An Experimental Comparison of LED and Eddy Current Pulse-Compression Thermography on an Impact Damage CFRP Benchmark Sample

A comparison of the application of Pulse-Compression to eddy-current and LED thermography is presented. Results achieved on a dedicated sample with an impact damage are reported and discussed. The same pulse compression procedure based on the use of Barker codes has been applied to both techniques and it proved to be work properly in both cases. Two different features have been extracted and imaged from the retrieved time-responses after pulse-compression.

AT always relies on the application of an external heating stimulus over the SUT to reach the desired thermal contrast [10]. In most of the applications, the heating stimulus is realized with some light source such as flashlamps, quartz lamps linear systems halogen lamps, laser heating and LED chips [11][12][13][14][15][16][17][18]. In all these cases, the heat is firstly transferred to the illuminated sample surface and then it diffuses inside the sample [19]. In addition to lightbased AT, alternative methods of thermal excitation have been proposed and are currently applied, which are based on phenomena used in other NDT techniques: Ultrasound Vibro-Thermography [20], Eddy-current Pulsed Thermography (ECPT) [21], Microwave Thermography [22]. With respect to the light-based techniques, in these cases the heat can be provided directly to the whole sample volume (bulk heating) and, in some cases, the heat can be selectively generated only in presence of an inhomogeneity or of a defect [23][24].
Despite the employed heating source, the classification of the existing AT schemes is strictly related to the timefrequency characteristic of the heating modulation signal. In Pulsed Thermography (PT), a short time duration heating stimulus that well-approximates the Dirac's delta excites the SUT. Features of interest can be extracted by analysing both the heating and the cooling trend of the recorded impulse response pixelwise. It is worth to note that the maximum achievable Signal-to-Noise Ratio (SNR) in PT is set by the excitation source power. Improved SNR values can be obtained by modulating the heating stimulus with a sinusoid at given frequency, i.e. Lock-in Thermography (LT) at the cost of less information with respect to the one obtainable by using PT [25][26]. In the recent years, efforts have been made to merge effectiveness of PT and the SNR values achievable via LT, leading to Pulsed Phase Thermography and Multi-Frequency Lock-in Thermography. To this aim, the use of a coded signal to modulate the heating stimulus together with Pulse Compression (PuC) to retrieve the sample impulse response has been also investigated [27][28][29][30][31][32]. In Pulse Compression Thermography (PuCT), the heating emission is commonly modulated either by a phasemodulated signal, e.g. Barker code, or by a frequencymodulated "Chirp" signal. These coded signals have a unique characteristic: their duration T and bandwidth B are uncorrelated. Thus, the signal frequency content can be designed to suit the investigation of a given sample, while its duration can be enhanced almost arbitrarily to reach the desired SNR [33]. Although PuCT is now a quite established technique, very few attempts are reported in literature to extend this approach to Eddy-Current Thermography [34]. To further investigate this possibility, in this paper experiments were carried out both with Eddy-Current Pulse Compression Thermography (ECPuCT) and with LED-stimulated PuCT on an impact damaged Carbon Fibre Reinforced Plastic (CFRP) benchmark sample described in detail in Section 3.3. The results are then compared to evaluate the effectiveness and the feasibility of Eddy-Current Pulse Compression Thermography (ECPuCT) with respect to LED-stimulated PuCT. Evalua-

Proceedings of FENDT 2018
978-1-5386-6230-4/18/$31.00 ©2018 IEEE tion of the impact damage on the composite materials using ECPT have been reported in literature [35][36][37], and the use of various feature extraction techniques for automatic detection and classification has been also reported [38]. Here the main goal is to apply to ECPT the Pulse-Compression procedure described in [18] and analyse the performances achieved compared by those provided by LED-PuCT. To do this, Barker code [39] signal was employed to drive the two different heating systems: an induction coil (ECPT) and LED chips. Although the physics underlying the two exploited approaches is different, i.e. heat dissipation by Joule's effect and volumetric heating for the Eddy Current case with respect to the ideallyconsidered 1D diffusion in the LED case, it is shown that PuC is a robust strategy that could be exploited in both cases. In addition, the suitability of the non-linear fitting approach for subtracting the step heating contribution proposed for LED PuCT in [18] was successfully verified for the proposed ECPuCT scheme.

