Defect Detection in Bidirectional Glass Fabric Reinforced Thermoplastics Based on 3-D-THz Imaging

Nonhomogeneous, bidirectional glass fabric reinforced thermoplastic (GFRT) composites are used in the automobile industry to produce economic and environment friendly vehicles. Despite the fact that GFRT materials have superior mechanical characteristics and lower manufacturing cost compared with traditional materials, they are prone to significant deformation and defect emergence during the forming/manufacturing of parts from prefabricated laminates. Some defects, especially hidden ones like delamination and consolidation defects, can significantly reduce the mechanical properties of the final parts. It is essential to use 3-D nondestructive testing technology to detect and identify such hidden defects. This research investigates the ability of 3-D-THz imaging to detect hidden defects (i.e., delamination and consolidation) in GFRT composite materials. Frequency modulated continuous wave (FMCW) technique is used to generate 3-D-THz cross-sectional images. Nondefective and defective samples with a thickness of 1.5 mm have been measured with two 3-D THz imaging systems operating in the frequency range from 238 to 316 GHz and from 499 to 733 GHz. In addition, a simulation of the reflected FMCW spectrum is presented to analyze the measurements and intercompare the performance of the two imaging systems. A systematic analysis in a large sample set demonstrates that a delamination defect with a thickness of 125 $\mu$m as well as a consolidation defect with a difference in thickness less than 0.4 mm can be successfully detected by the 700-GHz imaging system.

Fiber reinforced polymers (FRP) are lightweight composite materials consisting of many layers of fibers encased in polymer matrices [5]. The FRP composites have diverse forms based on the type of fiber and the category of polymer. Each of them has several mechanical properties and is adequate for a specific application area [6], [7]. In the automotive industry, the glass fiber reinforced thermoplastic (GFRT) is intensely envisaged as an attractive new alternative to produce vehicle parts. This is not only due to the low-cost processing of the parts by conventional forming procedures in presses but also due to the high stiffness, low weight, corrosion resistance, and recyclability in comparison to traditional materials, such as iron, copper, and steel [8], [9]. However, during the production of the GFRT panels (semifinished products), the fibers can be displaced, damaged, or torn. Besides, any insufficient temperature, pressure, or cooling can cause a layers separation or a difference in the material thickness. Moreover, during forming the finished parts, additional defective areas can be created or existing defects can be enlarged [10]. As such materials are starting to be used in mechanically relevant parts of cars, any deformation in the material or deficiency in the fiber distribution decreases material's stiffness and strength that, hence, affects the car safety. For that reason, nondestructive testing (NDT) is crucial to detect outer and/or inner distortions in such materials without causing damage to the original part.
Various NDT techniques are available, such as visual, ultrasound, infrared thermography, and X-ray testing, and each has strengths and shortcomings [11]. For example, the visual inspection is widely used to detect defects in the surface due to its simplicity and reliability, but it is not sufficient for hidden defect detection. The X-ray technology can provide a high resolution 3-D-image with an accurate position of hidden defects, but, unfortunately, it ionizes molecules. For this reason, it requires shielding which makes it impractical for direct use in manufacturing lines in factories. The ultrasonic testing (UT) is a safe solution, but it is not suitable to inspect thin, rough, or heterogeneous materials. The infrared thermography (IRT) is a safe, low-energy measurement system that can be used easily for in-line inspection, but cannot provide information about the defect position in depth. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ Imaging at terahertz frequencies offers an interesting direct 3-D subsurface imaging capability without requiring ionizing radiation.
The main weakness of the THz image compared to the X-ray and optical ones is the lower resolution. Nevertheless, due to the great advances of THz technology, recently the enhancement of the THz image resolution has been given a great focus by researchers either by improving the THz imaging techniques [12], [13], [14], [15] or developing super resolution algorithms [16], [17], [18], [19], [20].
The rest of this article is organized as follows. In Section II, prior research in the field of NDT of composite materials based on THz technologies are summarized. In Section III, the measured samples and the THz-FMCW imaging system are explained. Furthermore, the THz-FMCW reflected signal is simulated. The measurement results of 300 and 700-GHz imaging systems are presented and discussed in detail throughout Section IV. Finally, Section V concludes this article.

