnes et al. (2003). This paper reviews the portion of the study dealing with the effects of discontinuity width, including estimation of discontinuity width and the minimum detectable discontinuity width. Simulation Models To reduce computation times, the numerical simulations were performed using two dimensional finite element models. The simulation object consisted of a 100 mm (3.9 in.) long by 20 mm (0.8 in.) thick concrete slab covered with several layers of carbon fiber reinforced polymer. Hand layup carbon fiber reinforced polymer composite laminates were modeled in the simulations. Each laminate of carbon fiber reinforced polymer was 0.5 mm (0.02 in.) thick. Subsurface discontinuities varying in depth and width were placed at the center of the model. Since the subsurface discontinuity was located at the centerline of the model, the simulation was simplified by using plane symmetry. Thus, only one half of the object was modeled. Internal discontinuities were modeled as air gaps with a thickness of 0.1 mm (3.9 10-3 in.). Discontinuity widths ranged from 3.0 to 25 mm (0.1 to 1 in.). The models were arranged into three different groups depending on the depth of the discontinuity: delaminations in the plates at 0.5 mm (0.02 in.) deep; debonds at the laminate concrete interface at 1.5 mm (0.06 in.) deep; and concrete spalls 2.5 mm (0.1 in.) deep. Figure 1 illustrates the geometries of the models. The material properties of the model were those of concrete for the substrate, air for the discontinuities and carbon fiber reinforced polymer for the bonded composite. The carbon fiber reinforced polymer layer in direct contact with the concrete had the fibers running in the longitudinal X direction, while each of the adjacent laminates had its fiber direction rotated by 90 degrees. The model was composed of four noded solid quadrilateral elements. Each node had one degree of freedom: temperature. The simulation object was meshed using mapped meshing, since it allows the user to directly control the element size and type. The global element size was set to 0.5 mm (0.02 in.). Mesh refinement was applied to the thin composite layers and at the fiber reinforced polymer/concrete interface.

The analysis was defined as a transient heat transfer problem. A square pulse of intensity 20 kW/m2 (6340 Btu/hft2) and duration 2 s was applied uniformly to the top surface of the test object. Adiabatic conditions (dT/dx = 0 and dT/dy = 0) were assumed for the additional surfaces. The assumption that there is no temperature gradient at the bottom of the concrete is reasonable because, during the short time interval for thermography testing, the heat pulse does not reach this surface. The assumption that there is no temperature gradient on the vertical surface implies that the specimen is sufficiently large relative to the discontinuity width so that heat flow is unaffected at the boundary. The latter assumption was verified from calculated temperature contours. The initial temperature for all the simulations was 296 K (73.4 F), which represented ambient temperature. For simplification, uniform heating, no convective and radiative losses and perfect contact between layers were assumed for the analyses. Thermal Response Parameters In these analyses, surface temperatures above the discontinuity and in the background where no internal discontinuity was present were recorded as functions of time (Figure 2), from which the following quantities were determined: maximum surface temperature; thermal signal; and time to maximum signal. The maximum surface temperature occurs above the discontinuity and at the end of the thermal pulse. The thermal signal is defined as (1) T = Td Tb

where T = the thermal signal Td = the surface temperature above the discontinuity Tb = the surface temperature above the background (sound material). The time to reach maximum signal tS is the time from the start of the heating until Tmax is reached. Effect of the Width of the Discontinuity The size of subsurface discontinuities could affect the performance of the fiber reinforced polymer composite bonded to concrete.

Figure 2 Schematic of the infrared thermographic method to detect the presence of a discontinuity based on surface temperature differences.

(a)

(b)

(c)

Figure 1 Geometry of models used in simulations to investigate the effects of discontinuity width: (a) delamination; (b) debond; (c) concrete spall.
Materials Evaluation/January 2005 57

Estimation Error (mm)

