A Novel, Non-Contact NDT Scanner Case Study: Thickness Measurement, Debonding and Defects Detection in Metallic and Composite Parts

The NDT methods currently used in aviation MRO are predominantly labour-intensive and time-consuming processes performed by human operators throughout the lifespan of an aircraft. These techniques are time-consuming, require perpetual training and are highly dependent on the operator’s skills. Thus, there is a growing need for more efficient, automated, and accurate NDT tools that will be able to provide faster and less labour-intensive assessments. This study presents a novel, non-contact, automated NDT scanning system under development, which aims to reduce the inspection time significantly. The proposed technique uses a non-contact, Lamb wave-based approach. A further essential step during the process is to use an automated positioning system. Thickness mapping and defect detection in metal and composite structures have been performed. A local thickness map in the order of 1 mm has been obtained through a fast-scanning process with comparable resolution to conventional inspection techniques. Overall, it is currently concluded that the proposed NDT scanner is a promising tool that potentially can reduce the inspection time while also having the potential to automate the damage assessment resulting in more efficient MRO inspection processes.


Introduction
During the maintenance process of aircraft systems, there is a growing need for efficient and automated non-destructive techniques (NDT) that can perform processing steps without highly trained staff.At this moment, a wide spectrum of methodologies make use of a limited number of techniques to perform diagnostic inspections.As illustrated in Figure 1, these NDT techniques include wave/diffusion-based mechanical and electromagnetic methods, ranging from ultrasonic testing (UT), acoustic emission, thermography, shearography, x-ray radiography, to some emerging methods such as computed tomography.The current, certified, state-of-the-art NDT used within aviation Maintenance, Repair and Overhaul (MRO) are largely labour-intensive and slow processes that human operators carry out with extensive ASNT/ NAS/ ISO training.Despite this training, which is periodically updated, the measurements' outcome still can depend on the operator, the applied NDT technique as well as the surface conditions.At the same time, the increasing use of composites in place of traditional metallic alloys creates new challenges in inspection and damage assessment, as composites are non-homogeneous and anisotropic materials.Moreover, it is well established that aircraft composite structures are particularly susceptible to impact damage, where induced damage can occur within numerous locations and at various scale levels, making it difficult to detect and assess the damage [1] efficiently.Especially, the occurrence of barely visible impact damages (BVID) in composite components is a serious problem, which threatens the structural safety of an aircraft, and should be timely detected [2].Each NDT technique has its own benefits and weaknesses.However, it rarely achieves its full potential when considering a full-scale diagnosis of possible defects [3].To bridge this gap, the objective of this project is to construct a modular, automated non-contact NDT approach in which, besides the proposed detection system, also the associated positioning, reporting, and data-handling are included.Use will be made of an UTbased non-contact array sensor system, in conjunction with a multi-applicable scanning system.At a later stage, other types of sensors are planned be used as well.In this study, the setup will be discussed in combination with the requirements for an automated positioning system and a corresponding datagathering system to generate maintenance reports and lay the groundwork for adaptive and, ultimately, predictive maintenance.The present paper is structured as follows.First, the applied methodology will be discussed, with the specifications of the experimental setup and the experiments.It will be followed by the obtained results, further work, conclusions, and recommendations.

Method
An ultrasonic sensor, consisting of a MEMS (Micro-Electro-Mechanical Systems) microphone array connected to a translation system, is used to scan the radiated Lamb wave energy.The sensor head further contains a number of laser profilers to measure the shape of the part under inspection to provide a full 3D representation.

Experimental setup
The setup consists of a vacuum activated rubber sleeve with a piezoelectric actuator on top (Figure 2a), that functions as the main source of waves.It can generate waves in the lowest ultrasonic frequency band [4], with a vacuum pressure of about 35 kPa.The actuator is connected to the aeroplane surface under consideration and should be located within the same aeroplane component inspected.The sweep is generated with a Keysight 33631A Waveform generator that can create signals from DC up to 120 MHz [5] and a Krohn-hite 7500 Power amplifier, which has a range from DC up to 1 MHz and that can generate a 200 V peak-to-peak [6].The non-contact MEMS-sensor array is composed of four identical modules that each contain 32 off-the-shelf MEMS omnidirectional acoustic/ultrasonic sensors, as illustrated in Figure 2b.They are placed 3 mm apart from each other, making the total size of the 128-unit array to be 384 mm.The sensors are integrated on a printed circuit board (PCB) shielded by a housing and connected to an A/D digitiser [7].The digitiser is an ACQ2106, 6 site ELF carrier, which can handle up to 192 channels, with up to 5Gbps fibre optic link and a gigabit Ethernet connection [8].In addition, the system uses 3 micro-epsilon scan control high performance LLT3000 red laser scanners.It can reach a 10kHz measurement speed, with a line linearity of up to 1.5  [9].The sensor system can be placed on a translational system or a collaborative robot (cobot).The translational system uses two MC1-20 ISEL single axis controllers.The cobot used is a Universal Robots UR10e, which can lift up to 12.5kg and has a reach of 1300mm, with an arm consisting of 6 rotating joins [10].To be noted here that the sensor head in use weights approximately 4kg, which well below the cobot's limits.An advanced data processing scheme has been developed, translating the wave field measurements into a thickness map of the part with a spatial resolution of 1 mm. Figure 2a and b.Rubber sleeve with a piezoelectric actuator and MEMS sensor array with lasers.