II. STATE OF THE ART
In the field of composite materials inspection based on THz technologies, most prior research implemented terahertz time domain spectroscopy (THz-TDS) systems to detect different kinds of defects in various types of composite materials like carbon fiber reinforced polymer (CFRP) [21], [22], [23], [24] and glass fiber reinforced polymer (GFRP).
The THz-TDS is a broadband system (up to 10-THz bandwidth) based on femtosecond lasers to generate THz-pulses; therefore, it is a good choice to inspect thin composite materials and measure thicknesses in the μm range. For example, in [25] multidelamination defects in a commercial unidirectional glass fiber reinforced plastic composite material with a thickness of 1.5 mm have been detected based on the THz-TDS imaging system with a frequency range from 0.1 to 3 THz. The delamination layers have been produced by inserting Teflon films with a thickness of 100 μm between the GFRP layers. Although the THz-TDS is not a 3-D imaging system, the authors could successfully determine the location of each delamination in Z-direction as well as calculate the thickness of each one by analyzing the THz-waveforms in the time-domain.
Since the composite materials have to be inspected in-line; as a consequence, the researchers in this field have recently focused more on the THz systems based on all-electronic devices in spite of the comparatively lower bandwidth (in the GHz range) of those systems. The THz-FMCW is a narrow-band low-power system based on all-electronic devices to generate a continuous frequency. This technique gives information about the Z-direction in the frequency domain and by moving the system along X and Y directions, a 3-D-image can be measured easily. Although this system has some limitations in both depth and lateral resolutions, some researches have shown promising results in detecting defects in the composite materials based on this system. For instance, Becker et al., [30] reported some results about detecting a delamination defect in the oxide fiber reinforced ceramic composites with two different thicknesses of 10 and 3 mm using the THz-FMCW system running at 100 and 300 GHz, respectively. In [31], a hole defect has been detected, but, in a homogeneous material [polytetrafluoroethylene (PTFE)] with a thickness of 1.5 cm based on FMCW THz radar with an operation frequency of 100 GHz and a bandwidth of 35 GHz.
However, to speed up the scanning time of the singletransmitter-single-receiver THz-FMCW 3-D-imaging system, currently we and many more researchers are working on expanding such systems to high-speed multiple-input-multipleoutput (MIMO) imaging system [12], [13] in order to reach real-time imaging capabilities. In [32], the authors have designed a THz system with six sensors (three receivers facing three transceivers) based on the FMCW technology and the system has been tested on a glass fiber reinforced plastics material with different thicknesses of 5, 6, 7, and 10 mm.
In previous research, up to our knowledge, THz-FMCW systems have been basically used to detect defects in materials with thicknesses higher than 3 mm due to the depth resolution limitation. In addition, none of the previous researches (neither those based on the THz-TDS nor the THz-FMCW) have inspected the GFRP composites in which the thermoplastic material is used as polymer matrices [glass fiber reinforced thermoplastic (GFRT)]. Our main contribution is to build high-bandwidth all-electronic 3-D-THz imaging system to detect subsurface defects for the first time in a thin (1.5-mm thickness), inhomogeneous, bidirectional E-glass fiber reinforced thermoplastic composite material. A detailed description and introduction to our system is presented in [33], [34], [35], and [36].

A. Inspected GFRT Laminates
The analyzed samples are made from a bidirectional twill 2/2 E-glass roving fiber reinforced polypropylene. The twill 2/2 indicates that two wrap yarns (the horizontal fibers) travel over and under two weft yarns (the vertical fibers) that performs a diagonal pattern which is seen in the photograph of the nondefective sample in Fig. 4(g). E-glass is a type of the glass fiber that has low electrical conductivity. Polypropylene is a thermoplastic material which softens at a temperature from 160 to 208 • C [37] and hardens again when it is cooled down. This property gives the GFRT composite material the ability to be reshaped and recycled [38].
In general, the semifinished product of the GFRT composite material is produced by layering the glass fabric and the thermoplastic plies then applying a glass-transition temperature and a certain pressure to combine them. After that, the compressed layers are cooled down [38], [39], [40], [41]. However, any inadequacy in the temperature, pressure, or cooling can lead to different sorts of distortions. The inspected specimens consist of three layers, i.e., two outside layers and one middle layer. They have dimensions of 80*25 mm with a thickness between 1.3 to 1.5 mm. In this research, three different specimens of GFRT laminates have been measured as follows.
2) Defective Specimens With a Delamination Defect: the delamination defect can be defined as a separation in layers. It is considered the most critical defect in GFRT laminates [10], [42], [43], [44]. It has been produced in this project by inserting a kapton film with a thickness of 125 μm to separate the sample layers. 3) Defective Specimens With a Consolidation Defect: the consolidation defect is a variation in the specimen thickness. It happens due to the implementation of low/insufficient pressure during the incorporation of the GFRT layers. This causes the layers of the fabric to rise up, creating holes which are subsequently filled with the melted thermoplastics.