With this concern, requirements on allowable discontinuity size have been introduced by the International Code Council (ICC) Evaluation Services (2003). Among the conditions of acceptance, the ICC criteria state that discontinuities with areas larger than 1300 mm2 (2 in.2) should be repaired. The objectives of this parametric study were to understand the effect of discontinuity width on the thermal response, establish a procedure to estimate discontinuity width and determine the minimum width of a detectable discontinuity. With these objectives, the variation of surface temperature with distance from the center of the discontinuity must be considered. Figure 3 illustrates a typical temperature/distance plot at the time to maximum signal. It is seen that the surface temperature is fairly constant over the central part of the discontinuity and drops rapidly at the discontinuitys edge. It was decided to investigate whether the location of the inflection point in the temperature/distance curve could be used to estimate the discontinuity width. A similar idea was proposed independently by Vavilov (2000). The location of the inflection point may be computed by setting the second derivative of the distance versus temperature curve equal to zero (2) 2 TS = 0 at west x 2

as the estimated discontinuity width. The results of the estimation using the location of the inflection point are presented in Table 1. The estimation error, that is, the difference between the estimated and the actual width of the discontinuity, is also presented in Table 1. The results indicate that with this approach, discontinuity width may be underestimated or overestimated depending on discontinuity size and depth. For example, the width of large discontinuities tends to be underestimated, while the width of smaller discontinuities tends to be overestimated. The results given in Table 1 are presented in Figure 4a and they reveal that the estimation error increases with discontinuity depth, that is, the estimation of the width of delaminations (discontinuities between layers of fiber reinforced polymer) is more accurate than the estimation of the width of debonds (discontinuities at the fiber

4.0 3.0 2.0 1.0 0.0 -1.0 -2.0 0.0

(a)

Delaminations (0.5 mm deep) Debonds (1.5 mm deep) Spalls (2.5 mm deep)

where TS = the surface temperature x = the distance from the centerline along the X axis west = the estimated distance of the edge of the discontinuity from the centerline.

60.0 Surface Temperature (C) 55.0

5.0

10.0

15.0

20.0

25.0

Estimated Width (mm)

25.0
T
o

Minimum Width (mm)

50.0 45.0 40.0

Inflection point

max

= 0.1 C = 1.0 C = 2.0 C

o o

20.0 15.0 10.0 5.0 0.0 0.0

T T

max

max

Estimation of 1/2 width of discontinuity 5.0 10.0 15.0 20.0 25.0 30.0 Distance from Centerline (mm)

35.0 0.0

Figure 3 Estimation of discontinuity width from a temperature profile. For each simulation, the surface temperature at the time of the maximum signal was recorded as a function of distance from the centerline of the discontinuity. The second derivative was approximated by numerical differentiation using a spreadsheet and the point where the second derivative equaled zero was taken

0.5

1.0

1.5

2.0

2.5

3.0

(b)

Discontinuity Depth (mm)

Figure 4 Discontinuity width data: (a) discontinuity width estimation error as a function of discontinuity width and depth; (b) minimum width of detectable discontinuity as a function of discontinuity depth and instrument sensitivity.

Table 1 Estimated discontinuity width based on point of inflection in temperature profiles from numerical simulations Discontinuity Depth 0.5 mm (0.02 in.) 0.5 mm (0.02 in.) 0.5 mm (0.02 in.) 0.5 mm (0.02 in.) 1.5 mm (0.06 in.) 1.5 mm (0.06 in.) 1.5 mm (0.06 in.) 1.5 mm (0.06 in.) 2.5 mm (0.1 in.) 2.5 mm (0.1 in.) 2.5 mm (0.1 in.) 2.5 mm (0.1 in.) Actual Width 25.0 mm (0.984 in.) 14.0 mm (0.551 in.) 5.0 mm (0.197 in.) 3.0 mm (0.118 in.) 25.0 mm (0.984 in.) 14.0 mm (0.551 in.) 5.0 mm (0.197 in.) 3.0 mm (0.118 in.) 25.0 mm (0.984 in.) 14.0 mm (0.551 in.) 5.0 mm (0.197 in.) 3.0 mm (0.118 in.) Estimated Width 24.0 mm (0.95 in.) 13.1 mm (0.52 in.) 4.6 mm (0.18 in.) 3.1 mm (0.12 in.) 23.2 mm (0.91 in.) 12.5 mm (0.49 in.) 5.7 mm (0.22 in.) 4.9 mm (0.19 in.) 23.3 mm (0.92 in.) 12.7 mm (0.50 in.) 6.7 mm (0.26 in.) 6.0 mm (0.24 in.) Estimation Error 1.0 mm (0.039 in.) 0.9 mm (0.035 in.) 0.4 mm (0.016 in.) +0.1 mm (0.004 in.) 1.8 mm (0.071 in.) 1.5 mm (0.059 in.) +0.7 mm (0.028 in.) +1.9 mm (0.075 in.) 1.7 mm (0.067 in.) 1.3 mm (0.051 in.) +1.7 mm (0.067 in.) +3.0 mm (0.118 in.)