Experiments
The MEMS sensor array and actuator communicate with the positional system with a digital handshake, where electric pulses activate the positioning system.Each time the cobot moves along pre-defined "waypoints" on the track, a trigger pulse is generated.The actuator is located at a fixed location while the MEMS is moved along the intended track.Depending on the size of the area that needs to be inspected, one or multiple actuators may be used simultaneously.Despite that, different track patterns can be followed, currently the translational axis uses a lawnmower pattern, where a stepping interval of 1 mm is applied.During the experiments, the piezoelectric transducer uses a single shot record.It generates a 2 ms linear frequency sweep in the frequency range between 20 kHz and 250 kHz, enhancing the signal-tonoise ratio.This range is chosen to create Lamb waves in the thin-walled structure under inspection.In principle, the two fundamental Lamb wave modes (symmetric,  0 and asymmetric,  0 ) are created in the part.Due to the limited out of plane displacement of the  0 -mode, it will not radiate efficiently in air, unlike the desired  0 -mode (Figure 3) [7].The radiation losses of the  0 -mode is very low.The range that can be covered by a single actuator is mainly determined by loss-mechanisms in the part like viscoelastic attenuation in fibre composites.The distance between the MEMS array and the object under inspection is typically 100 mm or more.This minimises the risk of sensor surface collision and avoids multiple reflections between the sensor head and the surface of the part being inspected.The phase velocity of the Lamb waves is frequency dependent, consequently, the leaky Lamb wave's refraction angle is also frequency dependent.To avoid a blurring effect due to this frequency dependency of the refraction angle, the wavefield needs to be backpropagated to the surface.The actual position of the surface with respect to the sensor array is obtained with the laser scanners.The wave field is further processed to obtain a local thickness measurement using the dispersion curve (see Figure 3).For the metallic part, the dispersion curve is accurately known.For composite parts, the dispersion curve is estimated with a non-linear curve fitting algorithm based on a semi-analytical model.It incorporates material properties and directional effects, encountered especially in composite materials, with their highly anisotropic and inhomogeneous composition [7].A contour-following mode is applied with the translational system for curved surfaces, which is also expected to be part of the cobot steering.The distance between the MEMS array and the surface is monitored using lasers that scan along the locations on which the sensors are directed.When this information is included in the wavefield analysis, topological artefacts in the thickness map can be minimised [7].In this research, two samples are considered an aluminium wing flap (Figure 4a) and a GLARE 2 (GLAss REinforced aluminium laminate) panel (Figure 4b).The wing flap has impact damage located at the trailing edge (Figure 4a and Figure 7).The GLARE2 panel used is a hybrid composite made of aluminium alloy and pre-impregnated glass fiber reinforced polymer (GFRP) laminates [12,13,14] with varying thickness (Table 1).The panel has been subjected to artificial damage, consisting of 3 sets of 9 Teflon "defect" layers (located at different depths with varying diameters, see Table 1).These defects were added during the fabrication of the panel [7].
Figure 4a and 4b.Aluminium wing flap inspected with a translational system and GLARE 2 Panel inspected with industrial robot [7].