B. 3-D THz Imaging Setup
In this project, two all-electronic 3-D-THz imaging systems with identical configuration but different operation frequencies and bandwidths have been built to examine the effect of the frequency and bandwidth on a defect detection in GFRT laminates. Fig. 1 shows the platform of the 300 and 700-GHz imaging systems. The two 3-D imaging systems are based on the FMCW technology [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57] and they are similar to the system in [33] and [34]. The imaging unit in both systems consists of a hollow wave guide transmitter and receiver mounted on the same platform. The 2-D image is generated by driving the imaging unit (transceiver) along the X and Y directions using two stepper motors with a step size of 0.2625 mm. The sawtooth FMCW signal is continuously transmitted to measure along the Z-direction. At each lateral position the depth is measured and by scanning the entire sample, the cross-section images are generated. Table I presents the system parameters. The 300-GHz imaging system operates within the frequency range from 238.73 to 316.9 GHz and a bandwidth of 78.17 GHz, whereas the frequency range of the 700-GHz imaging system is from 499 to 733 GHz and the bandwidth is 234 GHz.
The lateral resolution is calculated based on (1), where λ is the wave length and N A is the numerical aperture The depth resolution is determined by the Rayleigh limit (2), where c 0 is the speed of light, n is the refractive index, and B is the bandwidth The complex refractive indices of the GFRT and kapton film materials have been measured using the transmission mode of the time-domain spectroscopy [58]. Based on (2) and the measured refractive index, the depth resolution in the GFRT and kapton film materials can be calculated. Table II presents the measured complex refractive index and the calculated depth resolution for both materials and both system frequency operation ranges.

C. Simulation of Reflected THz-FMCW Signal
In this section, the mathematical model of the THz reflected signal from a material with a thickness of Δz 1 are derived to help us analyze the real measurements and compare the performance of the two imaging systems. In general, the number of reflections from different mediums equals to where r N r is the number of reflections; r N m is the number of mediums.
Accordingly, the number of reflections from a material in air (air-material-air) is two reflections. Fig. 2 presents the block diagram of the transmission and reflection components of the reflected THz-FMCW signal from a material with a thickness of Δz 1 in air. Z 0 is the reference plane. The ideal reflected signal E ideal (ω) at an angular frequency ω is given by where E 1 is the reflected signal from the material surface Z 1 and E 2 is the reflected signal from the material backside Z 2 . The sum term represents the backward and forward reflections inside the material [Fabry-Pérot effect (FP)] [58]. Since it is a geometric series it can be written as r r is the reflection from a medium m when hitting the next medium m + 1 r t is the transmission coefficient from a medium m to the next medium m + 1 This quantity is measured around a center function ω c with a finite bandwidth B. Thus, the overall signal is given by Assuming the reflection interfaces of the material are perfectly flat resulting in completely separate reflections (p 2 1 (ω, Δz 1 ) · r 12 (ω) · r 10 (ω) 1 ⇒ F P (ω) ∼ = 1). By implementing the inverse Fourier transform on (11) and folding by the Dirac signal, the expected reflected signal in the time-domain can be calculated as follows: where r E 0 is the initial transmitted signal; r α m is a constant equals to From (12a) it can be seen that the expected reflected signal from a material with a thickness of Δz 1 is the summation of two sinc functions E 1 (t) (12b) and E 2 (t) (12c), which symbolize the reflection from the surface and the backside, respectively. Moreover, the term e −α 1 Δz 1 in (12c) shows that the amplitude of the second reflection is inversely proportional to the material extinction coefficient k m and its thickness Δz 1 .
The model of the reflected THz-FMCW signal (12) has been simulated in MATLAB. For simplicity, noise is neglected and the reference plane Z 0 is assumed to be on the sample surface (Δz 0 = 0 mm). The GFRT material with a thickness of Δz 1 = 1.5 mm has been considered. Fig. 3(a) exhibits the simulated E 1 (t) (the blue curve) and E 2 (t) (the red curve) of the 300-GHz imaging system and Fig. 3(c) represents their sum. Similarly, Fig. 3(b) and (d) display the simulated signals of the 700-GHz imaging system. It can be clearly observed that the depth resolution of the 700-GHz imaging system has been dramatically improved due to the bandwidth expansion from 78.17 to 234 GHz, which cannot be achieved in the 300-GHz imaging system due to the system limitations. On the other hand, based on the characteristics of the GFRT material (Table II), the extinction coefficient k is larger at higher operation frequency. This will, accordingly, affect the penetration and reduce the amplitude. Please notice from Fig. 3(b) that the difference in amplitude between two reflections in the 700-GHz imaging system is around −21.8 dB, while it is approximately −8 dB in the 300-GHz imaging system in Fig. 3(a).
From Fig. 3(c) and (d) the side-lobes effect on the depth resolution can be clearly seen. For example, in Fig. 3(d) instead of observing the second pick, which presents the reflection from the backside of the material, at depth = 1.5 mm two picks appear at depths 1.45 and 1.61 mm. This is due to the summation of the second side-lobe of the reflected spectrum from the front surface with the main pick of the reflection from the backside. However, to suppress the effect of the side-lobes on the depth resolution, Hamming window has been implemented on the measured data.

