Non-Destructive Testing of CFRP DCB Specimens Using Active Thermography ^†

Abstract

A Mode I Interlaminar Fracture Toughness test of carbon fiber-reinforced polymer laminates requires a double cantilever beam (DCB) specimen with a pre-implanted non-adhesive insert at the mid-plane to initiate delamination. However, the insert’s quality and placement within the DCB specimen can be problematic, necessitating non-destructive testing methods. In this study, active thermography is employed to inspect potential defects around the Teflon insert in the DCB specimens. Both uniform and non-uniform heating methods have been applied, and thermal images were analyzed to obtain quantitative information, such as the insert’s location and non-contact area. TSR-enhanced images were obtained using two variations in the classical thermographic signal reconstruction. The analyzed results confirmed the presence of non-contact areas in the DCB structures composed of both 22-layer and 24-layer CFRP prepregs. These areas may be attributed to residual air gaps formed during the hot-press molding of the DCB structures.

1. Research Motivation

Carbon fiber-reinforced polymer (CFRP) composites are prone to interlaminar cracking in structural applications [1]. As such, investigating their interlaminar fracture toughness is critical for the development and selection of composite structure designs. To facilitate subsequent fracture toughness testing, double cantilever beam (DCB) specimens with three different stacking sequences were fabricated, incorporating a non-adhesive film at the mid-plane on one side to initiate delamination [2,3]. To assess the integrity of the specimens, active infrared thermography [4] was employed to examine potential unbonded regions between the inserted film and the adjacent CFRP prepreg layers.

2. Infrared Thermography Experimental Method

Standard CFRP DCB coupon specimens are typically prepared from large test pieces fabricated using unidirectional carbon fiber (0.1mm thick) prepregs. We will refer to the test piece as the specimen hereafter. The specimens, measuring 160 mm long, 135 mm wide, and 2.1 mm thick, were created in 22-, 24-, and 26-layer configurations. To pre-induce cracks, a 0.03 mm thick layer of Teflon was inserted in the middle of each specimen. All specimens were hot-pressed at 40 kgf/cm² to achieve a final thickness of 2.1 mm.

On the back side of each completed CFRP specimen, two 500 W halogen lamps were used as the thermal excitation source. Infrared imaging was performed on the front side using an AVIO R500 thermal camera (Nippon Avionics Co., Ltd., Yokohama, Japan), with data recorded over a 20 s heating period at a frame rate of 3 Hz.

3. Thermal Image Analysis

After data extraction, relative temperature differences were calculated by subtracting the initial frame (baseline temperature) from each subsequent thermal image. These temperature differences were then analyzed as a function of time and fitted using polynomial models of degree, deg. In addition to direct time-temperature analysis, logarithmic transformations (lnt and ΔlnT) were also employed to construct polynomial fittings (Equations (1) and (2)). Polynomial degrees of 3 and 5 were adopted, resulting in a total of four different processing approaches (Figure 1).

ΔΤ(t) = a₀ + a₁ t + a₂ t² + a₃ t³ + ⋯ + a_deg t^deg

(1)

Δln(T) = a₀ + a₁ ln(t) + a₂ ln(t)² + a₃ ln(t)³ + ⋯ + a_deg ln(t)^deg

(2)

In order to improve signal clarity, the natural logarithm of temperature and time data was taken and then subtracted. This is equivalent to dividing the original data, and it was performed to avoid errors that can occur when polynomial fitting is applied to temperature difference data, especially when it contains zeros or negative numbers. The polynomial-fitted data was further processed by applying non-derivative, first-order, and second-order derivatives to create enhanced images. The clearest image showing the defect was selected from Figure 1 for further analysis. Temperature differences were categorized using a normal distribution model, dividing the thermal data into high-temperature (specimen background) and low-temperature regions (defects and aluminum foil) (Figure 2).

An analysis region was first defined within the thermal images, with its size set to be at least n times larger than the target area to be evaluated—for example, 6~8 times larger for the aluminum foil and 3~4 times larger for defect areas, as this yielded more stable results (Figure 3). A normal distribution analysis was then applied to this region. The number of low-temperature pixels was determined using a threshold defined as the mean temperature minus n times the standard deviation. Three classification criteria were used:

(1)

Pixels with temperature differences lower than the average (Avg);

(2)

Lower than the average minus one standard deviation (Avg − σ, or SD);

(3)

Lower than the average minus two standard deviations (Avg − 2σ, or SD2).

