Aeronautics composite material inspection with a terahertz time-domain spectroscopy system

Abstract. The usability of pulsed broadband terahertz radiation for the inspection of composite materials from the aeronautics industry is investigated, with the goal of developing a mobile time-domain spectroscopy system that operates in reflection geometry. A wide range of samples based on glass and carbon fiber reinforced plastics with various types of defects is examined using an imaging system; the results are evaluated both in time and frequency domain. The conductivity of carbon fibers prevents penetration of the respective samples but also allows analysis of coatings from the reflected THz pulses. Glass fiber composites are, in principle, transparent for THz radiation, but commonly with significant absorption for wavelengths >1  THz. Depending on depth, matrix material, and size, defects like foreign material inserts, delaminations, or moisture contamination can be visualized. If a defect is not too deep in the sample, its location can be correctly identified from the delay between partial reflections at the surface and the defect itself.

Aeronautics composite material inspection with a terahertz time-domain spectroscopy system 1 Introduction The desire to simultaneously fulfill the seemingly contradicting requirements of reduced weight and enhanced stability drives the development of sophisticated, complex, and heterogeneous composite materials. The aircraft industry, both civilian and military, naturally is a key player in this field. The advancement of materials calls for correspondingly advanced methods and technologies for inspection and failure detection during production or maintenance, especially in the area of nondestructive testing (NDT). A wide range of NDT techniques has been established and matured; however, a silver bullet method that can identify all kinds of defects and failure modes with equally high precision does not yet exist. THz spectroscopy, imaging, and tomography, 1-8 on the other hand, are becoming increasingly attractive tools for NDT purposes, since THz radiation penetrates most nonmetallic and nonpolarized media, can be applied contact-free, and does not require special safety precautions like radiographic imaging. THz NDT on aeronautics materials first came to the limelight with the inspection of space shuttle isolation foam after the 2003 Columbia accident 9 and has since been applied to various NDT tasks on aeronautics materials and coatings. This includes ceramics, 10 glass fiber composites 11,12 as well as surface roughness and corrosion of metal under coatings. 13,14 Conductive materials like metal or carbon fibers are excellent THz reflectors, which commonly limits the applicability of THz radiation to surface defects for these cases. Reflectance and transmittance in carbon fiber sheets, however, depend on the orientation of the fibers with respect to the polarization of the THz radiation: Not unlike in a wire-grid polarizer, THz radiation which is polarized perpendicularly to the fibers experiences a higher transmittance than for parallel polarization. The anisotropic THz characteristics of carbon fibers have been utilized to examine heat damage 15 as well as delamination 16 at or close to the surface. There are numerous further examples of THz methodologies and applications in NDT of composites. An extensive review is given in Ref. 17.
The project DOTNAC (Development and Optimization of THz NDT on Aeronautics Composite Multilayered Structures; http://www.dotnac-project.eu) aims for the development of a fast, high resolution, noninvasive, and noncontact inspection tool for assessing aeronautic composite parts either during production or maintenance on the basis of THz radiation. It is carried out by a consortium of 10 partner institutions that include research organizations, NDT specialists, and aircraft manufacturers. The tool should be easy to integrate into industrial facilities and fill performance gaps present in existing NDT methods. It is therefore expected to be a complimentary technique with the potential to be more precise for certain kinds of defects and/or structures. The assessment of that potential and the performance of THz material inspection are essential parts of the project. All THz measurements will therefore be compared to reference data taken with established NDT techniques, such as ultrasound, radiographic testing, and infrared thermography, to deliver a THz feasibility study for the inspection of aeronautics composite materials.
DOTNAC takes a two-pronged approach to deliver a more comprehensive picture of the THz capabilities on aeronautics composite materials: both a frequency-modulated continuous wave (FMCW) and a pulsed time-domain spectroscopy (TDS) system are set up and used for the inspection of a broad range of typical samples. While both systems are designed to work as stand-alone devices, they possess portable sensors that can be operated remotely via a cable harness in a three-dimensional (3-D) scanner working in reflection mode, which is the third hardware component developed in the project. This scanner unit can alternately incorporate the FMCW or TDS sensor and capture data on a nonflat part, specifically a test radome (see Fig. 1). The functionality and performance of the FMCW tool are described elsewhere; 18 a brief overview is given in Sec. 3.1.
In the main part of the article, we will first give an overview of the various sample types that have been manufactured for the project. Next, we will explain the two THz-TDS systems from which data are presented here. Besides the mobile fiber-coupled DOTNAC TDS tool, we use a laboratory system installed at the TeraTec application center of Fraunhofer IPM as a reference system. These descriptions are followed by a presentation of our imaging results. An assessment of the potential of THz-TDS for aeronautics NDT and an outline of possible further developments conclude this study.