Introduction to Pulse Compression
The PuC algorithm has been exploited on two different setups, which are described in detail in Section 3. To gain insight on how the PuC can be successfully implemented for the two setups, a brief introduction about the mathematical theory of the PuC technique is given here. The reader is suggested to refer for example to [33] for further details. PuC is a powerful measurement technique widely used to estimate the impulse response of a Linear-Time Invariant (LTI) system in poor SNR conditions. Given a pair of coded signals, the excitation that modulates the heating source, and the so-called matched filter , such that their convolution (denoted by ) approximates the Dirac's Delta function , , an estimate of can be obtained by exciting the LTI system with and then by convolving the system output with . This is demonstrated in (1) and in Fig. 6 for a single pixel of the acquired thermogram. The presence of an Additive White Gaussian Noise (AWGN) as the term, uncorrelated to , is considered. Note that the SUT is assumed to be an LTI system.
The main pros of PuC is directly related to the peculiar characteristics of the coded signal used, which has uncorrelated T and B. It can be demonstrated that the SNR gain achieved by using PuC with respect to a pulsed excitation of the same peak power is ~T×B. Consequently, the higher is the T×B product, the closer is the approximation of the Dirac's Delta function: the amplitude of the sidelobes of , which affect , decreases as T×B increases. In this work, the employed coded signal for mod-ulating the heat sources was a Barker Code (BC) [39][40] with a bit length of 13, that in turn assures sidelobe level suppression of 22.3 dB. The BC signal used is plotted versus time in Fig. 5(a), in which each original bit lasts one second at the chosen frame rates used for the ECPuCT and the PuCT setups, as described in Sections 3.1 and 3.2. Note, that also the SUT cooling trend was acquired for additional thirty seconds after the ending of the BC. Although the physics underlying the heat diffusion on the two setups is different, the chosen padding duration onsets thermal waves with increased thermal diffusion wavelength, thus assuring enough sensitivity at deeper depth within the inspected SUT [10,19,24].

Experimental Setup
In this Section, the experimental setup and exploited signals are shown and described for both methods.

LED PuCT
The setup used in this experiment was the same introduced in [6,18,41]. A 'National Instrument PCI-6711 Arbitrary Waveform Generator (AWG) board and a National Instrument 1433 Camera Link Frame Grabber were connected to a PC, where an ad-hoc developed virtual instrument (VI) in LabVIEW managed the signal generation and acquisition. The AWG board provided both the wanted excitation signal, i.e. Barker code, and a reference

Proceedings of FENDT 2018
clock signal (CLK) for triggering the IR camera acquisition. The camera used was a 'Xenics Onca' equipped with an InSb cooled detector with the spectral range of 3.6 to 4.9 m. Camera offers a resolution of 320 × 256 pixels and NETD < 17mK. The coded signal was input into a 'TDK Lambda GEN 750W' power supply that fed eight LED chips placed at about 30 cm from the SUT. The LED chips are capable to provide a maximum overall power of 400W. The thermograms were acquired at the frame rate of 40 Hz. A block diagram of the experimental setup used is depicted in Fig. 1, whilst Fig. 2 shows a picture of the SUT and LED setup.

Eddy Current PuCT
The ECPT configuration used in this experiment is shown in Fig. 3. It mainly consists of 5 units, i.e. a heating module, an induction coil, a signal generator, an infrared (IR) camera and a PC. In this work, the heating module was Easyheat 224 from Cheltenham Induction Heating with the working frequency of 260 kHz and RMS current of 300 A; one side of a rectangle coil was used; the signal generator was used for synchronously triggering the heating module and IR camera; the type of IR camera was the FLIR SC655 equipped with an uncooled microbolometer detector array with the resolution of 640 × 480 pixels, the spectral range of 7.5 -14.0 m and NETD < 30 mK. The IR camera records the surface temperature distribution of the 13 seconds excitation period as well as additional 30 seconds cooling period at 50 FPS.

