BUP69 - PReSeLA

Physics-informed Reinforcement learning for Self-Learning Laser Ablation

The production of future mobility components requires sub-micrometer precision. Ultrafast lasers are the key technology for such high-precision manufacturing. However, even minor variations in material properties can lead to change in process behavior and quality defects. Currently, human experts must manually recalibrate the process parameters for new material batches. This rigid process slows down manufacturing and prevents highly flexible production. Self-learning systems enable automated and rapid adaptation to new boundary conditions.

Goal

The aim of this project is to develop a self-learning algorithm as a foundation for autonomous and adaptive laser-based production systems. Using machine learning algorithms, the system recognizes quality deviations and autonomously adapts the manufacturing parameters. This paves the way for the implementation of “Manufacturing Skills” for laser ablation-based manufacturing processes in skill-based manufacturing systems of the future.

Approach

Physics-Informed Reinforcement Learning (PI-RL) to enable rapid and safe adaptation of process parameter.
Integrated sensor systems to monitor and evaluate quality features immediately after the ablation process.
Feedback signal to the PI-RL agent to learn the relationship between manufacturing parameters and quality based on the acquired data.
Autonomous adjustment of key process parameters (pulse energy, repetition rate, and scanning speed) to maintain quality requirements.

Benefit

For production: Increased manufacturing agility through fast, automated, and autonomous adaptation of the production machine to new boundary conditions. Cost reduction due to less scrap and lower requirements for operator expertise.
For the future vehicle owner: Components can be produced more reliably and at a lower cost.