Conventionally, industrial processes are controlled using sensor information and attempting to hold the process variables constant. Nowadays, however, society seeks high-quality products and high-performance materials. The tolerances are tightened, driving the development of new materials and processes with reduced processability windows. Thus, the conventional industrial automation approach to process design and control needs to be reconsidered, but the attempts are inhibited by poor quality or missing sensor data, e.g. due to failed or faulty instruments or wrong manual inputs. Analyses of process performance are therefore executed rarely and can be unsystematic. Results are rarely incorporated into comprehensive process models and almost never implemented as operational tools. This in turn leads to poor financial gains and slow rates of process and financial improvements, characteristic of several European industries today.
Summing up, industrial automation of complex scenarios needs a systematic integration of domain knowledge with self-learning capabilities for advanced process analytics and customized actuation for reliable automated manipulation. For achieving this ambition, the research group is involved in the development of the following key technologies:
- Digital twins and virtualized production for process understanding and optimization
- Advanced sensing, actuation, and cyber-physical systems are capable to operate in a harsh environment
- Lifelong and self-explainable machine learning (ML) for self-adaptive analytics and decision-making
- Collaborative robotics smart actuation and safe human-machine-robot interaction
