Parallelsession: New Technologies

Moderation: Alexander Numrich, GMAR

Talk 1: SONIC – AI-Driven Robotic Tomography for Non-Destructive Testing, Thomas Gallien, Joanneum Research Robotics

SONIC is an FFG-funded research project investigating flexible, AI-supported non-destructive testing for complex industrial components. The project combines robotic Laser Ultrasound and X-ray CT with 3D surface acquisition, automated trajectory planning, and an Inference-as-a-Service concept in which measurement data are transferred via 5G to cloud/HPC resources for reconstruction. A particular focus is the use of physics-based simulation to generate synthetic training data for surrogate models and, subsequently, generative AI reconstruction methods. The talk will present the SONIC concept, its industrial use cases, and the current development status of the simulation and data-generation pipeline.

Talk 2: Simulated First, Live Later: When AI Agents Take Control of the Machine, Michael Eberle, digifai

Digital twins have arrived in mechanical engineering. They enable machines and control software to be developed and tested in a safe virtual environment before the actual system is built and commissioned. However, creating these digital twins takes time. This is where AI agents come in: they support the creation of simulation models and significantly shorten the path to a fully functional digital twin. And then what? Another agent takes over control. First in the simulation, where mistakes come at no cost. Then on the real machine, where they certainly do. Michael Eberle from digifai demonstrates this approach using twin as an example and explains why robust, reliable simulations are essential before AI agents can be allowed anywhere near a real machine.