Ultrasonic guided wave-based debond identification in a GFRP plate with L-stiffener

This paper presents a robust assessment of debond in a glass fibre-reinforced polymer composite structure with L-stiffener attachment. Towards this, the ultrasonic guided wave (GW) propagation based laboratory experiments have been carried out on a stiffened composite panel with piezoelectric transducers for the excitation of GWs and a scanning laser Doppler vibrometer for sensing the GW propagation. To study the changes caused by the stiffener and debond a signal processing based multi-point analysis has been carried out. The proposed methodology consists of two steps. Step 1 using the full wavefield root mean square energy map-based approach to check the presence of debond. Step 2 using point-wise measurements to study debond localization and size estimation using a baseline free signal coefficient difference algorithm (SCDA). The proposed processing approaches are applied for an in-depth analysis of the experimental signals that provide information about the interaction of GWs with stiffener and debond. The mentioned approaches take advantage of the asymmetry caused by the damage. For the applied SCDA methodology there is no need for full-wavefield measurements, healthy case measurements, as only a few measurement points can be enough for the assessment of stiffener debond in such structures.

(Some figures may appear in colour only in the online journal) * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
Fibre-reinforced composite structures offer many advantages like high stiffness-to-weight ratio [1-3] acoustic damping, and construction flexibilities, amongst others. Besides several attractive properties, these structures are prone to damage like crack, delamination and debond [1]. Such hidden damage can propagate further while in-service and leads to a sudden collapse of the structure [4,5]. Hence, early-stage detection of such structural damage is necessary.
Ultrasonic guided wave (GW) propagation-based structural health monitoring (SHM) has shown its potential for the detection and localization of structural defects. GWs are surface waves that can travel longer distances and could be utilized for damage identification based on wave-damage interaction phenomena [6]. GW based methods are very effective in identifying hidden debonding in composite structures [7,8]. Application of low-cost piezoelectric transducers (PZTs) in GW based SHM are widely adopted for defect identifications in composites [9,10]. PZT based GW methods using the numerical and experimental study of elastic wave propagation in composite laminate was conducted by Gresil et al [11]. The authors also noticed that the damage detection results depend on the temperature, PZT bond line thickness, geometry and mechanical properties of the sample.
The most common problem that occurs when bonding substructures together is the debonding issue caused by various scenarios [8,9]. Breathing debond analysis in composite structures with stiffener using GW measurements was analysed by Sikdar et al [12]. Golub et al [13] analysed GW scattering at the interface regions of structures. Bekas et al studied debond and impact analysis in mono stringer using PZT [14]. Zhang et al compared experimental, numerical signals to locate the debond [15].
Debonding analysis using scanning laser Doppler vibrometer (SLDV) techniques are gaining importance as they allow for analysis and visualization of GW velocity spread from measured scanning points [16]. GW analysis in the stiffened plate using SLDV and artificial neural network-assisted codes were studied in [17,18]. Balasubramaniam et al [19] studied the interaction of Lamb waves with a glass fibre-reinforced polymer (GFRP) I-joint stiffener and Philibert et al [20] analysed T joint stiffener made of carbon fibre-reinforced polymer (CFRP). SLDV wavefield visualization and processing for the damage in the bonding area between the composite plates were analysed by Sohn et al [21]. Delamination location between the composite plates was determined by Duflo et al [22]. SLDV based analysis in the anisotropic plate was analysed by Owens et al [23]. Damage detection analysis based on varying velocities using the SLDV approach was performed by Maio et al [24]. Stiffened carbon fibre plate was analysed using SLDV [25] by Zheng et al with changes in velocity and time of arrival [26] approaches. The use of the root mean square (RMS) energy of GW in damage detection was analysed by many researchers [27][28][29][30]. The author of [28] apart from the RMS energy studies also used weighted RMS (WRMS) and radial WRMS to study damage detection.
Marks et al employed the SLDV and windowed RMS studies but focused on an aluminium plate with an unequal angle stiffener with a debond [30]. Also, an aluminium plate but with a T stiffener arrangement was studied by Mandal et al employing a permanent array of transducers and looking at the nonlinear waves generated by the debond [31]. As far as composites are concerned the efforts to assess the debond in CFRP were reported in [31,32] based on referential measurements. The GFRP structure (part of a wind turbine blade) was studied in [32]. The focus was on the debond of the spar from the blade skin. The authors analysed the wave amplitude behaviour but did not focus too much on the signal processing indicating it as future work. SHM as mentioned earlier is used by many industries with various signal processing methods and GW is one of them [1,19].
The research in this paper focuses on processing the data obtained from the SLDV using various methods of GW analysis. Investigated damage is the debond of the L-stiffener from the GFRP plate. A comparison is done to understand the GW propagation in the debond region and without the debond region using the SLDV measurements and analysis of the full wavefield (FWF), RMS visualization. These studies of step 1 help us to verify if the debond was properly introduced by differentiating it from the region without it (perfectly bonded). In step 2 point-wise measurements are utilized to study debond localization and size estimation based on signal coefficient difference algorithm (SCDA). Four cases were studied using SCDA (a) case 1: 50 kHz-horizontal sensing path analysis, (b) case 2: 50 kHz-cross sensing path analysis, (c) case 3: 100 kHz-horizontal sensing path analysis, (d) case 4: 100 kHz-cross sensing path analysis. The debond area plotted using SCDA is compared with the original debond size using pixel counting. This is done as a part of error estimation.
Taking into consideration of the published works the novelty of the research presented here is the use of a fully baselinefree approach of debond assessment, localization and size estimation based on SCDA. The SCDA method developed is modified and improved to check the SCD's from the signals and does not require a healthy sample as described in the previous research [33,34]. The methodology was applied to the signals collected by SLDV (single actuator and multi-sensing points for analysis). All the analyses were made for data gathered from the GFRP composite plate with the stiffener and debond. No referential state of the plate was needed. The paper highlights that there is no need for the selection of higher harmonic waves in the referential free assessment of debond in a stiffener.