4. Results and Discussions

The low-temperature pixel regions were retained and visualized as binary images (Figure 4, showing S40-22 Deg3 as an example) to facilitate comparison. Since the number of aluminum foil pixels under the non-derivative condition using the Avg − 2σ (SD2) criterion closely corresponded to 1 cm²—the actual area of the foil—we adopted the SD2 threshold for aluminum foil area estimation. Subsequently, we identified the frame in the first- and second-order derivative results where the aluminum foil pixel count matched or closely approximated that of the non-derivative SD2 case. This frame was then used for further defect area analysis.

As shown in Figure 4, both potential defective and foil regions were compared. However, it was observed that in the second-order derivative condition, the aluminum foil pixel count significantly deviated from that of the non-derivative SD2 reference, and the defect region exhibited considerable noise. Therefore, the results from the second-order derivative condition were excluded from further analysis.

In the TSR-enhanced images, a suspected rectangular-shaped defect was observed in the 22-ply specimen, while a triangular-shaped suspected defect appeared in the 24-ply specimen. No distinct temperature anomalies were detected in the 26-ply specimen. For defect area estimation, the aluminum foil pixel count was consistently determined using the Avg − 2σ (SD2) criterion, and the corresponding defect pixel count was taken from the same frame. Additionally, several temperature profiles were drawn to extract the vertical and horizontal dimensions of the suspected defect regions using the Full Width at Half Maximum (FWHM) method.

For the 22- and 26-ply specimens, the upper and lower bounds of the defect area were estimated by multiplying the vertical and horizontal lengths. In contrast, the 24-ply specimen’s suspected defect area was calculated using the triangle area formula due to its shape. The estimated defect area ranges for each specimen, based on four different processing approaches—including polynomial fitting and logarithmic transformations—are summarized in

Table 1. Estimated defect area range (cm²) for S40-22 under different polynomial and statistical methods.

Method	Deg3	Deg3ln	Deg5	Deg5ln
Avg	42.2~44.7	40.9~44.4	41.2~45	42~44.6
SD	25.5~26.2	24.8~26.6	26~26.6	26~26.2
SD2	4.3~5.5	4~6.1	3.8~5.5	4.6~5.3

,

Table 2. Estimated defect area range (cm²) for S40-24 under different polynomial and statistical methods.

Method	Deg3	Deg3ln	Deg5	Deg5ln
Avg	54.2~54.4	53.4~54.4	53.9~54.4	53.8~54.2
SD	33.3~35.9	33.6~35	33.5~36.4	33.2~36.1
SD2	4.8~7.8	6~8	4~7.2	5~8.5

and

Table 3. Estimated defect area range (cm²) for S40-26 under different polynomial and statistical methods.

Method	Deg3	Deg3ln	Deg5	Deg5ln
Avg	55.8~57.8	55.2~56.9	55.9~57.1	55.5~57.6
SD	23.7~24.8	23.8~24.9	23.9~24.4	23.5~24.6
SD2	1.6~1.8	1.5~1.8	1.8~2	1.7~1.8

.

In particular, the selected pixel count within the region of interest for the 26-ply specimen was based on the average temperature (Avg) criterion (Table 3). This adjustment was made because the temperature difference between the low-temperature region and the surrounding area in the 26-ply specimen was relatively small (less than 1 °C), compared to over 2 °C in the other two specimens. As a result, the FWHM method yielded a broader width for the low-temperature region, leading to an overly large estimation of defect area, as shown in

Table 4. Defect area estimates (cm²) selected by FWHM analysis and defect-to-aluminum statistical threshold ratios.

	S40-22	S40-24	S40-26
Calculation	Defect SD/Aluminum	Defect SD/Aluminum	Defect Avg/Aluminum
Method	SD2	SD2	SD2
Area (cm2)	24.8~26.6	33.2~36.4	55.2~57.8

.