Defect Detectability in CFRP Composites Using Long Pulse Thermography: A Comparative Study of Five Defect Types

Compared with short-pulse thermography, the longer heating period used in LPT can improve sensitivity to defects located deeper below the surface because more thermal energy is introduced into the laminate. Thermal contrast is then evaluated primarily during the cooling phase, when local differences in thermal diffusivity, thermal conductivity, and heat capacity influence the rate of temperature decay (Almond et al., 2017; Kalyanavalli et al., 2018; Huang et al., 2022).

Step Heating Thermography (SHT) uses a gradual or rectangular heating input and records the temperature evolution during and after heating (Balageas & Roche, 2014; Roche & Balageas, 2015; Zhuo et al., 2023). LPT can be regarded as a long-pulse implementation of active thermography in which the excitation resembles a step-heating input; however, the analysis in the present work focuses on the post-heating transient cooling response rather than on a steady-state temperature rise. Therefore, the experimental setup may be described as SHT-type excitation combined with LPT analysis: SHT defines the heating input, whereas LPT defines the post-excitation transient evaluation used to assess defect detectability.

Both LPT and SHT are used in aerospace, automotive, and manufacturing applications where safety and structural integrity are critical (Carslaw & Jaeger, 1959; Maldague, 2001b; Vedula, 2010). Their non-contact and non-invasive character makes them suitable for inspecting composite parts, bonded joints, and layered structures without damaging the component (Shepard, 2003; Sun, 2006; Kalyanavalli et al., 2018; Wang et al., 2018).

Recent advances in SHT and related active thermographic methods include the use of Thermographic Signal Reconstruction (TSR), early detection and characterization procedures, and virtual heat-flux reconstruction (Favro et al., 1995; Deák & Czigány, 2009; Świderski & Vavilov, 2009; Manohar & di Scalea, 2013; Balageas & Roche, 2014; Ghadermazi et al., 2015; Zhuo et al., 2023). These approaches improve defect contrast, reduce noise, and support quantitative interpretation of thermal sequences.

For the present study, the important implication is that post-processing can enhance defect visibility, but the final detectability of a defect still depends on its geometry, embedded depth, and thermophysical mismatch with the surrounding CFRP laminate (Carslaw & Jaeger, 1959; Ringermacher et al., 1998; Almond & Pickering, 2012; Wang et al., 2022).

Building on these principles, the following section describes the specimen design and experimental procedure used to evaluate LPT performance across five defect classes.

The specimen, with dimensions of 850 × 450 mm, was manufactured from a unidirectional carbon-fiber/epoxy prepreg (CFRP UD) cured in an autoclave. The material system therefore represents a typical aerospace CFRP laminate with continuous UD carbon reinforcement and an epoxy matrix. Additional/foreign elements used to simulate defects included prepreg foil, paper inserts (Figure 1) and graphite foil.

Figure 1.

Paper inserts used to simulate interlayer foreign-body inclusions.

The specimen contained five types of defects, simulating three types of composite damage:

• Row 1 – prepreg foil (simulates interlayer insertion, a manufacturing defect caused by the worker placing the prepreg layers in the mold, Figure 2);

• Row 2 – flat-bottom holes (simulate delamination caused, for example, by low-energy impacts, Figure 2);

• Row 3 – paper inserts (simulate interlayer insertion caused by the inclusion of a foreign body, e.g., a piece of nitrile glove, in the structure during the laying of layers in the mold at the production stage, Figure 2);

• Row 4 – flat-bottom holes filled with epoxy resin (simulate resin-rich areas, Figure 2);

• Row 5 – graphite foil (simulates interlayer insertion, a manufacturing defect caused by the worker placing the prepreg layers in the mold, Figure 2).

Figure 2.

A – Prepreg foil; B – Flat-bottom hole; C – Paper insert; D – Flat-bottom hole filled with epoxy resin; E – Graphite foil.

The triangular planform of defect A and the square planform of defect C were intentionally selected to distinguish insert-type defects from the circular flat-bottom holes and to verify whether non-circular geometries could be reproduced in the thermal indication. The remaining defects were circular because they were manufactured as drilled flat-bottom holes or circular inserts. In LPT images, lateral heat diffusion tends to smooth sharp boundaries; therefore, triangular and square defects may appear as rounded thermal footprints during post-processing.

The specimen was a carbon composite plate of variable thickness. Based on thickness, the specimen can be divided into three zones:

• Zone 1 – 16 layers;

• Zone 2 – transitional;

• Zone 3 – 24 layers.

Holes and inclusions simulating defects were placed in all three zones. Zones 1 and 3 each contain 25 defects arranged in 5 rows and 5 columns. The transitional zone contains 10 defects arranged in 5 rows and 2 columns. Photographs of the specimen from both sides, detailing the transition zone, are shown in Figure 3. A schematic diagram of the specimen is shown in Figure 4.