Two-step process in debond analysis
A two-step methodology is utilized (figure 1) in debond analysis of GFRP structure with the L stiffener.
Step 2: SCDA-SHM analysis for localizing and size estimation of debond.

Composite structure studied and experimental process
The stiffened GFRP composite panel as shown in figure 2 measures 50 cm (L) by 50 cm (W) with a thickness of 0.27 cm (H). The plate is equipped with a bonded L-shaped stiffener of length 50 cm, a height of 3 cm, and a thickness of 0.3 cm. The plate is made of 12 layers of GFRP prepreg (VV192T202 IMP503 from G. ANGELONI, www.g-angeloni.com/). The stiffener is bonded using an adhesive film (AF3109 2U from 3M, www.3m.com). The PZT disc transducers (Ø 0.1 cm) were attached at two different locations as shown in figure 2. The discs were manufactured by CeramTec (www.ceramtec.com/) from SONOX P502 material. The debond with a length of 5 cm and a width of 3 cm (stiffener width) is located at the centre of the stiffener. The debond was created by the local cutout of the adhesive film. A room temperature of 21 ± 1 • C was maintained using air conditioning during the experiment.
The PZT 1 and PZT 2 were glued to the stiffened side of the composite plate (figure 2). The frequency test of 50, 100 kHz were performed with five cycle burst Hann window modulation. The wave measurement was based on the SLDV. In the research, Polytec PSV 400 SLDV instrument was utilized and the setup is shown in figure 3. The device measures the surface vibration velocity along the laser beam. Since the laser head was kept relatively far from the measured object, the measured response is approximately the out-of-plane component of velocity. The measurements were registered for a set of scanning points distributed uniformly on the panel surface. The measured surface was the flat one (without the stiffener and PZTs).
PZT 1 was placed at the location facing the centre region of the stiffener to analyse wave interaction with the debonded region, while the PZT 2 was placed slightly away from PZT 1 for comparison with the results obtained for PZT 1 excitation. In both cases, the wave firstly propagated in the panel then entered the stiffener region and then left the stiffened region. The signals are registered with the help of one SLDV head. Firstly, the laser beam is focused on the GFRP panel. The panel is attached to the stand with the help of nylon strings to prevent any external source of vibration. The filter cutoff frequency values are set properly in the SLDV instrument to filter out the frequencies outside the excitation range. The chosen sampling frequency is ten times higher than the frequency of excitation, so it respects the Nyquist-Shannon sampling theorem. The very dense sensing mesh (at least 60 000 measurement points-2 mm point spacing in x and y direction) allowed for correct space sampling of the shortest wave propagating in the structure.

Results: GW characteristics study and debond identification
The aim of step 1 is to visualize the GW propagation in the properly bonded region of the structure against the debond region. The section involves FWF analysis, RMS checks between debond region and properly bonded region.
Step 2 using SCDA to localize debond and to estimate its size using pixel theory.