Experimental Results
Figure 6a shows a snapshot of the propagating wavefield in the GLARE 2 panel.The wave velocity increases in the thicker layers that are on top of the sample.The next step, after processing, is to visualise the thickness and the detected "defects" which show up as deviations in a local thickness map.From the results (Figure 6b), it can be observed that the "defects" with ID 1, 6 and 7 (3mm diameter) are more difficult to observe over the entire panel.The ID's, 2, 5 and 8 with 6mm diameter are more challenging to distinguish in the thicker layers.Overall, 6mm in diameter defects could be detected for thick panels, and defects as small as 3mm could be detected for thin panels.In the second experiment an aluminium wing flap is considered.In Figure 7, the impact at the trailing edge can clearly be detected as a geometrical deformation.During this experiment, three piezoelectric actuators where used at the same time (due to transmission losses between separated panels that we riveted to an underlying support structure), to receive a high quality signal across the region of interest.A detailed inspection of the results reveals debonding of the trailing edge rivets.At the leading edge also debonding is observed of the internal structure, as well as a variation in thickness (around a 1000 mm) over the entire airfoil.

Further work
To improve MRO practice, a variety of steps are considered besides using a novel detection system.Besides the used translational system, a cobot-based positional system, situational awareness, datawarehousing, certification, digital twin capabilities and automated reporting of the obtained results are currently studied.

Cobot-based positioning platform and situational awareness
Using pre-defined trajectory scripts, a cobot is currently able to perform surface scans, similar to the translational system.As a next step the cobot will be expanded with situational awareness sensors (separate from the additional sensors on the MEMS array), to prevent collisions of the sensor and or shoulder of the cobot with its surroundings (safety plane regions), as illustrated in Figure 8.The obtained dataset creates the possibility to automate the attitude of the sensor-head and keep it directed to the surface, even when double curved surfaces are considered.A pre-defined trajectory for a flat surface can also be translated to a curved surface trajectory with this surface dataset.At a later stage, a deep reinforced learning will be considered for autonomous trajectory planning to perform IOP Publishing doi:10.1088/1742-6596/2692/1/0120247 surface scans.The lidar and camera data obtained during the situational awareness scan, generates a general dataset (this will be a reference coordinate system) of the full airframe, followed by a dataset with information where the sensor is positioned for inspection.

Digital twin capabilities development
A Digital Twin (DT) might cover a different scope (broader or narrower) as compared to the MRO inspection applications that have been presented above.In this specific case, a DT is the development of a digital entity with the use of a virtual reference, that can replicate and visualise the process and results of the inspection and the respective measurements.In addition, a DT can be as detailed as needed.The granularity and accuracy can be the same as the one of the physical system at most, but also lower if this is considered sufficient.Similarly, the fidelity and the dimensions of the model can be anywhere between 2D, quasi-3D, 3D or also include the temporal element and make it dynamic.In Figure 9, the main layout of the intended DT and the model layers are illustrated.A first capability is related to data analytics and the data can be visualised in a way that assists the responsible engineers to assess the physical status of an asset.A visualisation can also help to identify related issues in the same structure, or even unnoticed fleet-wide issues.Building on that capability, a DT can have additional AI-based diagnostic and prognostic layers, that could either identify and assess specific types of damage or support any decisions regarding the potential maintenance actions [15].

Conclusion and recommendation
This paper presents a novel, non-contact inspection method for metallic and composite parts using a MEMS-sensor array to record the wavefield.The recorded wavefield is processed to provide a highresolution thickness map.The proposed NDT scanner has the potential to become an alternative to conventional inspection approaches that result in long, labour-intensive MRO processes.Results shown that the MEMS-sensor array is suited for an efficient, contactless inspection due to its ability to detect the radiated Lamb waves at a large stand-off distance from the inspection material.Further, the obtained thickness measurements and the size of the detected defects show that the proposed approach can provide comparable inspection results with the conventional techniques while expected to significantly reducing the inspection time and having the potential to automate the damage assessment processes.Finally, an equivalency process of the proposed system as compared to the certified, standard tool recommended by manufacturer and currently used in aircraft inspections is prepared.For our case, the

Figure 1 .
Figure 1.Classification of NDT techniques in groups associated to used methods (adapted from [1]).

7thFigure 5
Figure 5 shows the layering of Teflon patches inside GLARE 2 panel.It can be observed that the thickness of the sample gradually increases.

Figure 6a and 6b .
Figure 6a and 6b.Wavefield in GLARE2 panel and wavefield converted to thickness map.

Figure 7 .
Figure 7. Results of thickness measurements in aluminium aircraft wing structure.
a c tu a to r lo c a ti o n s impact damage debonded internal structure d e b o n d e d r i v e t s 7th International Conference of Engineering Against Failure Journal of Physics: Conference Series 2692 (2024) 012024

Figure 9 .
Figure 9. Layout of the intended Digital Twin system and respective use cases and applications.

Table 1 .
Layup of GLARE 2 panel and ID/diameter of artificial defects.