Figure 3.

Photographs of the specimen (A – bottom side, B – top side).

Figure 4.

Schematic diagram of the specimen.

The specimen was manufactured using the autoclave process, following the thermal cycle shown in Figure 5.

Figure 5.

Heating cycle graph for specimen manufacturing.

The experiment used a rectangular long-pulse excitation. From the system point of view, this corresponds to the SHT excitation mode available in the C-CheckIR system; from the data-analysis point of view, the inspection was evaluated as LPT because the defect response was assessed primarily during the transient cooling phase after heating. Thus, SHT and LPT are not separate inspections in this work: SHT describes the heating input, whereas LPT describes the post-excitation transient analysis. The specimen, with dimensions of 850 × 450 mm, was divided into four test areas because the field of view of the C-CheckIR device is 430 × 340 mm at an operating distance of 400 mm. The sensor head stimulated the test area with an optical excitation source (halogen lamp) of 2.2 kW at 230 V. The standoff distance of the heat source was selected to avoid local heat concentration and to provide nearly uniform thermal stimulation over the inspected surface. The camera used was an Automation Technology GmbH IRS-336 infrared camera with a maximum resolution of 640 × 512 pixels, a frame rate of 60 Hz, and thermal sensitivity below 30 mK. The camera was positioned so that only the test specimen was visible, in order to minimize reflections from the surrounding area. Thermal image sequences were acquired using irNDT-Mobile and analyzed with the Pulse/Transient Measurement module, which processes the time-dependent temperature response after excitation. The Lock-In Evaluation module is also available in the software for phase and amplitude evaluation of periodically modulated thermal data; however, the present detectability assessment used the transient cooling response after long-pulse heating. The measurement system is shown schematically in Figure 6.

Figure 6.

Schematic diagram of the thermography test configuration.

Figure 7 shows a photograph of the C-CheckIR. The system comprises a control unit with software, a testing unit (infrared camera, excitation source, and stand), and connecting cables.

Figure 7.

Photograph of the C-CheckIR system during specimen testing.

The acquisition and excitation settings used in the experiment are listed in Tables 1 and 2.

In Table 1, Sync Mode describes how image acquisition was triggered; Free Run indicates that the camera recorded continuously without an external synchronization trigger. Acquisition Duration is the total recording time of the thermal sequence, and Acquisition Frame Rate is the number of infrared frames recorded per second. In Table 2, Rect. Modulation denotes a rectangular on/off heating signal; Amplitude Low and Amplitude High define the relative lamp power levels; Excitation Period is the complete modulation cycle; Duty Time is the active heating duration within that cycle; and the Rising and Falling Ramp Times smooth the transition between power levels. The number of measurements corresponds to the total number of recorded frames, whereas the Best Results Measurement identifies the frame selected as providing the most interpretable defect contrast.

The LPT results reveal a pronounced dependence of detectability on defect type. The highest detection rate was obtained for flat-bottom holes filled with epoxy resin, for which all 12 defects were identified, corresponding to 100% detectability. This defect class also exhibited the most stable response, with the lowest mean standard deviation of 0.92 mm. At the opposite extreme, graphite foil remained undetected at all 12 locations, indicating that, under the present LPT configuration, this defect class produced insufficient thermal contrast relative to the surrounding CFRP.

The intermediate classes showed distinct error patterns. Prepreg foil was detected in 8 out of 12 cases (66.7%); once detected, it produced a mean error of only −0.58 mm (−2.9%), indicating that defect size was reproduced relatively accurately when a measurable response existed. Flat-bottom holes were detected in 10 out of 12 cases (83.3%), yet their dimensions were systematically underestimated, with a mean error of −2.61 mm (−12.4%) and a mean measured-to-nominal ratio of 0.88. Paper inserts were also detected in 10 out of 12 cases (83.3%), but this group showed the most unstable sizing behavior. The overall mean relative error was +21.0%, although this value was strongly influenced by an outlier at the 6 mm location, where a mean measured size of 19.63 mm was obtained. This result is best interpreted as an anomalous or merged indication rather than a physically realistic size estimate. Flat-bottom holes filled with epoxy resin were not only detected in all cases, but also systematically overestimated. The mean sizing error for this group was +2.51 mm (+19.3%), and the corresponding mean measured-to-nominal ratio reached 1.19. In practical terms, LPT was highly sensitive to resin-rich features, but the thermographic footprint extended beyond the nominal geometrical boundary of the defect. The overall detectability ranking can therefore be expressed as: flat-bottom holes filled with epoxy resin > flat-bottom holes ≈ paper inserts > prepreg foil >>> graphite foil.