Samples
Two general categories of composite structures are available for investigation with pulsed THz radiation: glass fiber reinforced plastic (GFRP) and carbon fiber reinforced plastic (CFRP). While the relevance of CFRP for the construction of aircraft is ever growing, especially for high-performance components, its conductivity is an inevitable obstacle to penetration with THz radiation. In tests conducted by ourselves, the transmitted THz amplitude in the case of perpendicular polarization was 11 times higher than for parallel polarization on a single-carbon fiber sheet but still very low compared to the incident signal (1.2% of the incident amplitude was transmitted for perpendicular polarization compared to 0.11% for parallel). In a set of samples with impact-damaged CFRP panels with 5 mm thickness, for example, the deformation of the surface by the impactor could be visualized as a dent, but the critical crack formation inside the sample is invisible to THz inspection. On the other hand, the good THz reflectivity of CFRP enables the inspection of coating layers on CFRP substrates.
The GFRP samples are either solid laminates, consisting of several plies, or sandwich structures that incorporate one (A-sandwich) or two (C-sandwich) cores between GFRP skins. Two kinds of sandwich cores are used: Rohacell foam and a honeycomb structure. The samples further differentiate into samples with artificial, well-defined defects (socalled "calibration samples") and such with defects created through intentional faults in their fabrication process (test samples). The calibration samples include rectangular inserts serving as mock defects, water contaminations of different volumes and impact damages with various kinetic energies. The test samples include debonds and delaminations as well as a sample with enhanced porosity level. The misprocessed coatings on CFRP panels are also classified as test samples. Figure 2 summarizes the selection of samples.

FMCW Setup
For the sake of completeness, the key parameters of the complimentary, fully electronic FMCW system should be mentioned. This tool employs interchangeable transceiver heads whose frequency ramps are operated via a voltage-controlled oscillator, a frequency multiplier chain, and a Schottky diode mixer for heterodyne detection each. Three sensors with center frequencies of 100, 150, and 300 GHz and corresponding output powers of 2, 0.1, and 0.06 mW were used. Another high-frequency measurement head at 850 GHz (2 μW output power) was found not to have sufficient penetration capability on the samples for satisfactory imaging results. The sensors were operated both in focused and synthetic aperture geometry; furthermore, datasets from the 100 and 150 GHz heads could mathematically be fused together into an ultra-wideband frequency range to create a virtual sensor that combines the relatively high material transparency at low frequencies with enhanced depth resolution through a broadband frequency ramp.

