|4. Juli 2023
|Download als iCal:
Institute for Systems Theory and Automatic Control
University of Stuttgart
Tuesday 2023-07-04 4 p.m.
IST Seminar Room 2.255 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen
In this talk, we consider data-driven analysis as well as prediction-based design of aperiodic sampling in control systems. For analysis, the aperiodicity of the sampling instants is seen as an undesired phenomenon, and it is necessary to ensure that stability and a certain performance level are retained. This situation emerges for instance in Networked Control Systems. We provide methods to analyze stability of control systems under arbitrary aperiodic sampling directly from recorded data, without requiring any model knowledge. For design, the aperiodicity seen as a degree of freedom, and the sampling pattern must be shaped so that control goals are satisfied with a minimum amount of samples or with maximum performance. Conventional methods for this purpose, for instance event-triggered control (ETC), suffer from a lacking predictability of the resulting sampling pattern and often do not come with guarantees on sample efficiency. Thus, we propose a prediction-based approach for design of the aperiodic sampling pattern, called rollout ETC, which allows for predictability of the required resources and a guarantee of improved sample efficiency compared to periodic sampling. In this talk, we give an overview over both topics and present some insights on the given theoretical guarantees. The efficacy of the developed methods is demonstrated by numerical examples.
Stefan Wildhagen received the Master’s degree in Engineering Cybernetics from the University of Stuttgart, Germany, in 2018. He has since been a doctoral student at the Institute for Systems Theory and Automatic Control under supervision of Prof. Allgöwer and a member of the Graduate School Simulation Technology at the University of Stuttgart. His research interests are in the area of Networked Control Systems, with a focus on optimization-based scheduling and control as well as on data-driven methods.