58 Materials Evaluation/January 2005

reinforced polymer/concrete interface) or concrete spalls. Most estimation errors, however, are only on the order of a few millimeters, which is very encouraging. An important issue that needs to be addressed is the minimum width of detectable discontinuities. The minimum width of detectable discontinuities is influenced highly by the thermal sensitivity of the infrared detector or camera and environmental noise. Thus, the minimum detectable discontinuity depends on the detectable maximum thermal signal. The minimum detectable discontinuity was estimated for three cases: Tmax equal to 0.1 K (0.18 F), 1 K (1.8 F) and 2 K (3.6 F). Analysis of the finite element mesh output revealed that, for a 0.1 mm (3.9 10-3 in.) thick air void, the maximum signal Tmax could be expressed as the following empirical hyperbolic functions of the width of the discontinuity: for delaminations (depth = 0.5 mm [0.02 in.]): (3) Tmax = 20.95 0.4 ( w 1.86) 1 + 0.4 ( w 1.86)

decreases, the detectable discontinuity width also decreases. As the depth of the discontinuity increases, so does the detectable discontinuity size. While these relationships are not unexpected, the simulations revealed the quantitative aspects of the relationships.

EXPERIMENTAL STUDY
Test Configuration The second phase of the research program involves laboratory studies. The results presented in this paper focus on measurement of emissivity, detection of debonds and the estimation of the width of a debond in a controlled discontinuity specimen. A controlled discontinuity specimen was constructed by using a 610 by 250 by 45 mm (24 by 9.8 by 1.8 in.) precast concrete slab as the substrate. Two pultruded carbon fiber reinforced polymer plates were bonded to the substrate parallel to each other and running lengthwise as shown in Figure 5. The plates contained unidirectional carbon fibers. Each plate had the following dimensions: 609 by 102 by 1.3 mm (24 by 4 by 0.05 in.). The composite plates were bonded to the concrete substrate using bonding epoxy supplied by the composite manufacturer. Eight discontinuities were created by placing different low thermal conductivity materials at the interface between the concrete substrate and the fiber reinforced polymer. Each discontinuity was approximately 25 by 25 mm (1 by 1 in.) in plan. The materials used to simulate debonds included air, a low conductivity fabric, a ceramic insulating sheet and several types of plastic (Starnes et al., 2002). The components of the test system are shown in Figure 6. The infrared camera was connected to a computer with software for data capture and analysis.

for debonds (depth = 1.5 mm [0.06 in.]): (4) Tmax = 5.95 0.14 ( w 1.9) 1 + 0.14 ( w 1.9)

for concrete spalls (depth = 2.5 mm [0.1 in.]): (5) Tmax = 3.12 0.1 ( w 1.87 ) 1 + 0.1 ( w 1.87 )

In these equations, the constants in front of the quotients represent maximum values of a thermal signal for large, 0.1 mm (3.9 10-3 in.) thick discontinuities. Using Equations 6, 7 and 8, one can estimate the minimum width of detectable discontinuities for different values of a maximum signal. These estimated values are shown in Figure 4b for three values of Tmax. Thus, the minimum width of detectable discontinuities could be computed using Equations 6, 7 and 8. The results from the calculations are presented in Figure 4b. Observation of the results presented in Figure 4b indicates that as the magnitude of the signal that can be detected decreases, the minimum width of detectable discontinuities also decreases. For a given magnitude of maximum signal, the relationships between minimum width of detectable discontinuity and depth could be approximated by the following power functions: (6) w min = 1.61 + 0.36 d 0.53 for Tmax = 0.1 K (0.18 F )

(7)

w min = 1.86 + 0.61d 2.27 for Tmax = 1 K (1.8 F )

(8)

w min = 1.68 + 1.33d 2.87 for Tmax = 2 K ( 3.6 F )

Figure 5 Controlled discontinuity test specimen.

where wmin = the minimum width of detectable discontinuities in millimeters d = the depth of the discontinuity in millimeters. The results indicate that near surface delaminations should be detectable down to widths of 2 mm (0.08 in.). To summarize, the results of these numerical simulations indicate that the minimum discontinuity width required for detection depends on the depth of the discontinuity and the detection limit of the infrared system. As the maximum signal required for detection

Figure 6 Infrared thermographic test configuration.

