ÂÌñÏ×ÆÞ

The complexity of composite structures for aircraft/aerospace is increasing, says Nicolas Colin, technical expert in nondestructive testing (NDT) for (Bouguenais, France), an industrial collaborative research center with composites expertise. “Traditionally, NDT machines used in composites production operations are large, specialist machines that are not easily adaptable to complex geometries. Instead, they tend to be C-scan systems using ultrasound testing [UT] heads on gantry or dual robot systems with water squirters that require a significant concrete foundation, large footprint in the factory and a lot of water, up to 100 liters/minute.”

In 2019, IRT Jules Verne started exploring how to reduce cost and increase flexibility in composites NDT by applying automation. “Automation has existed,” notes Colin, “but using older technology. Airbus asked if we could develop an NDT system that is more flexible. So, we built a consortium with Airbus, its NDT company , Tier 1 composite aerostructures supplier , which is located next door to us, as well as the robotics integrator [Aizenay] and the , which has expertise in UT inspection and simulation.”

The resulting Flexible & Automated NDT PlatfOrm for Manufacturing (FANTOM) project ran from January 2022 to December 2024 and resulted in a prototype demonstrator that combines an autonomous mobile robot (AMR) and cobot with a Creaform (Lévis, Quebec, Canada) geometric scanner, visual inspection cameras and UT hardware and software to meet a variety of inspection requirements from Airbus and Daher using actual carbon fiber-reinforced polymer (CFRP) parts.

FANTOM project roadmap

IRT Jules Verne is known for leading collaborative projects, says Aurélien Lunion, project leader for FANTOM at IRT Jules Verne. “We also have a lot of know-how in our five areas of expertise.” These include:

“We used these multiple sets of expertise and multiple in-house teams,” says Lunion, “as well as our project partners in FANTOM to develop an automated robotic inspection demonstrator that could fulfill the requirements for CFRP structures that Daher and Airbus gave to us.”

He explains that IRT sees inspection moving away from fixed stations for specific parts. These use thousands of liters of water and cost €1-5 million per system, depending on the size of structure to be inspected. Instead, the vision is to use more inline sensing to ensure quality and predict performance. Such systems would integrate different inspection methods into a simple platform that reduces manual operations, minimizes use of floor space and water usage and reduces cost versus traditional systems.

“During FANTOM, we demonstrated a mobile system that implemented geometric inspection using Creaform, visual inspection using cameras and UT scans using a low water usage spraying system for coupling with the part surface,” says Lunion. “We proved that we can use less manpower and achieve more inspection thanks to a smart way of performing the UT scanning, and we need much less space because the platform moves to the structure. We minimized water consumption by using only mist while significantly reducing the system cost — the prototype unit cost less than €1 million.”

This system was demonstrated on two CFRP use cases: An Airbus A350 wing box front spar measuring 6 × 2.4 meters × 10 millimeters with ply drop-offs and a falling edge, while Daher’s use case was a 2 × 0.4-meter × 10-millimeter beam with integrated stiffener, thickness variations and restricted access.

Prototype NDT system

IRT Jules Verne worked with Axiome to design the robotic platform and then tested its mobility, says Colin. “Using an AMR and a cobot are not revolutionary concepts, but they enable assembling the NDT and robotic systems in a way that is fully integrated and can precisely localize inspection with the cobot end effector.” IRT Jules Verne has a patent pending on this system and inspection technique.

“We can localize inspection by combining position data from the AMR, cobot arm and the C-track on the Creaform system that registers where the end effector is with high precision,” Colin adds.