Step 1: results-FWF analysis in damage detection
At the beginning of the studies, signals (50 kHz) gathered from the whole surface were analysed. The FWF for excitation at PZT 1 is shown in figure 4 for chosen time instants. The wave propagation is visualized just after (78 µs) the excitation. The waves propagate in an elliptical fashion. One can see the slower wave mode (GW-mode II) in the dark shades near the PZT location, whereas the faster (GW-mode I) wave mode already reached the stiffener location and converted to a slower wave mode with a shorter wavelength. Later, the faster wave reaches (128 µs) the top edge in and the wave that was created at the stiffener propagates up and down from the stiffener. Whereas, the slower wave mode from the PZT 1 propagating upwards has not reached the stiffener yet. Since PZT 1 is close to the bottom edge the slower wave mode is already reflected from it and we see it propagating upwards.
The slower wave mode finally reaches the stiffener location (250 µs) and is partially reflected and transmitted through the stiffener region. Also, this wave mode reached the side edges and is reflected from them. The FWF plot shows some disturbance in the GW. This confirms the presence of the adhesion failure region due to debond. The debond region at the centre of the panel has created some changes to the wave pattern, reflection and some interference of the wave. After crossing the stiffener location, the GW-mode I waves experience multiple reflections from the boundaries throughout the panel. The GW-mode II wave mode attains further reflection at the stiffener, side, and top edges.
In the case of excitation at PZT 2 (figure 4), the general wave behaviour is similar to the observed GW behaviour of PZT 1 excitation. But the reflection from the right edge is seen earlier as the PZT 2 is placed close to the boundary. However, there is an important difference at the stiffener location. In this case, (250 µs-PZT 2) the wavefront at the stiffener is clear and continuous, while for the possible debond case as predicted earlier (250 µs-PZT 1) the wavefront is disturbed indicating different conditions encountered by the wave. This will be further analysed with the RMS visualization.

Step 1: results-RMS analysis in damage detection
To analyse the wave energy distribution, the registered wavefields are studied with RMS methodology (equation (1)). It is a standard RMS equation [1, 27,28] that was applied to the SLDV data. Results in the form of RMS energy maps plotted in the logarithmic scale (log RMS) are presented in figure 5: where k i -ith sample amplitude for points along with x, y-axis, and N-the number of time (t) samples. The RMS result for PZT 1 excitation is presented in figure 5(a) for 50 kHz. The region of debond is visualized as a brighter area at the stiffener. Also, there is a bright area above the stiffener indicating the energy of the wave that travelled through the debonded region. This is not present for the PZT 2 excitation ( figure 5(b)) because the stiffener is properly bonded in this case. The fringes visible in this case are related to waves reflecting from the right edge that is close to PZT 2. The excitation frequency of 100 kHz also produced similar results.