Table 3 confirms that detectability and accuracy are not equivalent. For example, prepreg foil exhibited a lower detection rate than flat-bottom holes, but once detected it was sized more accurately. Conversely, flat-bottom holes filled with epoxy resin achieved the best detectability and repeatability, although they were consistently oversized. This distinction is important for NDT practice because the best-detected defect is not necessarily the most accurately sized one.

When all defect classes are considered together, the detection rate increases with nominal size. For 6 mm defects, only 3 out of 10 cases were detected (30.0%); for 13 mm defects, 10 out of 15 (66.7%); for 19 mm defects, 11 out of 15 (73.3%); and for both 25 mm and 38 mm defects, 8 out of 10 (80.0%). These results indicate that the practical detection threshold for the present LPT configuration lies close to 13 mm. Defects of 6 mm are generally below the robust detectability limit, except when the defect class generates particularly strong thermal contrast, as observed for flat-bottom holes filled with epoxy resin. The size-dependent trend is illustrated in Figure 10.

Figure 10.

Detection rate as a function of nominal defect dimension for each defect type. The figure shows that 6 mm defects are generally below the robust LPT detectability limit, whereas detectability increases markedly from approximately 13 mm onward.

Dimension alone, however, does not fully explain the observed behavior. The defect response also depends strongly on depth and defect character, as reflected by the row-to-row differences defined during specimen fabrication. Figure 11 compares the measured and nominal dimensions for each defect type, and the following local comparisons show that defects of the same nominal dimension can produce different LPT responses. At a nominal dimension of 13 mm, the mean measured size followed the order: prepreg foil (12.08 mm) ≈ flat-bottom hole (12.15 mm) < paper inserts (14.08 mm) < flat-bottom hole filled with epoxy resin (15.58 mm). At 19 mm, the order was: flat-bottom hole (14.46 mm) < paper inserts (18.23 mm) ≈ prepreg foil (18.50 mm) < flat-bottom hole filled with epoxy resin (20.75 mm). At 25 mm: paper inserts (22.26 mm) ≈ flat-bottom hole (22.72 mm) < prepreg foil (25.38 mm) < flat-bottom hole filled with epoxy resin (28.63 mm). At 38 mm: flat-bottom hole (35.30 mm) < prepreg foil (36.72 mm) < paper inserts (38.10 mm) < flat-bottom hole filled with epoxy resin (39.72 mm).

Figure 11.

Mean measured dimension versus nominal dimension for each defect type. The diagonal line represents perfect agreement between the measured and nominal dimensions. Flat-bottom holes are predominantly underestimated, whereas flat-bottom holes filled with epoxy resin are systematically overestimated.

These sequences show that defects of the same nominal diameter may produce substantially different apparent dimensions in LPT, indicating that the thermal signature is governed by more than geometry alone. The through-thickness position defined in the specimen design, the local stacking sequence, and the thermophysical contrast between the defect and the surrounding laminate all contribute to the final response. This effect is visible in Figure 9 and in the local comparisons between columns. For example, for the 19 mm defects located in column 6, the mean measured size was 11.28 mm for flat-bottom holes, 21.20 mm for paper inserts, and 20.77 mm for flat-bottom holes filled with epoxy resin, whereas the prepreg foil at the same nominal location was not detected. Similarly, for the 13 mm defects in column 9, the mean measured sizes were 11.18 mm for prepreg foil, 13.56 mm for flat-bottom holes, 17.04 mm for paper inserts, and 17.71 mm for flat-bottom holes filled with epoxy resin.

The smallest defects provide the clearest evidence of the combined depth–dimension effect. The 6 mm flat-bottom holes were not detected, nor were the 6 mm prepreg foil or graphite foil inserts, while the 6 mm flat-bottom holes filled with epoxy resin were detected at both locations, with mean measured dimensions of 8.72 mm and 9.63 mm. The only other 6 mm response was the aforementioned 19.63 mm paper-insert indication, which should be interpreted as an anomalous case. Therefore, for small defects, the decisive factor is not merely nominal size, but whether the defect class and its depth placement are capable of generating sufficient thermographic contrast in LPT.

In summary, the LPT results demonstrate that defect detectability in CFRP is governed by the combined action of nominal dimension, through-thickness location, and defect nature. Flat-bottom holes filled with epoxy resin were the most visible and the most repeatable. Flat-bottom holes and paper inserts showed good but imperfect detectability with different sizing biases, prepreg foil remained moderately detectable but relatively accurate when visible, and graphite foil was effectively invisible under the present test conditions. These observations support the conclusion that defect detectability in CFRP cannot be characterized by size alone, and that any comparative evaluation of LPT performance must explicitly account for the defect class and its embedded depth.