Creaform is a company that develops advanced metrology solutions. Its R-series products are made for mounting onto robots and cobots. The system used by IRT Jules Verne includes a MetraSCAN head mounted onto the cobot and a C-Track optical tracker. The C-Track has two infrared cameras that emit pulses of infrared light that illuminate reflector targets on the MetraSCAN head and the object being scanned. The cameras track these reflectors to determine their exact position in space, creating a precise 3D reference frame for localizing the inspection area. The MetraSCAN then uses blue laser triangulation to capture the object’s 3D surface data within the tracked volume, providing high-accuracy, real-time measurements that are immune to vibrations and environmental changes. This allows the C-Track to continuously track the position of the scanning system and the object, enabling the 3D scanner to capture accurate measurements even as the cobot arm and AMR move.

Inspection process flow

As shown in the process flow diagram above, the Creaform system was used to complete the first three steps in the robotic NDT sequence:

• Acquisition of the actual geometry for the part being inspected
• Digital reconstruction of the surface
• Generation of trajectories for the cobot inspection.

This was key, notes Lunion, because it also enables optimization of the inspection trajectories. “Airbus has large parts which can bend due to their own weight,” he explains. “Paths generated from CAD files for these parts are not always accurate, which can cause a lack of coupling during the UT scan so that you don’t acquire NDT data in those locations.”

“We first needed to establish the correct geometry of the part to enable capturing all the necessary data during inspection,” Lunion continues. “So, we first perform a 3D scan of the part using Creaform and then generate the trajectory for the UT scan from that 3D data.”

“Once we have the 3D mesh [obtained with the Creaform system software], we load it in our trajectory planner [in-house software] to generate trajectories for the UT scan,” says Colin. “We then perform the UT inspection.” The team used CIVA software developed by CEA List and distributed by  (Massy, France) to simulate the UT inspection, modeling the propagation of the UT waves through the part. “Traditionally, this physics-based modeling was successfully carried out for metals and CFRP applications, but with limited validity range due to the assumptions of the ray-tracing models,” he notes. “For this project, the CEA List used a specific version of CIVA that incorporated finite element methods [FEM]. This enabled us to develop and optimize a phased array probe design using focal laws that are adapted to the part material and geometry.”

UT inspection

UT inspection typically uses water to achieve coupling between the ultrasound waves and the inspected part. The water fills any air gaps and enables ultrasonic waves emitted from the transducer to travel into the composite part and back with minimal reflection and loss of energy. However, water is an issue, explains Colin, “because it can migrate into holes and unfinished edges, so the parts must be prepared with caps to prevent this, which consumes time and money.”

IRT Jules Verne sought to minimize water use and storage. Iterations of the UT phased array probe shown at right used small misters on either side of the phased array UT probe. “We used a membrane that can also include a kind of water pocket to supply water, a solution provided by the company” says Colin.

“We needed only 1 liter of water to inspect one side of the Airbus wing box spar which had a surface area of 12-15 square meters,” he continues. “We typically use between 0.5 and 1 liter of water during a 2-hour inspection. This is acceptable to Airbus, especially compared to traditional UT systems with squirters which require at least 100 liters/minute, though in those systems, much of the water is recycled rather than lost.”

“We used a rigid UT end effector, but we also developed a compliant end effector with springs on it so that the blue membrane with the water mist is always perpendicular to the surface,” notes Colin. “We would lose some signal especially at ply drop-offs, but we had no signal loss with the rigid end effector. However, for this, you must be very accurate in terms of the end effector orientation and trajectory. For example, even when the probe is rotated just ±1° off perpendicular, you lose 2 decibels of the UT signal.”

“During the project, the robotics team determined the right trajectory, and they were accurate enough to use the rigid end effector,” says Colin. “The Creaform system tracks where the UT array is and that’s how we follow the trajectories.” It also helped deal with the part’s geometrical complexities. “We could see the ply drop-offs, for example, and the robotic arm followed these, maintaining the probe orthogonal to the part surface. Inspection of edges are also an issue, which are usually done manually. In our setup, the design of the end effector, a rigid structure with a flexible membrane and a linear array, allows the probe to scan along the edges effectively. The membrane of the end effector is pressed partly on the part and partly over the surrounding space, allowing the probe to maintain contact and acquire ultrasonic signals even along complex boundaries.”

