Introduction

Carbon fibre composite structures are used in a wide variety of applications because of their favourable strength/stiffness to weight ratios, especially in the aerospace and wind turbine industries. Carbon fibre reinforced polymers (CFRP) are suspectable to direct lightning strikes due to their semiconductive nature and ability to conduct current [1,2,3]. The internal damage mechanisms resulting from resistive heating and impact due to lightning strike have been described in detail by several researchers [4,5,6,7]. Lightning protection solutions have been proposed and studied [7,8,9,10,11]. However, lightning strike continues to damage CFRP structures and cause outages to aircraft and wind turbine (WT) blades because the damage has significant effects on the mechanical performance of the structure [5, 12]. Therefore, the identification and evaluation of lightning strike damage is crucial in maintaining the life of the structure.

Direct lightning strike damage is typically identified and assessed by visual or ultrasonic inspection (UT) non-destructive evaluation (NDE) techniques [5, 13,14,15]. However, the results are usually presented as a projected damaged area over the surface and do not provide the severity of the damage over the area. Furthermore, UT inspection also requires a couplant (typically water) and for high-resolution scans it can take hours to scan a relatively small area (< 2,000 mm2). Additionally, UT inspection equipment is challenging to install on turbines in operation. An attractive alternative to UT is pulsed thermography (PT), particularly as it takes minutes and does not require a couplant or contact. PT is a NDE technique based on the use of infrared imaging to detect subsurface damage [16]. The infrared (IR) images show thermal decay from a pulsed energy source as it passes through a component or sample. Methods developed to detect subsurface damage for thermography are utilizing the temperature maps obtained from an IR image series captured from a thermal camera.

Thermography techniques are typically categorized as either active or passive, characterized by their operational principles and applications. Passive thermography relies on the inherent thermal radiation emitted by the object itself, without any external stimulus. This methodology leverages variations in natural temperature gradients to elucidate thermal patterns indicative of hidden flaws or irregularities and is often suitable for evaluation of materials and structures in diverse settings [17,18,19]. In contrast, active thermography involves the deliberate introduction of an external energy source, typically in the form of a heat pulse or radiation, to induce thermal contrasts in the object under examination. Subsequent monitoring of the surface thermal response allows for the detection and analysis of structural anomalies or material defects, such as described in [20,21,22]. Active thermography offers enhanced control over the testing conditions, using targeted heat excitation to probe further into materials and structures.

Postprocessing of the thermal data is required to detect damage. The most widely used techniques include Absolute Thermal Contrast (ATC) [23, 24], Thermographic Signal Reconstruction (TSR) [25], Pulse Phase Thermography (PPT) [26,27,28,29] and Principal Component Analysis (PCA) [30, 31]. Processing of the image series using fast Fourier transforms (FFTs) converts the image time series to the frequency domain and enables the phase of the thermal response to be obtained; typically known as pulse phase thermography. These data processing techniques correct for the artefacts in the images caused by non-uniform heating, anisotropic diffusivities, and emissivity variations [32, 33]. A technique that has no need to smooth the data has not been established and most well-established thermography techniques use temperature contrast to determine the damage area and depth, but do not make an estimation of the damage severity.

Thermography has been used to identify typical defects in WT blades such as cracks, delaminations, and impacts [34,35,36,37,38]. The thermography inspection of WT blades typically take thermographic images from the ground and several companies have implemented standalone systems. Other researchers have used in-service aerial systems with thermal cameras [34, 36] for higher resolution inspection than those obtained from the ground. Recently a new type of thermal camera has become available that reduces camera costs significantly, is in miniature format, but can detect damage similar to the more expensive thermography systems [39]. This advance allows for a lightweight IR system to be deployed on an unmanned vehicle such as a drone, or could be safely manually handled by a technician inspecting the blade, thus providing routes for thermographic inspections for onshore or offshore WT blades. Although paint coating might obscuring damage [36], lightning strike penetrates and removes the paint enabling an effective inspection.