IV. RESULTS AND DISCUSSION
This section presents the experimental results of the defective and nondefective samples measured by the 300 and 700-GHz imaging systems. In addition, a comparison of the results of the two imaging systems is presented.
The specimens have been scanned by placing each one on a low refractive index foam sheet to minimize spurious reflections. The foam sheet is mounted in front of the transceiver unit at a distance R 0 equals to 41.4 mm in the 300-GHz imaging system  (Table I). Scanning one sample with dimensions of 80*25 mm takes approximately 14 min. During the measurement, each measured point is normalized based on where r M max is the maximum reflection measured from a metallic target.
r M min is the minimum reflection measured without any target.
r M is the reflection measured from a specimen.
The measurements and data collection are controlled by a LabVIEW program. After data collection, the 3-D-THz images are calculated and plotted in MATLAB by implementing the inverse Fourier transform to the Hamming window of the zero-padded data. The Hamming window and zeropadding have been implemented to suppress side-lobes and smooth out the reflected signals, respectively. It is worth to mention that the maximum signal-to-noise ratio (SNR) in the 300-GHz imaging system is around −35 dB, while the maximum SNR in the 700-GHz imaging system is about −20 dB.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.

A. Nondefective Sample
The nondefective sample has been measured by both FMCW imaging systems. Fig. 4 compares the results from both systems, where the left column represents the 300-GHz imaging system results and the right one shows the 700-GHz imaging system results. The photograph of the measured nondefective sample can be seen in Fig. 4(g). Fig. 4(a) presents the reflected spectrum from the nondefective sample at x = 65 mm and y = 4 mm measured by the 300-GHz imaging system, while Fig. 4(b) represents the reflected spectrum from the nondefective sample at x = 63 mm and y = 4 mm measured by the 700-GHz imaging system. From Fig. 4(a) and (b) it can be noticed that the reflected spectrum in both systems has only a single peak which denotes that the reflection from the rear face of the nondefective specimen cannot be distinguished. Looking at the depth layer (XZ layer at y = 12 mm) in Fig. 4(c) and (d), where the dashed yellow line in both figures refers to the surface reflection, it can be observed that the reason for the inability to distinguish the reflection from the rear face of the GFRT laminates in both systems is given by two different effects.
In the 300-GHz imaging system, the depth resolution in the air δ Z air,300 = 1.9 mm is on the order of the sample thickness Δz 1 = 1.5 mm; hence, the reflection from the backside cannot be separated clearly from the surface reflection. For this reason, in the XY amplitude image at z = 0 mm in Fig. 4(e), the fiber pattern of the surface cannot be seen obviously because this reflection includes not only the reflection from the front surface but also the reflections from the backside.
In the 700-GHz imaging system, this problem is solved because the depth resolution in the air δ Z air,700 = 0.64 mm is significantly smaller than the sample thickness Δz 1 = 1.5 mm. However, it can be observed from Table II that the extinction coefficient k is higher at higher frequencies (k GFRT,700 = 0.06 > k GFRT,300 = 0.039); therefore, increasing operation frequency inevitably decreases the amplitude of the backside reflection.
From the simulation results, it is expected that the difference in the amplitude between the front and rear interface signals at 616-GHz operation frequency is around −21.8 dB. Taking into account that in the real measurement, the SNR is around −20 dB, this means that the backside reflection cannot be distinguished from noise. In contrast to the 300-GHz imaging system, the XY amplitude image at z = 0 mm in the 700-GHz imaging system in Fig. 4(f) shows the fiber pattern of the GFRT surface clearly due to the higher depth resolution of the 700-GHz imaging system δ Z GFRT,700 = 0.267 mm.