4.3.
Step 2: results-debond size estimation using SCDA After analysing the GFRP structure with step 1 signal processing methods, it was shown that debond exists in the centre region of the stiffener, and other regions look properly bonded.
SCDA based baseline free debond localization approach is applied to identify the hidden debond and to approximate its size in the stiffened GFRP panel. This was done by taking a dozen equidistant scanning points as shown in figure 6 by exciting PZT 1. The time-domain Hilbert transformation (HT) which allows creating the envelope of the signal [35,36]  is utilized for time-domain signals. As mentioned earlier four cases (horizontal and cross paths) were studied as shown in table 1 and figure 6.
An example signal comparison of sensing points (S1, S2) and (S7, S8) is shown in figure 7(a) for 50 kHz excitation frequency to illustrate the changes in the amplitude values of signals near to the debond. The signal coefficient differences (SCDs) are obtained by comparing the HT-based envelopes ( figure 7(b)) within the GW-mode II amplitude in the The debond probability distribution is computed to visualize the possible defected regions by applying the extracted SCDs as inputs at each pixel. The image quality is further improved by using an image-fusion technique at each pixel. The debond localization probability, D d of any arbitrary position (x, y) within the selected sensor network is described as: where, d ij (x, y) is the debond signature distribution probability, measured from the sensor-sensor pair: i-j and the SCD is expressed as: where, s b is the HT of the initial-sensor (i.e. 'i') signal and s d represent the HT of the targeted-sensor (i.e. 'j') signals in each sensor-sensor pair: i-j, t 1 (variable/path) is the time of arrival of GW-mode II in the signal and t 2 = (t 1 + duration of the selected wave mode), is an elliptical contour-shaped spatial distribution function with non-negative values as represented in figure 8, where, where the k ij represents the distance between the initialsensing point 'i' and the target sensing-point 'j' for each sensing pair (figure 8) and α is the scaling parameter, which reduces the size of the predicted debond regions and is independent of propagating GW velocity. The magnitude α is obtained empirically and in this study, it is assumed as 1. 15. This debond localization code is prepared in MATLAB that calculates a (12 ×12) matrix with 144 combinations of HT signals as inputs. In the process, the combinations of sensor paths: S#1-1,…, 1-12; S#2-1,…, 2-12; 3-1; S#3-1,…, 3-12; S#4-1,…, 4-12; S#5-1,…, 5-12; S#6-1,…, 6-12; S#7-1,…, 7-12; S#8-1,…, 8-12; S#9-1,…, 9-12; S#10-1,…, 10-12; S#11-1,…, 11-12 and S#12-1,…, 12-12 as illustrated in figure 6(b) are considered to image the debond region in the structure. The value of debond index (SCD magnitudes) at each pixel is determined by processing the sensor signals from the sensing paths (i.e. sensor (i)-to-sensor (j)) and plotted accordingly. The debond localization maps in contour pattern and debond localized is shown in figure 9. The cases correspond to the previously mentioned cases in table 1.
The contour plots indicate the stiffener-baseplate debond regions in the experimental GFRP panel, corresponds to the higher SCD magnitudes. Threshold plots show the approximated size of the calculated debond with the red colour rectangle being the original debond size and the white threshold (90% of the contour map values) region being the approximated size. An example of threshold determination was carried out as shown in figure 10 for case 2. The threshold for the pixels was swept from 86% to 96% at steps of 2% and the mishits (false detections) and true positive indication (TPI) (true detections) were determined. The number of mishits should be as low as possible as it indicates the pixels which are detected as damaged but are healthy (false negative). On the other hand, the TPI index should be as high as possible as it shows pixels that are damaged and are detected as damaged (true positive).
As can be seen at 92% thresholds the mishits are 0, but the TPI is significantly lower. Qualitatively, the shape of the damage is not correctly identified. Similarly, at the threshold value of 88%, TPI is 100% which is desirable, but for that value, the number of mishits is quite high. As a result, a trade-off was made and the 90% threshold was chosen which shows an intermediate performance in terms of mishits and the TPI. Qualitatively, the 90% threshold also shows a very good estimate of the shape of the debond which is desirable.
In other applications, it is customary to define the threshold levels based on sensitivity studies done on simulated damage scenarios [37] or measurements under different ambient conditions to capture the uncertainties in the measurement [38]. Typically the threshold value that yields probability of detection (POD) and the probability of false alarms (PFAs) within the acceptable ranges is chosen. Such a sensitivity study is beyond the scope of the current paper but is identified as part of future work. It should be noted that the TPI defined here is analogous to the POD, while the number of mishits is analogous to the PFA. It was observed that 50 kHz identified the debond size better than 100 kHz results. Table 2

Conclusions
The paper shows an approach for assessing the influence of stiffener-baseplate debond on GW propagation in the GFRP composite panel. The step 1 results show wave reflections and transmissions influences the proposed assessment approaches and helps to assess the debond location in the structure.
Step 2 helps to locate debond and estimate its size approximately. Summarizing the important observation and results obtained as follows; • The SLDV based laser vibrometry and SHM results showed that it is not necessary to conduct the time-consuming FWF measurement, only a few measurement points could be enough for stiffener debond localization and size estimation. • The results are useful for a typical nondestructive testing (NDT) approach since measuring a few scanning points with a laser or contact probe is a fast process. Moreover, currently, there are possibilities to both excite and sense the GWs by laser sources [39] giving the proposed approach more flexibility in future applications. • The FWF results (step 1) of section 4.1 provide a lot of information about wave reflection, are seen as well as the difference in energy distribution at the debond location. However, the disadvantage is the measurement time needed to gather the whole surface response. The FWF and RMS visualisation (step 1) of section 4.2 showed wave interaction with debond of the stiffener.
• Analysing the HT coefficient and signals gave conclusive evidence that GW propagation changes before and after the stiffener (figure 7). • SCDA (section 4.3-step 2) a baseline free method does not require a healthy sample [33,34] for measurement and thus reduces overall calculation time in debond size estimation. • The SCDA at the end gave us a conclusive location of debond with SCD based contour plot. The pixel counting gave us the approximate size of the debond and the estimated error was relatively low for cross paths (table 2).
In this research study, the excitation was made by PZT and the laser only sensed the response. But if we consider the SHM approach with permanently placed PZT this could be realized by putting the PZTs at the locations of the laser sensing points. This would give even more flexibility to such a system since such a sensor network allows to excite waves from more locations (each sensor node).

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).