CENAERO
Hybrid Modelling Methods towards an Augmented Engineering: The Use Case of AI-enabled Additive Manufacturing

Embedded AI

Aim

The aim is to develop hybrid modelling methods for industrial products/processes combining physics/knowledge and machine learning to develop models that are more reliable and explainable notably in production processes using additive manufacturing.

What's at stake

In past decades, Additive Manufacturing (AM), also known as 3D printing, has revolutionised the manufacturing sector. While the traditional production methods removed material through machining, carving, shaping etc., 3D printing does the opposite: producing by adding material. This approach offers an extraordinary freedom of design while offering the potential of (nearly) zero-waste manufacturing solution. AM serves industries in domains as diverse as aerospace, defence, or health, of particular strategic importance in Wallonia. Yet its adoption by the industry is not widespread because it involves numerous and complex parameters to be monitored and controlled during the production process to achieve an acceptable level of accuracy and quality. Introducing AI could potentially address this complexity and improve the efficiency of the printing, thus fostering the adoption of AM to reach its full industrial potential.

Challenges

There are two types of challenges: the complexity of the additive manufacturing process (described above) and the challenge of designing AI solutions to overcome this complexity. AI solutions can only be successful if they jointly integrate different sources of complex data relevant to the 3D printing process: numerical simulations, experimental/monitoring data and potentially prior human/expert knowledge.

AI possible solutions

A number of AI solutions will contribute to overcome the challenges described above:

  • Merge/fuse heterogeneous data (experimental data, simulated data of different levels of fidelity) to design efficient hybrid systems and reduce the amount of data required, towards Digital twins, i.e. digital up-to-date and accurate copy of the properties and states of the physical object, including its shape, position, state and movement;
  • Hybridize engineering/expert knowledge with machine learning via Informed/Guided Neural Networks
  • Use explainable AI to expose the correlations between process, properties and performances;
  • Combine reinforcement learning with human feedback to optimize process control

Efficient application of AI/Machine Learning to industrial processes requires close cooperation between experts in computer science and in manufacturing. This is an exciting challenge for our AI researchers.

Key AI tech topics

  • AI-based manufacturing models using digital twins
  • Object, feature and action recognition in manufacturing settings
  • Few sample learning (e.g. for defects)
  • Physics informed neural networks
  • Industrial IoT

Get in touch

Caroline Sainvitu (caroline.sainvitu@cenaero.be)