Contributions to Control Systems in the summer semester 2019

 

In order to support the professional exchange in control and automation technology far beyond the borders of our university, some years ago a colloquium series with the name “Contributions to Control Systems” was brought into being at the Institute of Automatic Control. Speakers from industry and research talk about their work in relevant developments and applications in the field of automatic control.

 

All parties interested are very welcome.

We would be happy to add you to our corresponding email list. Please contact our office to be added.

Venue

Conference room of the IRT,Campus-Boulevard 30, 52074 Aachen
1.OG, Room 052.G

Monday, May 13, 2019, 4:00pm

 

Model-based control of dedicated hybrid drivetrains

Johannes Rumetshofer

Technische Universität Graz
Institut für Regelungs- und Automatisierungstechnik
Germany

Abstrakt:

Dedicated hybrid drivetrains are expected to be a major mid-term competitor to zero-emission drivetrain concepts, considering the trends in automotive emission regulations. The research presented in this talk addresses potential for further improvement of drivability and efficiency of dedicated hybrid drivetrains with a single electric machine in gear shifting. A concept for active control of smooth and lossless gear shifting is introduced, which resolves the trade-off between dissipation in clutches and propulsion torque interruptions. A model-based control strategy, which applies this concept in a drivetrain control system, is proposed. In order to support usability of this model-based strategy a generic and modular modeling approach for drivetrain mechanics, applicable to all common geared drivetrains, including combined planetary gear sets, is presented. In combination with a proposed model-based drivetrain analysis method the presented research supports the development process of automotive drivetrains in general and of dedicated hybrid drivetrains in particular.

 

Friday, June 07, 2019, 11:00am

 

Machine Learning “in the wild”

Barbara Hammer

Technischen Fakultät der Universität Bielefeld
Lehrstuhl für Machine Learning

Germany

Abstrakt:

Machine learning technologies, in particular deep learning, has revolutionised domains such as vision or language processing and it is included in everyday’s consumer products such as natural language interfaces of smart phones. Classical machine learning can be substantiated by strong guarantees as offered by statistics and computational learning theory. When facing real-life problems in data analysis or industrial processing, however, quite a few challenges arise, which go beyond the classical learning setting.

Within the talk, I want to address three challenges of machine learning “in the wild”:

• How to learn in the presence of few samples only?
• How to learn in non-stationary environments where drift might occur?
• How to enhance machine learning models by an explicit reject option when the output class is not clear?

I will present approaches how to address these problems, which are based on distance-based and prototype-based models, and I will explain exemplary applications from the domain of driver assistance, biomechanics, and bioinformatics.

 

Thursday, July 11, 2019, 9:00am

 

Kombinierte Pfadplanung und prädiktive Regelung für den automatisierten und autonomen Schiffsbetrieb

Max Lutz

Christian-Albrechts-Universität zu Kiel
Lehrstuhl für Regelungstechnik
Germany

Abstrakt:

Eine zunehmende Automatisierung des Schiffsverkehrs kann Sicherheit und Warendurchsatz erhöhen und in Verbindung mit Verfahren der Regelungstechnik und Online-Optimierung zur Realisierung energie-, verbrauchs- oder zeitoptimierter Routen beitragen.
Im Vortrag werden die kombinierte Pfadplanung und prädiktive Regelung für den Betrieb und das Manövrieren von (autonomen) Schiffen in ruhigen Gewässern, wie beispielsweise Häfen, vorgestellt. Basierend auf einem mathematischen Modell der Schiffsdynamik wird ein dynami-sches Optimierungsproblem formuliert und gelöst. Zusätzlich zu den Stellgrößenbeschrän-kungen für den Antrieb und das Ruder werden statische und sich dynamisch bewegende Hin-dernisse explizit berücksichtigt. Um in Echtzeit auf Veränderungen in der Umgebung reagie-ren zu können wird das optimale Steuerungsproblem auf einem wandernden Zeithorizont ge-löst. Dieser modellprädiktive Ansatz erlaubt es, gezielt Online-Informationen in die Lösung des Führungs- und Regelungsproblems einzubeziehen, um einen automatisierten bzw. auto-nomen Betrieb zu ermöglichen.

 

Friday, July 12, 2019, 10:00am

 

Mikrosensoren für die Digitalisierung in der industriellen Fertigung

Gerd vom Bögel

Fraunhofer-Institut für Mikroelektronische Schaltungen und Systeme
Geschäftsfeld „Wireless & Transponder Systems“
Duisburg
Germany