TDS Reference Setup
The samples described above should be examined not only with the mobile TDS tool that will be presented in the next section, but also with a reference system with well-known specifications and proven performance. The aim for the reference measurements is the acquisition of high-quality THz images to analyze the general potential and limitations for composite materials' inspection. For this purpose, we used a stationary TDS raster scan imaging system with InAs surface emitter and LTG GaAs photoconductive antenna detector. Both are pumped in a free-space setup by a Ti:Sa laser at 800-nm center wavelength; the THz beam is guided in a dry air enclosure to suppress water absorption features in the data. The time delay between pump and probe pulse is varied with a shaker that oscillates at 10 Hz, so that effectively 20 data traces∕s are recorded, since the data from both half-cycles of the shaker motion are captured and one set afterward is mirrored in the time domain. The use of the shaker limits the available bandwidth to 2 THz, with the center frequency being at 500 GHz. The area raster scan algorithm moves the sample horizontally at 1 mm∕s velocity through the THz focus at perpendicular incidence and takes data in on-the-fly mode. At the end of a line scan, the sample is moved by a vertical increment of 0.1 mm and the next line scan, in reverse direction, is started. Oversampling is done both in horizontal and vertical directions. This results in high-resolution images and requires 30 h for a 100 × 100 mm 2 area at the reference settings. The setup can be used in reflection or transmission geometry; if not stated otherwise, the data shown in this article have been taken in reflection geometry (see schematic in Fig. 3) to match the mode of operation of the 3-D scanner demonstrator, which was built in the DOTNAC project.
The datasets taken in this way contain a time-domain trace of the electric field for every pixel, so that they are effectively four-dimensional (x, y, time, amplitude). After applying a Fourier transform, the corresponding frequency information is available. The images were manually evaluated in the time and frequency domain by browsing through the datasets along the time or frequency axis, comparable to a flip-book. Additionally, a "static" image, which displays the time-domain peak-to-peak (p2p) amplitude value for each pixel, was examined. Examples for the evaluation modes are given in Fig. 4.
The reflection peak from the sample surface is the dominating feature for most of our samples. As a consequence, the p2p images mainly probe that surface, i.e., only if a defect in the sample has an effect on the surface like a bulge that reduces reflection at its slopes, it can be visualized with the p2p mode.
The delay evaluation naturally is suited to discover defects hidden in the volume of the sample by identifying local discontinuities in the dielectric behavior that cause in-or decreased reflection and/or absorption, respectively, which manifest themselves as trailing peaks at various time delays. Depth information about features that are generated within a sample, e.g., partial reflection at a discontinuity, can be calculated from the time-of-flight difference between the surface reflection and the feature of interest. This requires information about the THz dielectric function of the sample or at least the average refractive index between ∼0.1 and 1 THz. The lower end of this range is given by the system characteristics, the upper by the materials under test themselves. Preliminary examinations gave a high THz opacity beyond 1 THz for virtually all materials that have been tested in DOTNAC, which restricts the penetrability of the samples with broadband radiation. This also limits the detectability of defects that are embedded deep in a sample.
The frequency spectrum of the main surface reflection pulse is modulated by trailing pulses which are generated at defects in the sample volume and can therefore also be used to discriminate between pristine material and embedded defects. However, small trailing pulses from interfaces in the sample will cause correspondingly small oscillations in the spectrum. Additionally, interference effects might manifest themselves in the frequency images and block spectral windows from analysis. Laminate samples contain by definition multiple internal interfaces between their constituting plies and hence exhibit a multitude of reflections which complicate defect identification.  Fig. 2 Overview of the sample structures and defect types available in DOTNAC.