Materials Evaluation/January 2005 59

(a)

(b)

Figure 7 Visual image and thermogram of a test object during qualitative detection of internal discontinuities created artificially with different materials. Two 250 W infrared heating lamps mounted at 200 mm (7.9 in.) on center were used for the thermal input. An aluminum frame was made to hold the heating lamps and an aluminum shutter. The shutter was necessary to block radiation from the lamps after they were turned off. At the end of the heating pulse, an electrical trigger turned off the lamps and the electromagnet and the shutter fell in front of the lamps. The shutter system was needed to create a sharp heat pulse as used in the numerical modeling. An infrared camera in combination with data acquisition and real time software was used to record the surface temperature. The infrared camera operated in the long wavelength infrared spectral band, thus minimizing the atmospheric attenuation of the received radiation. Data acquisition was automatically triggered when the heating lamps were turned on. Determination of Material Emissivity The infrared detector measures the radiation emitted by the test object. This radiation is a function of the surface temperature and the emissivity of the specimen. The value of the emissivity is needed for accurate measurement of the surface temperature of the test object. In this investigation, the emissivity of pultruded carbon fiber reinforced polymer was determined so that the surface temperature could be determined accurately for comparison with finite element mesh simulations. Estimation of the width of the discontinuity, however, does not require knowledge of the actual surface temperatures since the locations of the points of inflection of the surface temperature profile do not depend on the absolute value of the surface temperature. The emissivity measures the capability of a material to emit radiation. In particular, emissivity is the ratio of the radiance of a body at a given temperature to the radiance of a black body at the same temperature (ASTM, 2001). Accurate determination of surface emissivity is key for the correct measurement of surface temperatures by using an infrared camera. Standard methodology for determining emissivity is described in ASTM E 1933 (ASTM, 2000). The contact thermometer method described in the standard was used in this experimental study. First, surface temperatures of the fiber reinforced polymer plate were measured using a copper/constantan thermocouple (ANSI Type T, special limits, 0.01 mm [3.9 10-4 in.] diameter). The thermocouple was embedded in the fiber reinforced polymer so that half of the perimeter of the wire was in contact with the composite and the other half was in contact with air. A small notch was cut into the plate and epoxy adhesive was used to hold the thermocouple in place. The test specimen was located at a distance of 0.55 m (21.7 in.) from the infrared camera. Since emissivity varies with the angle of view, the specimen was placed perpendicular to the line of view of the infrared camera. A value of emissivity was assumed and surface temperatures were measured with the infrared camera at three locations next to the thermocouple using the real time analysis software. The emissivity value was varied until there was agreement
60 Materials Evaluation/January 2005

between the temperature recorded with the thermocouple and the temperature recorded with the camera. One hundred and sixty-five measurements were obtained and the average emissivity was 0.80, with a standard deviation of 0.016. Thus, the expanded uncertainty interval for the emissivity of the pultruded plate is 0.80 0.03. Infrared Thermographic Test Procedure The first series of measurements were qualitative in nature and were intended to test whether each simulated discontinuity embedded in the test object could be detected. The surface of the specimen was heated by sweeping an infrared heat lamp along the length of the fiber reinforced polymer at a distance of 50 mm (2 in.) from the surface and at a speed of approximately 150 mm/s (5.9 in./s). This technique is similar to the method used in practice (Hawkins et al., 1999). A visual image and a thermogram (infrared image) of the test object are presented in Figure 7. Observation of the thermogram revealed that all eight discontinuities were detectable. The most visible discontinuities were numbers 5, 6 and 8, which corresponded to low conductivity fabric, air and ceramic sheet, respectively. During the qualitative test, it was observed that the air filled discontinuity number 6 did not appear square as intended. The bonding epoxy apparently flooded over the wire dam that was used to exclude epoxy and create the intended discontinuity (air gap or debond). Thus, discontinuity number 6 was chosen for verification of the procedure for width determination. The specimen was tested using the experimental configuration described in Figure 6. The specimen was placed at 330 mm (13 in.) from the camera lens. The test object was placed so that the air filled discontinuity was positioned at the centerline between the two heating lamps. The duration of the thermal pulse was set to 10 s and the thermogram data were recorded at 15 Hz for a period of 60 s. The surface temperatures above the discontinuity (Td) and above the bonded plate near the discontinuity (Tb) were obtained as a function of time. The maximum thermal signal (Td Tb) was 2.7 K (4.9 F), which occurred at 12 s after the beginning of the test. The maximum surface temperature registered was 302 K (83.1 F). Estimation of Discontinuity Width To estimate the width of the air filled discontinuity, the thermogram at the time of maximum signal was analyzed. The surface temperatures of the region surrounding the discontinuity were retrieved using an area measuring tool. The retrieved area had dimensions of 35 by 28 pixels. Since the analysis software provides surface temperatures at each pixel, a conversion between pixel size and actual physical size was required. The conversion was done using the known dimension of a heat flux sensor and a piece of tape located on the surface of the specimen as references. The heat flux sensor and the tape had actual widths of 25.4 and 18.7 mm (1 and 0.7 in.), respectively, and in the thermogram image the corresponding dimensions were 23 and 17 pixels. Thus, each pixel corresponded to a 1.1 mm (0.04 in.) square and the analyzed region surrounding the discontinuity was 38.5 by 30.8 mm (1.5 by 1.2 in.). This area box is shown in Figure 8.