The aim of the work described in the paper is to explore pulse thermography as a tool for identifying lightning strike damage in CFRP panels. In addition, a novel approach is introduced to analyse the thermal images obtained from pulse thermography that provides a new means of categorizing, separating, and quantifying damage. The novel procedure is validated using X-ray CT scans of the damage, and is referred to as the “restricted heat detection method” throughout the paper.

Materials

Four CFRP laminate specimens were manufactured using a carbon/vinyl ester material system with PX-35 fibres from Zoltek Technologies. A 1.6 m unidirectional pultruded panel was manufactured and cut into specimens with dimensions 400 mm long x 200 mm wide x 5 mm thick. The simulated lightning strike experiments were conducted at Polytech A/S (Herning, Denmark). Two specimens were damaged with impulse current waveforms (Fig. 1a,1b) and two specimens were damaged with direct current (DC) (Fig. 1cd), both widely used to represent simulated lightning strike by the WT industry [4,5,6,7,8, 14, 40,41,42,43,44,45,46]. Full details of the experimental setup are described in [7]. For clarity, the impulse waveforms were defined by four characteristic parameters; the peak current value (\( {I}_{peak}\)), the rise time to reach the peak current (\( {t}_{1}\)), and the time at which the current decays to half of the peak current (\( {t}_{2}\)), also known as the half time, and the time at which the strike was extinguished (\( {t}_{f}\)). The DC current is defined by the peak current value (\( {I}_{peak}\)), and the duration (\( {t}_{DC}\)). These parameters are shown in Fig. 1a and b. The charge, \( Q\), and the specific energy or action integral, \( AI\), are used to compare different waveforms and are calculated from \( t=0\) to \( t={t}_{f}\) as follows:

$$ Q={\int }_{0}^{{t}_{f}}i\left(t\right) dt$$
(1)

and

$$ AI={\int }_{0}^{{t}_{f}}{i}^{2}\left(t\right) dt$$
(2)

The lightning strike waveform characteristics (peak current, charge, and specific energy) of the damaged panels are presented in Table 1. As can be seen in Fig. 1c-f, the two different types of lightning strike resulted in large differences in spread and depth of the damages. This difference allowed for the method to be tested for accuracy. Full details of the experimental setup used to achieve the waveforms have been presented in [7].

Fig. 1
figure 1

Lightning strike waveform representations with parameters for (a) impulse waveform characterization, and (b) DC waveform characterization. Demonstration CFRP panels damaged by simulated lightning strike waveforms: (c) Imp-1, (d) Imp-2, (e) DC-1, and (f) DC-2

Full size image
Table 1 Simulated lightning damage parameters including peak current, waveform characteristics, charge & specific energy [7]
Full size table

Pulsed Thermography

Thermography is a non-contact method in which a thermal camera is used to detect damage in a specimen by measuring the change in surface temperature. Pulse Thermography (PT) [47] is an active thermography technique, which uses a heat source to supply a thermal load to a component to produce a rapid excitation and consequent decay. The heat source is usually a flash or heat lamp. The thermal camera captures images prior to heat input, and readings are captured throughout the heating and cooling process of the component/sample. Defects arrest heat conduction because damaged material volumes generally have a different thermal conductivity than the undamaged material . In ‘reflection thermography’ the heat source is on the same side of the component as the thermal camera is viewing. In this case, the surface temperature of the component increases in the vicinity of the damage and allows subsurface defects to be identified.

The arrangement used for the reflection thermography inspection used in the present work is shown in Fig. 2. The arrangement has 4 components: (1) IR camera, (2) flash/heat source, (3) test specimen/sample, and (4) computer. The heat source was a Bowen studio flash and the thermal camera used was a FLIR silver 480 m. The specification of the equipment used are presented in Table 2 along with the recording frame rates and durations.

The extent of the damage in the panels can be estimated and categorised to fibre and/or matrix damage visually but requires individual interpretation by an operator that can be subjective. This paper presents a reliable thermographic approach to identify the extent and categorise damage types without the need for human interpretation. This would enable rapid on-site deployment following a lightning strike to assess if a repair was required.