DOTNAC TDS Setup
The concept for the DOTNAC project TDS system follows several key specifications, which arise from its purpose as a demonstrator for an industrial grade system that should be suitable for usage on the shop-floor and could be integrated into testing facilities. As such, robustness, flexibility, and speed are issues to be addressed. It is desirable to construct a rigid system that requires a minimum of readjustment after long-distance transport by employing as many fiber optical (in contrast to free-space) components as possible. This coincides with the need to provide remote sensors that can be operated via electrical cables and fiber optical patchcords over a distance of at least 5 m from a base unit, to allow for the motion of the sensors on the five axes (X, Y, Z, θ, ψ) 3-D THz scanner. Finally, the system should be able to scan a time-domain trace of at least 100 ps in at most 10 ms, that is to say at a data acquisition rate of 100 Hz.
These parameters differ significantly from the design goals of the reference system, which provides THz imaging with high quality at the expense of being complex, immobile, and limited in data acquisition speed. As an adequate response to the DOTNAC project requirements, a fiber-integrated electronically controlled optical sampling (ECOPS) setup was chosen. It consists of two nominally identical pulsed fiber lasers with 1560-nm center wavelength and 74.5-MHz repetition rate, which emit pulses of 100 fs duration and 70 mW average power at the end of a 6-m-long fiber delivery (customized versions of the FemtoFErb model by TOPTICA Photonics, 82166 Graefelfing, Germany). A phase-locking electronics module allows synchronizing the cycle of the slave with that of the master laser so that their pulses are emitted with a fixed time delay. Alternatively, the delay can be varied by applying a voltage to a phase shifter input, which modifies the repetition rate of the slave laser until a time delay has built up that corresponds to the applied voltage. 19 We chose a sinusoidal waveform, effectively generating an analogous time delay variation to that caused by an optomechanical shaker. Therefore, for each cycle of the modulation voltage, two THz traces can be recorded, one on the half-wave with a positive slope and one on that with a negative slope. Figure 5 shows a sketch of the main system components.
The ECOPS operating principle is very similar to the well-known asynchronous optical sampling (ASOPS) technique; the principal difference is that ASOPS works with a fixed repetition rate detuning of the slave laser, whereas ECOPS employs a variable one. In ASOPS, the time delay between the laser pulses thus grows in each cycle by  the increment determined through the repetition rate difference so that the pulses overlap in time again after the total delay has added up to the inverse of the master laser repetition rate. The time window that is captured, hence, corresponds to this. For common mode-locked fiber lasers, typical repetition rates are on the order of 100 MHz, which creates datasets of 10 ns length. The information of a THz trace, however, is in most cases contained in ∼100 ps so that up to 99% of an ASOPS dataset are junk data that do not include relevant information. The variable time delay of ECOPS allows one to limit the captured THz traces to meaningful information and thus enables a more efficient data acquisition. Note that these statements are made for a setting where the distance from sensor to target, and in this way the zero delay offset, is kept virtually constant. This is realized in our measurements and would also apply to other arrangements that utilize focusing optics and require the targets to stay within a Rayleigh length of a few millimeters. For a stand-off scenario, possibly with moving targets, the wide time delay window of ASOPS would be advantageous or even necessary.
Our sensor consists of two fiber-coupled measurement heads that include photoconductive antennas as emitter and detector of THz radiation and TPX lenses for beam shaping. The measurement heads are mounted under an angle of 20 deg on a baseplate which can be attached to the 3-D    scanner unit, alternatively to an FMCW transceiver head (see Fig. 6). The focal distance of the THz optics is 200 mm to match the working distance of the scanner. The focusing lenses can easily be dismounted to allow the use of other optics, for example, with sharper focusing for enhanced spatial resolution. Also mounted on the baseplate is a trans-impedance amplifier for signal preprocessing. With the maximum specified phase offset modulation voltage, our ECOPS system can scan a time window of 146 ps at 100-Hz modulation frequency. This range is calibrated with a nonlinear frequency-doubling crystal on which the beams of master and slave laser are collinearly overlapped. Note that within our measurement accuracy, the delay-voltage response at 100 Hz is still linear, with only a small decrease of the maximum scan range compared to lower modulation frequencies [see Fig. 7(b)]. At higher frequencies approaching 1 kHz, the hysteresis of the piezo actuator that modulates the repetition rate of the slave laser would lead to a saturation of the delay range at lower voltages.

Results
The majority of the results presented in the following sections have been measured with the reference system since the optimization of the mobile system and the taking of data with it are still ongoing. This section, therefore, aims to present a broad overview about the potential of THz-  TDS for the analysis of aircraft composite materials. Data from the DOTNAC TDS system, when available for comparison, are shown and explicitly denoted.