B. Defective Samples: Delamination Defect
To imitate the delamination defect, kapton film with a thickness of 125 μm has been inserted between GFRT layers before applying the glass-transition temperature and pressure. Due to the fact that the kapton film does not melt before +400 • C [59], it will separate the GFRT layers. This separation can be seen in the microscope image of the cross section (XZ side) of the defective sample with a delamination defect in Fig. 5, where the yellowish layer shown in Fig. 5 represents the inserted kapton film and the blue ellipse points to the separation. The photograph of the measured defective sample with a delamination defect can be seen in Fig. 6(i).
In order to get a rough idea about the position of the delamination defect in depth, the distance between the front surface of the GFRT specimen and the front face of the kapton film at different points in the microscope image has been measured. It has been found that the delamination position varies between 0.4 to 0.7 mm due to the inhomogeneity of the GFRT composite material.
From (3), the number of expected reflections is four, because the number of mediums are five (Air, GFRT, kapton film, GFRT, Air). However, because the thickness of the kapton film (125 μm) is less than the depth resolution in the kapton film in both systems δ Z kapton-film,300 = 1.3 mm and δ Z kapton-film,700 = 0.35 mm (Table II); consequently, the reflection from the backside of the kapton film will not be differentiated from the reflection from the front side of the kapton film. Besides, the reflection from the backside of the GFRT specimen cannot also be distinguished (it has been explained previously in the Section IV-A). Wherefore, the expected reflections are two; one from the front face of the sample at z = 0 mm and the another one from the front face of the kapton film at z = 0.62 mm. Fig. 6(a) and (b) are the reflected spectrum measured by the 300-GHz and 700-GHz imaging systems, respectively, at x = 35 mm and y = 8 mm. It can be noticed that the 300-GHz reflected spectrum has only a single peak, whereas the 700-GHz reflected spectrum has two peaks. This demonstrates that the reflection from the kapton film can only be recognized by the 700-GHz imaging system. Also, Fig. 6(c) and (d) which represent the XZ layer at y = 8 mm emphasize this notice, where the yellow dashed line in both figures represents the reflection from the surface at z = 0 mm and the black dashed line in Fig. 6(d) points to the reflection from the kapton film.
The reflected amplitude images from the surface (XY plane at z = 0 mm) measured by both systems can be seen in Fig. 6(e) and (f). In addition, the subsurface layers at z = 0.6 mm measured by both systems are presented in Fig. 6(g) and (h).
The presented spectrum in Fig. 6(b) shows that the difference in amplitude between the two peaks is around −3 dB. However,  from Fig. 6(h) it can be noticed that the amplitude values in the delamination area varies between −7 to −35 dB. This indicates that not all spectra reflected from the delamination area have two separated peaks because the depth resolution in the air δ Z air,700 = 0.64 mm is not always less than the delamination position in Z-direction which changes between 0.4 to 0.7 mm.
It should be noted that the studied sample (artificial delamination defect) is more critical than the nature delamination defect (only air-gap). The main reason for that is the difference in the complex refractive indices between the kapton film and air which will therefore affect the reflection coefficient r 12 as well as the transmission coefficient t 12 . However, from (12c) it can be observed that the reflection from the second surface is effected by the reflection coefficient r 12 which is in the case of GFRT-air r GFRT-air = 0.4106−0.0105i higher than the reflection coefficient in the case of GFRT-kapton film r GFRT-kapton film = 0.1333−0.0073i . This leads to lower reflected amplitude from GFRT-kapton film than the reflected one from GFRT-air. As a conclusion, if the THz system can detect the artificial delamination defect, it will surely detect the nature delamination which has the same position in depth and same thickness.

