We will develop a physics informed supervisory control system for robotic welding: A machine learning component will suggest on-the-fly corrective measures to the welding robot based on sophisticated physics simulations and live sensor data. By using inverse modelling in a scalable cluster of cloud computers, we will generate a real time digital twin of the currently welded joint. Validated microstructure models for duplex steels will be used together with a finite element simulation of the thermal transport in the macroscopic weld joint geometry. This multi-scale simulation will inform the robot about the integrity of the weld in real time, so that the robot can take corrective measures if needed. Welmax makes flexible robot solutions that employ laser scanning of the weld groove and the joint geometry to adapt the robot programming to the actual task at hand. The integration of IFE's cloud-based welding simulation software "Weldsim as a Service" with Welmax's adaptive robot welding solutions, will give increased weld consistency over an even wider range of geometric variations, materials and process environments. We focus on duplex steels, as this is an advanced material which is difficult to weld due to a microstructure which is highly dependent on the temperature history. IFE will improve the mathematical models for these steels, and Sintef Manufacturing will contribute with experiments and characterization to validate the models. Both the market for advanced materials and for adaptive robotics are growing, and this project aims for value creation directly in the intersection of these markets. For this reason we expect a tremendous potential for economic gains, provided that the project succeeds.
Project leader: Øyvind Jensen