Impact Damages
GFRP laminates panels, which had been mechanically damaged with four different impact energies between 15 and 40 J, were investigated. Albeit being visible, the impact with the lowest energy could not be identified in the THz images, as can be seen in Fig. 8. The other three impacts with energies of 22, 30, and 40 J could all be detected (see Figs. 8 and 9). While p2p and frequency images highlight the slopes of an impact damage where the back   Corresponding delay (c) and frequency (d) images. Delay is at 71 ps below surface, i.e., after reflection at the back of the core; frequency is 691 GHz. Third row and bottom row: same field of view, but from the rear side of the sample. Whereas the p2p image (f) does not allow identification of the defects, the delay (g) and frequency (h) evaluations still reflect the presence of the defects: The time-domain wavefront visible in the lower part of (g) is distorted at the debond sites (indicated with white dashed rectangles), while (h) shows a slight contrast difference in the frequency domain. Delay time is ∼68 ps below rear surface, frequency at 290 GHz.
Optical Engineering reflection is reduced, the delay image can additionally identify compression of material or delaminations caused by mechanical shock.

Artificial Inserts
To simulate foreign inclusions, debonds, and delaminations, five GFRP panels (one laminate, plus one A-sandwich and C-sandwich each with Rohacell and honeycomb cores, respectively) with artificial inserts as mock defects were fabricated. Each contains 42 rectangular inserts ranging from 6 × 6 to 25 × 25 mm 2 that are patterned in three different depths within the samples. In the laminate sample made of 20 plies, the inserts are placed between the second and third, fifth and sixth, and tenth and eleventh ply, i.e., at 10%, 25%, and 50% of the total thickness. A single ply is ∼0.25-mm thick. A-sandwiches have the inserts within the four-ply top skin, between the top skin and the adhesive which bonds it to the core and between this adhesive and the core, so that all defects are placed on one side of the core, which is defined as the front side. The C-sandwich structures have the defects centered in the middle six-ply skin between the two cores, between the central skin and the adhesive bonding it to the front side core and between that adhesive and the core, respectively. Each insert group consists of airgaps, polyethylene, and Teflon patches. Examples are shown in Fig. 10.
In general, the detectability scales down with increasing depth and also depends on the material of the inserts. Airgaps yield the highest contrast due to the maximum difference in refractive index with the host material (2.07 is typical for fiberglass plies), whereas polyethylene and Teflon, commonly having THz refractive indices around 1.5 and THz transmittance around 90%, are harder to identify. This applies especially when they are close to or embedded in adhesive, which was measured to have a THz refractive index around 1.6, so their visibility diminishes. Furthermore, in general, the best defect identification could be done in delay evaluation and slightly worse in frequency mode, whereas p2p images are only useful for airgaps. Examples for measurements on A-sandwich front sides are shown in Fig. 11 (Rohacell) and in Fig. 12 (honeycomb), whereas Fig. 13 presents a comparison of data from front and rear sides of the laminate panel.
Regarding sandwich structures, there is a significant difference in the response caused by the core material: While both Rohacell and honeycomb cores have an average refractive index close to 1 in the lower THz band (∼1.02 to 1.05) and are thus expected to have similar reflection characteristics, the attenuation in transmission is almost an order of magnitude higher for Rohacell. We measured an absorption coefficient between 0.24 and 0.32 mm −1 for Rohacell and only 0.039 mm −1 for honeycomb cores. In our C-sandwich samples, the inserts, debonds, and delaminations are positioned close to the center between the two cores. THz radiation that reflects at such a defect, therefore, must traverse a core twice on the way from emitter to detector. This limits the detectability already a lot in honeycomb C-sandwiches, and even more in Rohacell sandwiches. In fact, we could hardly detect these defects in Rohacell C-sandwiches at all; for the same reason, detection on Rohacell A-sandwiches was almost impossible from the rear side when the central (single) core is between the defect and the sensor. We suspect scattering in the foam matrix as the main reason for the high THz opacity of Rohacell. For honeycomb C-sandwiches, the deep defects leave traces in the delay images to some extent. For that reason, the inspection of honeycomb Fig. 15 Top row: photograph (a) and THz p2p image (b) of Rohacell core A-sandwich patch with debond between skin and adhesive sheet approximately 0.75 mm below front surface. Bottom row: corresponding delay (c) and frequency (d) images. Delay is at 7 ps below surface; frequency is 640 GHz. The core thickness of 9.5 mm prevents measurements from the rear side because of its strong THz opacity (compare with Fig. 14). A-sandwiches from the rear is best done in delay mode. Note, however, that the measurements were performed with tight focusing for high lateral resolution so that in C-sandwiches with more than 10 mm thickness, the beam divergence beyond the focal area is problematic. By changing to optics with a wider beam waist, but longer Rayleigh length, reflections from the whole sample volume can be captured.

