Research Seminar on AI: Improving closed-loop performance by combining process knowledge and process data

 

Tuesday, February 28, 2023, 4:00pm

Model predictive control is an advanced control method used in fields such as process control, automotive systems, and robotics. It relies on underlying process models for the control and enables the operation of such systems at their limits, while respecting existing constraints. While identifying models for nonlinear systems is complex and time-consuming, it is possible to use approximate models such as linear models or white-box models derived from first principles. On the other hand, data-driven models or black-box models can be created by exploiting measurement data and have the potential to describe most dynamic processes accurately. However, black-box models lack physical interpretability. To address this, a grey-box model approach is adopted, which combines the advantages of both white-box and black-box models. An LSTM, which is a type of recurrent neural network, is particularly suitable for MPC due to its ability to process entire sequences of data, making it well-suited for making predictions based on time series data. However, like other black-box models, it lacks physical interpretability.

Dominik Scheurenberg is currently a research associate with Prof. Dr.-Ing. Abel at the Institute of Automatic Control (IRT). His research is focused on grey-box modeling of dynamic systems, as well as the combination of Model Predictive Control with such models. Additionally, he is focused on extending industrial plants to cyber-physical production systems. He joined the Institute of Automatic Control  as a PhD student in 2019. Before his PhD, he received an M.Sc. in Mechanical Engineering, a B.Sc. in Electrical Engineering and Information Technology both from RWTH Aachen University and a B.Sc. in Biomimetics from Westphalian University of Applied Sciences.