Figure 8 Thermogram of an air void at the time of maximum signal.

The plan dimensions of the discontinuity were estimated by analyzing the surface temperatures along three lines, as shown in Figure 8. Unlike the finite element mesh output, the surface temperature data along each line did not vary smoothly and a five point moving average was computed to smooth the data. This smoothing was also applied to the second derivative of the temperature profile. The smoothed surface temperature and the smoothed second derivative along the horizontal line (LI01) are presented in Figure 9a. The roots of the second derivative (d2T/dx2 = 0) were located at 8.8 and 25.6 mm (0.3 and 1.01 in.) from the origin. Thus, the estimated width of the discontinuity along the horizontal line is 16.8 mm (0.7 in.). A similar procedure was used for the vertical line (LI02). In this

(a)

(b)

(c) Figure 9 Surface temperature profile and second derivative with respect to distance: (a) along a horizontal line (LI01), smoothed; (b) along a vertical line (LI02), smoothed; (c) along a diagonal line (LI03), unsmoothed.
Materials Evaluation/January 2005 61

(a)

(b)

(c)

Figure 10 Air void: (a) thermogram at time of maximum signal; (b) visual image following removal of polymer; (c) close up of the visual image.

case, the surface temperature profile and the second derivative curve were smoothed using the moving average of five points and three points, respectively. The resulting smoothed plots are shown in Figure 9b. The roots of the second derivative were located at 4.7 and 21.0 mm (0.2 and 0.8 in.) from the origin of the line. Thus, the estimated width of the discontinuity along the vertical line is 16.3 mm (0.6 in.). Finally, the diagonal dimension was estimated. For this case, temperatures along the diagonal of the surface temperature matrix were used. The surface temperatures were extracted every 3.97 mm (0.16 in.) along the diagonal. No smoothing of the temperature profiles and second derivative was required in this case. The temperature profile and its second derivative are presented in Figure 9c. The second derivative provided four roots. Based on observation of the surface temperature profile, the outer roots were used for estimation of the width. The outer roots of the second derivative were 1.4 and 29 mm (0.06 and 1.1 in.) and the estimated diagonal width of the air void is 27.6 mm (1.1 in.). To verify the estimated dimensions, the portion of fiber reinforced polymer plate above the air discontinuity was carefully removed by using a miniature high speed cutoff wheel. This procedure revealed the shape of the air void. A visual image of the exposed air void is presented in Figure 10 and compared with its image during thermographic testing. The measured dimensions of the actual air void indicated a horizontal width of 17 mm (0.7 in.), a vertical width of 17 mm (0.7 in.) and a diagonal width of 31 mm (1.2 in.). Table 2 summarizes the estimated and the actual widths of the air void.
Table 2 Estimated and actual widths of air void Horizontal line Vertical line Diagonal line Estimated Width 17 mm (0.67 in.) 16 mm (0.63 in.) 28 mm (1.1 in.) Actual Width 17 mm (0.67 in.) 17 mm (0.67 in.) 31 mm (1.2 in.)

discontinuity, where there was no epoxy between the fiber reinforced polymer and the concrete. These results show that curve smoothing using a moving average algorithm is an effective means for dealing with noisy signals. The number of points used in the smoothing may depend on the surface temperature profile. The goal is to easily identify the roots of the second derivative curve while minimizing the distortion of the original data. Thus, the number of points used to smooth the curves should be kept to a minimum. Additional studies on estimating discontinuity size should lead to guidelines for proper smoothing of temperature profile data.