Debonds and Delaminations
The utilizability of THz-TDS was then tested on several A-sandwiches which contain debonds (both Rohacell and honeycomb core), honeycomb C-sandwiches with debonds or delaminations, as well as solid fiberglass laminates that include delaminations. The A-sandwiches show a similar behavior as their calibration counterparts: detectability in  general is good from the front side and still possible from the rear side when the sample contains a honeycomb core, but degrades for a Rohacell core. In fact, detection from the rear side was not possible on the Rohacell test sandwiches. Figure 14 displays THz images from front and rear side of an A-sandwich with avhoneycomb core, whereas Fig. 15 shows data from a Rohacell core A-sandwich (front side only).
C-sandwich test samples gave lower contrast in THz-TDS compared to the A-sandwiches, as was expected from the calibration samples. Only larger defects could be detected via time-domain evaluation, as can be seen in Fig. 16.
The three fiberglass solid laminates had delaminations placed at 0.5, 1.25, and 2.5 mm depth below the front surface, like in the calibration laminate with inserts. Airgaps, The delay in (a) does not correspond to the actual depth of the defect below the rear surface, which would be ∼70 ps. This layer is inaccessible for direct visualization because of dispersion and absorption; instead, the bulging of a ply closer to the rear surface due to the delamination underneath is weakly visible. Bottom row: corresponding frequency images (c and d), both at 551 GHz. The second defect which is visible in Fig. 18 (c and d) is not in the observation window of Fig. 19 (b and d).

Optical Engineering
i.e., mock delaminations, were almost the only inserts in the calibration sample that gave a weak contrast in the p2p images. In the test samples, the detection rate in p2p is even lower (probably because a real delamination is a weaker damage than an airgap created with a blade) and negligible. One of the test samples has a polyurethane (PU) coating that prohibited any detection in p2p mode. Delaminations in the test fiberglass solid laminates could be detected in delay and frequency evaluation, however, with lower contrast than in the calibration sample. Even when measuring from the rear side, the deepest defects at 4.5 mm below the surface could be spotted. Data from fiberglass laminates are presented in Figs. 17-19.

Water Contamination
Samples with three different volumes of water contamination were fabricated (0.2, 0.35, and 0.75 ml). For each set of volumes, A-and C-sandwiches with Rohacell and honeycomb cores were available so that we inspected 12 samples in total. The water was injected into the core of the A-samples and one of the C-sandwich cores, respectively. Therefore, the contamination in the C-sandwiches was only identifiable from the sample surface which is closest to that core. Measurements from a honeycomb core c-sandwich are displayed in Fig. 20.
For honeycomb samples, all water contaminations could be detected in delay and frequency evaluation; for Rohacell samples, detection was limited to delay mode. The differences in the penetrability of the cores influenced the visual detection of the water contents in delay mode. For the honeycomb structures, the contrast was best when looking at the signal that was reflected at the rear side of the core, where the absorption by the water content had the strongest influence on the image. The signal from the Rohacell samples on the other hand rather showed signs of the modified delay in the reflection in front of the core (see Fig. 21). The THz radiation that passed the core twice was attenuated too much by the core structure itself already to discriminate the influence of the water contamination well.
Water inclusions are examples of a kind of defect that could be more efficiently detected if the measurements could be executed in transmission geometry. Since water is an excellent THz absorber, a patch of moisture in a sample will result in a zero-transmission area with good contrast to the dry surroundings, if the host material itself is not an extremely strong THz absorber (see example in Fig. 22).