Sample Under Test
The SUT (Fig. 4) was a quasi-isotropic CFRP plates made by Tencate (Almelo, The Netherlands), having 150 × 100 × 4.2 mm 3 , containing 12 layers of five harness satin carbon fibers woven with balanced woven fabric. The matrix was Polyphenylene Sulfide (PPS), a thermoplastic polymer consisting of aromatic rings linked with sulfide moieties, resistant to chemical and thermal attack. The carbon fibers are T300JB type and their volume is 0.5 ± 0.03 and the density is 1460 kg/m 3 . The specimen was subjected to an impact of 8 J of energy. The impact was made by using FRACTOVIS PLUS 9350-CEAST instrument (Instron, Norwood, MA, USA) with a hemispherical bumper head having 20 mm diameter and a mass equal to 2.045 kg. The sample exhibits electric properties that depend on the type of carbon fibers and on their volume fraction in the material. The transverse electric conductivity value was in the range [10-100] S/m, whilst the longitudinal conductivity varies between 5 × 10 3 S/m and 5 × 10 4 S/m.

Experimental results
As both the heating sources are monopolar, i.e. LED chips and induction coil, a step-heating contribution must be removed from the acquired thermograms to successfully implement the PuC algorithm, as for (1). To gain insight on this, Fig. 5(a) shows the BC modulation signal, whilst the colored line plot of figures 5(b) and 5(c) depict the corresponding acquired raw signals from a single pixel on the SUT, for the PuCT and ECPuCT re-   [18] proposed an optimized strategy based on a non-linear polynomial fitting for successfully subtracting the step-heating contribution in PuCT. This strategy has been applied here also on the ECPuCT data and the resulting fitting polynomial is shown as a black line plot in Fig. 5(c). It can be appreciated from Fig. 5(d) that the proposed approach allows the step-heating contribution to be faithfully removed from the raw data, thus the PuC algorithm being effectively carried out pixelwise as for (1). Fig. 6 depicts the s obtained from a single pixel onto the impact damaged area of the SUT for PuCT (red line plot) and ECPuCT (green line plot). It can be noted that the s are wellretrieved for both cases. In fact, the s follows an exponential decay trend as time elapses (neglecting fluctuation due to sidelobes), thus in a good agreement with the one expected from a pulsed excitation.
Thermograms of the sample are also shown, obtained by imaging the pixelwise amplitude of (i) the retrieved after performing PuC (as for (1)) and of (ii) the Time-Phase algorithm output [29][30][31]. The last one relies on the application of the Hilbert Transform (HT) over before performing the PuC algorithm, as schematically reported in Fig. 7. Results obtained as for (i) will be hereinafter referred as "Time", whilst results from (ii) as "Time-Phase". Fig. 8 shows a zoom view of the obtained thermograms with "Time" and "Time-Phase" algorithms for ECPuCT and PuCT. The damaged portion is visible as a bright pixel area on the images. Please note that the depicted thermograms refers to a time instant equal to one second on the . Finally, figure 9 reports for "Time" the images retrieved after PuC at non-equispaced time instant corresponding to linear spaced diffusion depth. For surface characteristics disappear and only the signature of the impact damage can be seen with maximum contrast. This can be positively exploited to improve defect detection and classification

Conclusions and Future Work
Comparison of the experimental results shows that the PuC process can be successfully applied on the ECPT and that the proposed ECPuCT scheme is capable to detect defects. This benchmark study provides the basis for the future work on the ECPuCT along with more theoretical analysis and exploiting the feature extraction algorithms and quantitative merit figures like SNR.

Acknowledgment
This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sk odowska-Curie grant agreement No 722134 -NDTonAIR.