AirCarbon III | Development of aerospace-specific carbon fiber with associated semi-finished products

The product quality of carbon fibers is typically assessed after production. To make a statement already during production, manual visual inspections or estimations based on correlating variables have been carried out up to now. However, these can only be carried out at longer intervals and also manually.¹ Automated monitoring solutions are necessary to ensure high-quality standards during production continuously.

Particularly in carbon fibers, automated quality control is difficult due to a lack of experience and comparable measurement technology. Fraunhofer IGCV has addressed this problem in the successive projects AirCarbon II and Air Carbon III and, in collaboration with the companies SGL Carbon GmbH and Chromasens GmbH, has developed an optical monitoring solution for carbon fiber production. The solution detects protruding filaments, lint, and similar foreign bodies and distinguishes the types of defects. To reduce the data volume, the partners developed an innovative optical concept. The associated software was developed entirely by Fraunhofer IGCV team. For this purpose, image data was acquired, prepared and processed, and also classified according to anomalies. The work was supplemented by data management and a user-centered interface. All in all, an inline-capable monitoring system was created, with which carbon fiber production processes can be continuously and automatically monitored: from precursor production through the various intermediate steps (oxidized precursor, carbonized, and graphitized fibers) to the sized fiber (see Figure 1).³

The basis for the functional demonstrator is a line scan camera from Chromasens GmbH and illumination systems suitable for the application. The significantly higher resolution of the CMOS color line sensor than conventional CCD sensors makes the detection of the protruding filaments, which range from 7-12 µm in diameter, possible in the first place. Since the sensors are limited in their measurement width, a novel and patented optical concept was designed and implemented that triples the camera's field of view while keeping the data volume the same. The system was designed primarily for large-format line sensors (up to 85 mm). However, it is easily transferable to smaller sensors or area sensors.⁴

This makes the solution particularly suitable for scaling, as additional hardware costs on the camera side and for computing capacity can be avoided.

Extensive automation through autofocus

To integrate the system quickly and easily into the process, it also provides autofocus. This uses a sharp evaluation of the captured images and focuses the system fully automatically. This also allows process-related fluctuations in the object plane to be compensated for.

More than image processing

In addition to the measuring range extension, an image processing algorithm developed for fiber production was ported to a Field Programmable Gate Array (FPGA). This is because high-volume image data is captured at up to one gigabyte per second. Thanks to FGPA, they can be processed in real-time at any time.

Downstream data processing was implemented in software developed in-house by Fraunhofer IGCV, which reliably handles defect detection. With a machine learning approach and selected training data, the reliable differentiation of the detected defects is successful. Thus, not only defects on the fiber material can be reliably detected, but also a separation according to defect types can be performed automatically. This eliminates the need for manual analysis while increasing the information content. In summary, a prototype was presented with an architecture that enables simple and cost-effective scaling of the system so that production plants can be equipped with it in the future.⁵

A special aspect of the research work is the consideration of upstream and downstream processes in the environment of carbon fiber production. Thus, not only the process for the production of aerospace-grade carbon fiber is considered, but also the upstream PAN spinning. Here, the precursor for carbon fiber production is obtained from polyacrylonitrile.

Another aspect is the further processing of the rovings (fiber bundles). By spreading, a continuous, flat fiber web can be formed from several rovings. This process can also be checked online using the optical monitoring system from Optima II. This makes Optima II holistic and enables online inspection along the value chain from raw material to a textile semi-finished product.

After filtering and preprocessing, the data is stored in a local database. Fraunhofer IGCV has equipped its own vision PC with all the necessary interfaces for this purpose:

• CameraLink interface
  
• Lighting control
  
• Gig1000 Ethernet
  
• Enclosure to protect against flying fibers
  

Data storage on the local hard disk can be integrated into the company's own cloud infrastructure at any time via standard protocols. Data is analyzed using INFIMO (Inline Fiber Monitoring) software, which provides both roving-specific viewing of defects and a deeper understanding of the data. This allows the user to gain insights based on feature analysis, i.e., by size, orientation, or expression of the defects.
This way, trends and anomalies that occur during ongoing operations become visible. At the same time, data science and data mining methods can be used directly via the software. This means that a transfer to adjacent processes (CFRP production, PAN spinning, spreading) or entirely different domains can be realized quickly.⁷

In addition to defect detection based on images of the fiber surface, further data is evaluated for inline monitoring. Fraunhofer IKTS designs a sensor array based on the eddy current method, which encloses individual rovings and enables three-dimensional recording. The processed raw data is then passed to the software framework for data analysis and visualization.

The goal is a comprehensive evaluation of the roving, such as monitoring the sizing content or detecting fluff accumulation. The measuring principle is particularly suitable for inline use even at high process speeds. The possibilities of the measurement system are currently being researched in AirCarbon III, as there has been no comparable implementation and application to date. This is where Fraunhofer IGCV uses its expertise in the field of artificial intelligence and machine learning. Applied to the measurement data, the partners hope to derive new insights from the data that are not possible with conventional approaches.

Contact to the Eddy Current Methods Group of Fraunhofer ITKS

The partners and scientific participants are already in the final phase of the project. Here, tests are being carried out at the pilot plant of the partner SGL Carbon, and data is being evaluated. The developed systems are being tested and improved both in the laboratory and in the pilot plant.

Besides, further fields of technology and applications could be developed from the research activities. The following projects complement the work from AirCarbon III:

• MAI Preform 2.0
• DIDA²
• Saturn
  

In addition, findings and research results from the project are being utilized or supplemented elsewhere:

• AI in Fiber Technology
• KI Award: Best Paper

At this point, Fraunhofer IGCV is already in talks with interested parties from the industry. Are you also looking for research and development cooperation on the topic of AI in fiber technology? Would you like to know how you can use the system at your company and use it to your advantage? Please contact us!

Non-binding inquiry

[1] Geinitz, Steffen: Automatisch Fusselfrei. Carbon Composites Magazin. 1/2018.

[2] Brunsell, A. R.: Fibre reinforcements for composite materials. Composite materials series, Bd. 2. Amsterdam, New York: Elsevier 1988

[3] Geinitz, Steffen, et al.: Detection of filament misalignment in carbon fiber production using a stereovision line scan camera system. Proc. of 19th World Conference on Non-Destructive Testing. 2016.

[4] Geinitz, Steffen, et al.: Online detection and categorisation of defects along carbon fibre production using a high resolution, high width line scan vision system. Proc. of 17th European Conference on Composite Materials, ECCM 2016.

[5] Margraf, Andreas, et al.: Detection of Surface Defects on Carbon Fiber Rovings using Line Sensors and Image Processing Algorithms. SAMPE 2017.

[6] Margraf, Andreas, et al.: An Evolutionary Learning Approach to Self-configuring Image Pipelines in the Context of Carbon Fiber Fault Detection. 16th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2017.

[7] Margraf, Andreas, et al.: Towards Self-adaptive Defect Classification in Industrial Monitoring. Proc. of 9th International Conference on Data Science, Technology and Applications (DATA). 2020.