Porosity
Another case of improved detectability in a transmission measurement was the highly porous laminate test sample. The detection of increased porosity (∼4% of the volume) proved to be difficult with THz radiation. While effects of the enhanced porosity could be seen with the bare eye, and clearly in an ultrasonic image, the THz images did  not reflect an influence of the inhomogeneity within the high-porosity sample. The delay evaluation that allows peeking into the sample interior gave very similar results for both laminates, as can be seen in Fig. 23. Obviously, we cannot resolve the features that indicate elevated porosity in reflection.
However, the highly porous sample yielded more inhomogeneous, grainy images in p2p and frequency mode in comparison to the reference sample when both were measured in transmission (see the p2p images of Fig. 24).

Coating Misprocesses
Finally, for measurements of misprocessed coatings on CFRP panels, two different batches of samples were fabricated, one with PU and one with rain erosion (RE) coatings. Misprocesses included missing cleaning steps, deliberate contaminations with oil, release agent, water, or powder, processing without primer or with the wrong amount of primer, and application of the wrong amount of topcoat. THz-TDS lends itself to the analysis of coating misprocesses through changes in the propagation time of the THz radiation within the dielectric layers that are induced by the faults.
Our time-domain data traces always exhibited two main negative peaks that we identify with the coating surface reflection and the reflection at the CFPR substrate panel. The negative peaks are caused by the 180 deg phase shift from the reflection at an optically denser medium and the conductive panels, respectively. Unlike the weak and dispersion-broadened rear side reflection peaks from the GFRP samples, here the two peaks are comparable in amplitude and shape because the CFRP panels are conductive, and THz attenuation and dispersion in the coatings are low due to their limited thickness of a few 100 μm. The time difference between these two main peaks corresponds to twice the propagation time in the coating layer. Intermediate peaks can be caused by partial reflections at the interface between topcoat and primer. Examples can be seen in Fig. 25(c). In most cases of flawed coating examined by us, the propagation time showed a deviation from that measured on the respective reference sample, with the largest difference being caused by applying a too thick primer layer. Furthermore, a deliberate variation of the coating thickness translates into a proportional change in the propagation time within the coating layer [see Fig. 25(d)]. A more systematic examination of the delay variation that takes into account the dielectric properties of the different components of the coating layers (primer plus topcoat for PU coatings, two different primers plus topcoat for RE coatings) could be applied to gain a deeper understanding of the various processing flaws.