Fig. 2
figure 2

Experimental setup of pulse thermography in reflection mode

Full size image
Table 2 Thermography equipment specification
Full size table

Restricted Heat Detection Method

Many techniques have been developed to interpret the data, e.g. pulse phase thermography, subtraction, contrast, slope, thermal signal reconstruction, and dynamic thermal tomography [25, 48,49,50]. Herein, an alternative approach is devised to analyse the thermographic images and provide a quantitative metric related to the severity of damage induced by the lightning strike, which introduces a new post processing procedure. As mentioned above, in thermography, damage can be identified by monitoring the heat diffusion through the specimen. The proposed post processing procedure exploits this feature by taking the integral of the temperature recorded by each pixel in the recorded thermal image series with respect to time. Thereby the integration makes use of all the thermal data captured during the test, which increases the signal to noise ratio. The integral is then subtracted from an integral obtained from the temperature image series associated with a reference pixel located in a region that is assumed to be undamaged. The approach is termed the ‘restricted heat detection method’ and is carried out in three steps as shown in Fig. 3:

1. An image series is captured by the IR camera over a time period that is long enough to fully capture the thermal decay. The temperature data from each pixel in the image series is assembled into a 3D matrix as shown in Fig. 3a,

2. Any bias such as the reflection of the cooled IR detector is removed by subtracting the first image from the rest of the data set, as shown in Fig. 3b, and.

3. The temperature data is numerically integrated with respect to time from the initial flash (\( {t}_{0}\)) until the temperature reading returns to ambient (\( {t}_{f}\)) using the trapezoidal rule, implemented in Matlab [51]. The integrated area is indicated by the grey colour shown in Fig. 3c. The units for restricted heat are derived from the integration of the temperature temporal data and therefore are expressed in units of °C * seconds (i.e. dT * dt), which is notated as °Cs throughout the remainder of the paper.

The integration is executed from the time of the heat input, provided by a camera flash, until the specimen returns to ambient temperature. As shown in Fig. 3c, the integral for the reference pixel is subtracted from all the other pixel integrals. The result also shown in Fig. 3d, is a spatial array that includes the temporal information from the image series that provides a clear indication of the extent of the damage, with the restricted heat scale indicating the severity of the damage. Please note that a reference point does not need to be selected. If this reference value is assumed to be zero, the result will remain a heat energy emitted to the thermal camera, and can still be used to identify damage. However, subtracting a non-zero reference enables determination of additional heat emission due to a defect.

Fig. 3
figure 3

(a) Organization of thermal images stacked as 3D matrix for analysis in Matlab, (b) thermal images of simulated lightning strike damaged specimen at different time steps after the heat input with subtraction of the first image, (c) typical temperature profile from a thermography experiment where the dashed black line is the reference non-damaged inspected area and the solid black line is the damaged region, and (d) is the processed results in the form of restricted heat. The grey area in the temperature profile shows the integration obtained from the thermography analysis

Full size image

It is proposed that the integral of temperature with respect to time provides a metric related to heat transferred through the sample. In the areas where there is no damage, a reference value is obtained related to the heat transfer over the time it takes for the material to return to ambient conditions. Whereas in the defect (or damaged) area, the damage prevents the heat diffusion through the sample and instead the heat is diffused around the defect slowly causing an increase in temperature on the surface of the component. Subtracting the integral with respect to time obtained from the defect region from that of the reference region, provides a metric termed the “restricted heat”, which indicates the quantity of heat that was prevented from diffusing through the thickness of the sample. Hence, a larger value of restricted heat indicates a higher level of damage severity.

