Description

The development of new production processes or the start-up phases of new products often involve high development costs and long development times, which have a strong negative impact on profitability. In order to drastically reduce these costs and significantly accelerate time-to-market, partners from the Fraunhofer IPT and Fraunhofer ILT have developed an approach for achieving first-time-right production – producing high-quality products right from the first job. With the help of fast and precise process simulations, multi-criteria Bayesian optimization and smart system engineering, we show that processes can be optimized efficiently and quickly in simulation and enable a successful transfer to the real process, so that only a few optimization experiments are required for fine-tuning on the actual machine.

The optimization system was developed for two applications, the molding of precision optics made of glass and laser ablation and enables both human-in-the-loop optimization and fully autonomous closed-loop optimization. In precision glass molding, process design is usually time-consuming and involves several iterative procedures in simulation as well as real experiments. Our results show that by utilizing Bayesian optimization in a human-in-the-loop approach, with just a few iterations, process improvements can be achieved that also enable to uncover process insights to be fed back into the optimization. Our closed-loop optimization mode for laser ablation further enables large parameter studies allowing to quickly identify optimal parameters from a broad design space.

Our further research will focus on integrating process expertise into the optimization protocol, on increasing comprehensibility and user acceptance, and finally on integrating the optimization system into automated machine control.