C. Defective Samples: Consolidation Defect
The thicknesses of the nondefective and defective areas of the defective sample with a consolidation defect have been measured by a digital caliper. The thickness of the defective area is nearly 1.7 mm, while it is roughly 1.3 mm in the nondefective one. As a result, the number of expected reflections from this specimen is three; one from the front face of the defective area at z = 0 mm, one from the front face of the nondefective area at z = 0.4 mm and a reflection from the rear face of the specimen. However, because the reflection from the backside of the GFRT sample cannot be discriminated, as it has been explained in the Section IV-A, the number of expected reflections is only two. The photograph of the measured defective sample with a consolidation defect can be seen in Fig. 7(g).
The difference in thickness can clearly be seen by looking at the depth layer (XZ layer). Fig. 7(a) is the XZ layer at y = 12 mm measured by the 300-GHz imaging system. It is apparent that the 300-GHz imaging system cannot detect the difference in thickness because the depth resolution of the 300-GHz imaging system in the GFRT composite δ Z GFRT,300 = 0.692 mm is higher than the difference in thickness between the nondefective and defective areas 0.4 mm. Fig. 7(c) represents the amplitude image at z = 0 mm and Fig. 7(e) presents the subsurface image at z = 0.46 mm. Similar to what has been inferred from the Fig. 7(a), the reflection from the surface of the consolidation area and from the nondefective area cannot be separated.
The XZ layer data at y = 12 mm measured by the 700-GHz imaging system in Fig. 7(b) proves that the difference in thickness can be detected clearly with this system, because the depth resolution of the 700-GHz imaging system in the GFRT laminates δ Z GFRT,700 = 0.267 mm is significantly less than the difference in thickness between the nondefective and defective areas 0.4 mm. Fig. 7(d) presents the reflected amplitude image of the front face of the defective area at z = 0 mm and Fig. 7(f) shows the reflected amplitude image of the front face of the nondefective area at z = 0.417 mm.

V. CONCLUSION
This article discusses from a simulation and experimental viewpoint the ability of an all-electronic 3-D-THz imaging system to detect hidden defects in a lightweight composite materials with a thickness of 1.5 mm. The bidirectional E-glass fiber reinforced thermoplastic composite material has been inspected with a THz system for the first time in this research. The FMCW technology has been implemented to generate 3-D images. Two THz-FMCW 3-D imaging systems with two different operation frequencies; 277.8 and 616 GHz, have been utilized to inspect the nondefective and defective GFRT specimen. The performance of the two imaging systems has been simulated. In addition, the real measurement results collected from both 3-D imaging systems have been compared. It has been found that the 300-GHz imaging system is not sufficient to detect a hidden defect in a thin material due to the limitation of the depth resolution. However, the use of the 700-GHz imaging system which has a bandwidth three times higher than the bandwidth of the 300-GHz imaging system has improved the depth resolution remarkably. Due to this improvement, such a system can detect delamination defect with a thickness of 125 μm and measure the difference in material thickness which is less than 0.4 mm. In comparison to the THz-TDS system used in [25], our presented 700-GHz imaging system shows similar analysis capabilities (detecting delamination defect with a thickness of 125 μm in different kind of GFRP laminates but with the same thickness of 1.5 mm), but it has better properties. First of all, it is based on all-electronic devices and is a low-power system. It does not need special environmental conditions to work efficiently like the one in [25], where dry air needs to be provided to prevent water vapor absorption. Second, it is less expensive than the THz-TDS system used in [25]. In addition, determining the hidden defect position in depth is much simpler than in the THz-TDS system. Last but not least, it is significantly easier to expand such an electronic system into an MIMO system for real-time in-line inspection in a manufacturing environment.

ACKNOWLEDGMENT
All samples have been produced by Bond-Laminates -a company of the LANXESS Group.