Thermal Non-Uniformity Correction

On large composite panels the heat input from the camera flash will result in a non-uniform temperature distribution across the surface of the panel. This non-uniformity can obscure the temperature differences between the non-defective and damaged regions [25]. Figure 4a shows the effect of the non-uniformity on the restricted heat. The damaged region is identified by a large value of restricted heat. In Fig. 4a, the threshold has been set above the damage level. Then, the pixel values above that threshold are ascribed as Not a Number (NaN) in MATLAB, allowing a visualisation of the bias the heat input. Figure 4b shows a surface fitting of all the data when the damaged region ascribed the NaN is removed. The initial full data field of restricted heat (i.e. without application of the NaNs) is then subtracted from the non-uniform heat and the result is shown in Fig. 4c. The NaNs are reinstated to show that the data in the non-defective region returns a practically zero restricted heat value. This non-uniformity correction was applied to all specimens to enhance the data interpretation.

Fig. 4
figure 4

(a) Results of the integration method with damaged pixel values removed, (b) surface fit of the observed reference values, and (c) results of the integration that has been corrected with damage removed

Full size image

Damage Severity Categorization

The plot of the damage severity obtained for specimen DC-2 from the integration method is shown in Fig. 5a, which displays typical characteristic features as obtained for the other specimens. The extent of the area of damage from lightning is clearly visible, and the severity of the damage is indicated by the large numerical values of the damage metric obtained from the heat restricted detection method. Charred regions have a lower damage severity than damage at the lightning attachment point. The integration method was used to quantify similar types of damage as shown for specimen DC-2 (Fig. 5a), similar to [52].

To identify similar damage types, the values of the characteristic threshold pixel were selected. These threshold values were set based on visual inspection and evaluated as the values that were at least 3 standard deviations away from the reference integrated values of the non-damaged regions. The values at each pixel were compared with the selected threshold values and binarized by assigning a value of 1 if the value was above the threshold and 0 if the value was below the threshold. Each different test setup requires slightly different threshold values depending on the amount of heat input supplied. The heat input is affected by factors such as the flash lamp being located at different distances from the inspected test sample, power supplied by heating source, and exposure time of heating source. For the specimens, a threshold for ‘matrix damage’ was set to one standard deviation and the results are shown in Fig. 5b. With the image binarized, a count of the number of pixels above the threshold values can be used as the total pixel area of the damage. The count is then multiplied by the spatial resolution squared to determine the physical damaged area.

To validate the proposed Restricted Heat Detection Method, X-ray CT was conducted on all the specimens using a 225 kV/450 kV Nikon/Metris X-ray scanning machine at the University of Southampton’s µ-VIS X-ray Imaging Centre. A 225 kV X-ray source was used with a 2000 × 2000 sensor to capture the X-ray images. The resolution of the scans for the damaged specimens was 0.1088 mm/voxel. The area of the damage identified by the X-ray CT was then compared with those from the categorisation procedure.

To identify the damaged area, each slice was inspected for damage and segmented to quantify the damage in each slice. Afterwards, the damage from all the slices was projected to a plane representing the top surface of the panel. The damaged area was then identified using the number of pixels that were segmented as damage.

Fig. 5
figure 5

Thermography results for specimen DC-2 where (a) is the nominal integration method measurement, (b) is the different categories based on standard deviations

Full size image

Validation of Damage Categorisation Procedure

Figure 6 compares results obtained using the Restricted Heat Detection Method (Fig. 6a) and the damage categorisation procedure (Fig. 6b) with X-ray CT images (Fig. 6c). Figure 6a clearly indicates the areas of damage highlighted by the higher restricted heat values. The higher values of restricted heat are associated with fibre and resin damage, and the lower values are associated with undamaged regions. The clear distinction between fiber and resin damage shows the procedure can decouple the different damages. Figure 6b shows the damage categorisation based on the standard deviation in the restricted heat values. Using the proposed thresholding approach identifies regions that can be categorised into fibre breakage or only matrix damage. Figure 6c shows the CT images of the damage specimens overlayed with the extents of matrix and fibre damage identified by the categorisation procedure. The CT images capture only the fibre damage, which corresponds well with that identified by the Restricted Heat Detection Method. Although the CT does not identify the matrix damage, the visual inspection (Fig. 1) indicates the extent of the matrix damage which correspond well with that identified by the new procedure.