Conclusions and Outlook
We investigated the NDT inspection capabilities of THz-TDS on a wide choice of typical aeronautics composites with various defects. In most cases, it could be applied for detection. A-sandwiches and solid laminates yield a good detection rate for different defect types like debonds, delaminations, moisture ingress, and foreign material inclusions. Limited usability, besides the impossibility to directly visualize damages concealed by conductive carbon fibers, was found whenever a sandwich core, in particular  Rohacell foam, had to be penetrated twice. This affects especially thick C-sandwiches that are problematic structures for many NDT methods. It should be mentioned that tests with the fiber-coupled system showed improved penetration of sandwich and laminate structures because of its longer Rayleigh length compared to the reference system so that data from the whole sample volume could be gained. If lateral resolution can be traded off against the possibility to image the reflection from the opposite surface even in thick samples, softer focusing thus would be an option. Very promising is the possibility to analyze coating layers on CFRP panels with high depth resolution, which could not be done with any conventional NDT technique available in the project. Now that the general applicability of THz-TDS for composites has been explored, the logical next steps are a detailed analysis of the results with respect to characteristic signatures of specific failure modes and a possible automatic identification of defects based on this. Such enhanced data processing and evaluation must be accompanied by steps to increase data acquisition speed and hardware improvements, e.g., the development of a robust and compact TDS transceiver sensor. René Beigang is a professor for experimental physics and head of the Ultrafast Photonics and THz Physics group at the University of Kaiserslautern. He started his scientific career in the field of laser physics, nonlinear optics, atomic physics, and generation and application of ultrashort laser pulses. He has been working in the field of THz spectroscopy and applications for more than a decade.
Joachim Jonuscheit is a business development manager and deputy head of department in the "Materials Characterization and Testing" (MC) department of the Fraunhofer Institute for Physical Measurement Techniques IPM. He graduated in physics at the Ludwig-Maximilians University of Munich and received his PhD degree from the University of Erlangen-Nuremberg. He worked at various laser companies for several years in several management positions (project manager, R&D manager, and production manager).
Patrick Mounaix received the degree of Engineer of Material Science in 1988 from the Ecole Universitaire D'Ingénieur de Lille (EUDIL) and the third cycle thesis on quantum devices in 1992 from the University of Lille (PhD). Then he joined CNRS and the High Frequency Department of the Institutd'Electroniqueet de Micro-électronique du Nord (IEMN). He moved to Bordeaux in 2002, where he developed a new terahertz spectroscopy activity. Now he is currently interested in equilibrium and time-resolved terahertz spectroscopy techniques for solid and liquid materials and terahertz imaging applications, chiefly computer tomography for art and NDE industrial applications.
Pablo Venegas has graduated as an industrial engineer and has more than 7 years of experience in the aeronautical sector in the research and development of thermographic techniques applied as NDT tools. He has a master's degree in investigation in industrial technologies and is now a PhD student. He has been working in The Aeronautical Technologies Centre as a project manager for more than 6 years, where he has gained experience in national and European projects in the investigation of the thermographic technology as well as in the evaluation and validation of new NDT techniques. At the beginning of 2013, he has become head of NDT area in The Aeronautical Technologies Centre.
Laura Vega has graduated as an industrial engineer and has more than 5 years of experience in the aeronautical sector in the research and development of thermographic techniques applied as NDT tools. She has a master's degree in investigation in industrial technologies and is now a PhD student. She has been working in The Aeronautical Technologies Centre as a project engineer for more than 5 years, where she has gained experience in national and European projects in the investigation of the thermographic technology as well as in the evaluation and validation of new NDT techniques.
Ion López has graduated from E.T.S.I. of Bilbao as industrial engineer and was a PhD student at the University of Cantabria. He has been working as head of Singular Projects at CTA (Aeronautical Technologies Centre) since 2008, participating in different certification test campaigns for the aeronautical sector and different R&D projects. He has been involved as project manager in several structural test projects related to different aircraft models (A320, A340, A380 and A400M). His main research domains are combustion and scale model fire tests and infrared thermography as NDT technique for defect detection.
Marijke Vandewal graduated in 1996 from the Royal Military Academy, division Polytechnics, specialty Telecommunications. After a specialized avionics Air Force training, she was assigned at the flight Avionics in Kleine Brogel (Belgium), an F-16 Air Force Base. Then she was transferred to Brussels as a software engineer on C-130 transport aircraft. In 2000, she became an assistant at the Royal Military Academy, in optronics, radar, and microwave techniques. In cooperation with DLR (Deutsches Zentrum für Luft-und Raumfahrt) she obtained her PhD degree in engineering sciences in 2006. Since 2008, she has been teaching as a military professor at the Department CISS (communication, information, systems and sensors) of the Royal Military Academy.
Biographies and photographs of other authors not available.
Optical Engineering