Visual inspection

In addition to 3D scan and UT end effectors, the CASPR demonstrator was developed to use a third type of end effector. “During commercial aircraft manufacturing and operation, 70% of inspection is visual,” says Colin. “Airbus and Daher aim to use automation to achieve significant cost reduction in these operations and are also interested in AI defect detection.”

To achieve this, FANTOM developed two different vision end effectors for visual inspection. The first was designed for scanning flat or gently curved surfaces, covering relatively large areas with adjustable lighting modules that can vary in intensity, wavelength and orientation. A smaller, more compact end effector was developed to access confined or complex geometries such as concave regions, internal radii and corners, enabling inspection of areas that are typically difficult to reach.

“In addition to the 3D scan of the part, the visual inspection end effectors systems also provide information about scratches, bumps and dents in part surface,” says Lunion. “Airbus and Daher specified four types of defects that must be detected: scratch, dent, FOD and flaking. We took more than 100 examples of defects on structures from Airbus and used machine learning to teach the system to recognize and classify the defects.”

Mapping inspection data

One of the key challenges in FANTOM, notes Lunion, was accurately linking the position of each inspection end effector with the NDT data it collected. “Precise position information is essential to generate reliable cartographies,” he adds.

FANTOM partner Testia led this work. Lunion explains that the Creaform data gave X, Y, Z (3D space) and A, B, C (rotation) of the tool center point (TCP) for both the visual and ultrasonic sensors. “They achieved a high-frequency, precise measurement of the TCP position. This data was then combined with the data collected by the UT and camera systems.”

For FANTOM, Testia built upon Airbus’ existing NDT Kit UT software, originally developed for postprocessing and visualization of ultrasonic data. “They extended the software so that visual inspection images can also be stored in the NK3 file format, which was previously used only for 3D ultrasonic data,” says Colin. “This allows 3D positional data, UT cartographies and visual inspection results to be combined within the same software environment, enabling operators to view and analyze all relevant inspection data in one place and streamline the postprocessing workflow.”

“Today, we can see all sets of data simultaneously,” he continues, “but they are separate.” The image at right shows the UT data on the left and visual inspection data on the right. “In the future, we aim to generate a single, fused cartography, where each point represents the probability of defects based on all inspection methods. This would integrate the raw cartographies from ultrasonic, visual and 3D inspection into one fused map. The separate data would still be available for more detail, but this type of cartography would enable the NDT operator to move quickly and focus only on high-probability areas.”

Future developments

Although the FANTOM project ended at the end of 2024, IRT Jules Verne is continuing to explore further developments. “We achieved the objective Airbus and DAHER set for us: developing a mobile automated NDT prototype capable of performing the most common inspections on CFRP structures with reduced factory space, foundation requirements, cost and manual labor,” says Colin. “The system is still a prototype and requires further development to fully meet the needs of industrial partners. Current work focuses on improving the robotized inspection end effectors for all types of NDT methods, making the system easier to operate for non-robotics specialists, and leveraging its modular design so partners can select only the technological components they want to integrate into their own factories.”

“We also see a possibility to improve defect visualization through augmented reality [AR],” he continues. He notes that CAD/CAM and digital twin software provider Dassault Systèmes (Vélizy-Villacoublay, France) acquired the company (Paris, France) in 2022. Dassault Systèmes is integrating Diota’s solutions into its software to enhance the use of digitized processes and digital mock-ups for manufacturing by connecting virtual twins with real-world data in the field. “This illustrates our vision: The UT and visual inspection data we collect, combined with positional information and stored in .NKD files, can be projected onto a structure via AR. When moving an AR tablet or goggles, operators can instantly locate defects and their associated metadata, focusing on areas that require further inspection. This is particularly valuable for guiding them to perform targeted UT scans at precise defect locations, something that can be difficult when defects are only visible on conventional 2D maps, making the process more efficient.”