AI and automation, Part 1: The Avatar project, or how process monitoring got to be so smart…

Avatar is supported by Coriolis Composites and GeM, the Research Institute in Civil and Mechanical Engineering. GeM is a joint research unit composed of Nantes University, Centrale Nantes engineering school and CNRS, which is a partner.

Thanks to the advances made on calculation engines, it is now easier to develop AI-based solutions and automation concepts that make it possible to use raw materials more efficiently, therefore generating less waste during the production of composite parts. “Within the framework of the Avatar project, we designed a standalone box that was tested in real conditions in an industrial setting. It picks up on defects and alerts operator in real time, meaning shorter cycle times and lower costs for the production facility,” said Elena Syerko, research engineer at GeM. The device processes information transmitted continually in real time by the physical sensors placed at points along the production chain. Data is processed per precalculated metamodels, based on physical models that represent every step of production (preforming, injection, curing, etc.). Depending on the flow of information and the models, the criticality of a defect can be evaluated in real time on the production line.

Virtualising production process reality through digital calculations
Implementing AI-based solutions requires that large volumes of reliable, representative production data be available for quality machine learning. Rolling out AI-based solutions based on real physical models from production is not always an easy matter. “Collaborations between the academic and industrial worlds often run into the issue of production data confidentiality. However, there is an alternative – which is to train the engine on a digital twin, whose results can be analysed and exploited internally,” Syerko said.

To date, two sources of reliable data exist. On one hand, there are the measurements taken directly from the physical processes, which assumes having appropriate instrumentation and automation. On the other, there are synthetic data generated digitally, based on rules of physics that are assumed to be true. It is the latter that are used as input for the digital twin. “In the first case, care must be taken not to over-instrumentalise, since manufacturing processes are not well-suited to this,” Christophe Binetruy, professor at École Centrale de Nantes engineering school and GeM member.

Interaction and open innovation bring these technologies to SME & SMI
One of the main hardships encountered by research teams is related to the fact that process simulation software lag behind the technologies that companies are in the process of developing. While manufacturers who lead the way in their respective industries are familiar with these solutions, the vast majority of SME & SMI in the composites sector are still quite a way away from implementing them. Typical reasons for this gap include financial resources and human resources. “In order to make these solutions affordable for the widest audience, we need integrators,” Binetruy said. At Nantes University and Centrale Nantes, we steward creation of start-ups capable of navigating this twofold position of being present both in research and in technology transfer to industry.

But this position, with one foot in the academic world, and the other in French institutes for research and technology or the National Composites Centre (NCC) – which have similar approaches – is not a comfortable one. Technology evolves very quickly, and one must grab hold of the best option on the market, bring it to maturity, and then make it available to customers. A good way to cut out the slack periods is to turn to open innovation. “When the needs are the same and there’s no pressure to immediately bring a solution to market, the rationale of sharing innovation practices makes sense. Even if it means adapting results to the specific needs of the company later down the line, as well as to its commercial and industrial strategy,” said Binetruy.