Fig. 6
figure 6

Results of (a) integration method performed on pulse thermography, (b) damage categorization from integration method, and (c) comparison of CT of lightning damage to integration method for (left-right) Imp-1, Imp-2, DC-1, and DC-2

Full size image

Figure 7 provides a quantitative comparison of the fibre damage estimates obtained from the CT images, visual inspection, and the Restricted Heat Detection Method (RHDM). It is observed that there is an excellent correspondence between the CT and the restricted Heat Detection Method, with the visual inspection showing an overestimation of the damage. The range of differences between the RHDM and CT observation is 10–14% and the relative to visual inspection 74–239%.

Fig. 7
figure 7

Comparison of visual, CT, and PT data

Full size image

Figure 8 shows the signal to noise ratio (SNR) along the line shown in Fig. 6a that intersects with the restricted heat value. The noise in the SNR was calculated using the average data from the undamaged section. Since the damage was variable, the signal varied but the noise level is constant at ~ 1 as shown by the lowest values in the plot in Fig. 8. Therefore, Fig. 8 clearly indicates a sharp transition between signal from the damaged region and the noise without any noticeable lateral heat diffusion from the damaged to the undamaged regions. The maximum SNRs are presented in Table 3, which shows the maximum values with the largest damage depths provided the highest SNR and the specimens with the lowest damage depths have less signal compared to the noise.

Fig. 8
figure 8

Signal to noise ratio of vertical slices with the maximum signal to noise of each experiment

Full size image
Table 3 Maximum signal to noise ratios for each experiment
Full size table

Discussion

It has been shown that the proposed Restricted Heat Detection Method based on integrating the thermal response following the heat excitation was able to identify damage on in CFRP panels exposed to simulated lightning strike. It has also been demonstrated that the approach can be used to quantify damage severity by using the standard deviation of the restricted heat values as a metric to identify resin damage, between one and two standard deviations and fibre damage as greater than two standard deviations. This metric proved to be valid for the cases studied but may need to be refined for different damage types, different material systems, and experimental setups. However, it has been shown that using the standard deviation as a metric is an effective means of identifying different regions in the process thermal data and hence has the ability to characterise the damage types.

The simulated lightning strike damage investigated in this paper is representative for real direct lightning strike damage reported elsewhere [4,5,6,7,8, 14, 40,41,42,43,44,45,46], and identified using techniques such as ultrasound. The advantage of the restricted heat methodology is that it can rapidly identify damage in a non-contact manner over relatively large regions of structure. The approach provides a pathway for developing a lightweight system to be implemented on e.g. an unmanned vehicle such as a drone to perform in-service inspections for onshore or offshore WT blades. Further, as the infra-red cameras are small it also enable manual inspections by a technician. An inspection protocol could be developed based on laboratory trials, such as those described in the present paper, which could then be used to characterise damage in-situ on WT blades, thus aiding the decision process for repair or to remove the blade from service.

Conclusions

A new NDE methodology has been proposed based on the application of pulsed thermography. The novelty is in the processing of the thermal data to provide a new metric referred to as the restricted heat, which integrates the thermal response following the pulse heat excitation. It has been shown that a segmentation and categorisation procedure can be devised and applied to characterise different damage types. A simple procedure was devised to account for non-uniform thermal pulses, which enabled a consistent value of the restricted heat metric to be determined. The new approach has been demonstrated on CFRP panels damaged by lightning but could also be applied to impact damage cause by e.g. hail or bird strike. The technique does not rely on optical alignment and is insensitive to vibrations as rugged low-cost bolometers that can be mounted on a variety of platforms and used for in service inspections. It has been demonstrated that PT can be used to identify lightning damage. The damage area was enhanced with damage severity by adopting an integration procedure capturing the cumulative heat that was restricted through the sample to provide additional insight into the damage state. This provides a quantitative data set that allows for different damage types to be segmented and categorised, which is a marked improvement over other thermography analysis techniques.