CONCLUSION
The finite element method was used to gain quantitative understanding of the effect of discontinuity width on the surface temperature response to pulsed heating. The objective was to determine whether characteristics of the temperature distribution could be used to measure discontinuity widths and to establish the key factors affecting the smallest detectable discontinuity width. The results indicate that the smallest detectable discontinuity width increases nonlinearly with increasing depth. In addition, the smallest detectable discontinuity width is a function of the required thermal signal for detection by the test system. The experimental portion of the study involved thermographic testing of a controlled discontinuity specimen. The experiments included measurement of the emissivity of the fiber reinforced polymer surface, evaluation of different materials for simulating discontinuities and estimation of the width of an irregularly shaped air void. The estimation of the width was done by determining the inflection points of the surface temperature profile above the discontinuity. Smoothing of the temperature profiles and second derivatives using a moving average filter was effective in reducing noise in the data. With this procedure, the size of the discontinuity was estimated within a few millimeters of the actual size.

REFERENCES
The maximum estimation error was in the diagonal dimension, where the width was underestimated by 3 mm (0.12 in.). Observation of the region with discontinuities revealed that in the lower left corner of the discontinuity there was an unbonded area with a thin layer of epoxy on the concrete (Figure 10). Thus, the thickness of the air void was reduced in this region. The estimated diagonal width matched the distance between the upper right corner of the discontinuity and the edge of the thin layer of epoxy. The thinner air gap at the lower left corner produced a lower signal than the rest of the
62 Materials Evaluation/January 2005

ASTM International, E 193399a, Standard Test Methods for Measuring and Compensating for Emissivity Using Infrared Imaging Radiometers, West Conshohocken, Pennsylvania, ASTM International, 2000. ASTM International, E 1316, Standard Terminology for Nondestructive Examinations, West Conshohocken, Pennsylvania, ASTM International, 2001. Alampalli, S., O. Hag-Elsafi, J. OConnor, T. Conway and A. Aref, Use of FRPs for Bridge Components and Methods of Performance Evaluation, Proceedings of the 2001 Structures Congress and Exposition, Reston, Virginia, American Society of Civil Engineers, 2001. Hawkins, G.F., E. Johnson and J. Nokes, Typical Manufacturing Flaws in

FRP Retrofit Applications, Proceedings from NIST Workshop on Standards Development for the Use of Fiber Reinforced Polymers for the Rehabilitation of Concrete and Masonry Structures, NISTIR 6288, Tucson, Arizona, National Institute of Standards and Technology, 1999. ICC Evaluation Service, Interim Criteria for Inspection and Verification of Concrete Acceptance and Reinforcement and Unreinforced Masonry Strengthening Using Fiber-reinforced Polymer (FRP) Composite Systems, AC178, Whittier, California, ICC Evaluation Service, 2003. Starnes, M.A. and N.J. Carino, Infrared Thermography for Nondestructive Evaluation of Fiber Reinforced Polymer Composites Bonded to Concrete, Technical Report NISTIR 6949, Gaithersburg, Maryland, National Institute of Standards and Technology, 2003.

Starnes, M.A., N.J. Carino and E.A. Kausel, Quantitative Infrared Thermography for Quality Control of Concrete Structures Strengthened with FRP Composites, Proceedings of the American Concrete Institute 5th International Conference, ACI SP 209-41, Farmington Hills, Michigan, 2002. Starnes, M.A., N.J. Carino and E.A. Kausel, Infrared Thermography Testing for Quality Control of Concrete Structures Strengthened with FRP Composites, ASCE Journal of Materials in Civil Engineering, Vol. 15, No. 3, 2003, pp. 266-273. Vavilov, V.P., Three-dimensional Analysis of Transient Thermal NDT Problems by Data Simulation and Processing, Proceedings of SPIE, Thermosense XXII, Vol. 4020, Orlando, Florida, SPIE, 2000, pp. 152-163.

Materials Evaluation/January 2005 63