Einladung zum Vortrag im Kolloquium Technische Kybernetik

Capturing dominant nonlinearities in partly unknown systems

Prof. Luigi del Re

Institut für Design und Regelung mechatronischer Systeme
Johannes Kepler Universität
Linz, Österreich

    Zeit: Donnerstag · 16. 6. 2005 · 14:00 Uhr
    Ort: Raum 3.241 · Pfaffenwaldring 9 · Campus Stuttgart-Vaihingen

Abstract

A good description of nonlinear systems is central to many control and supervision problems. Increasing performance requirements on the final product are often related to increasing requirements in terms of nonlinear models. This again means that less known or even unknown dynamical phenomena must be described. Traditionally, this kind of problems has been tackled by using universal approximators, like the ubiquitous neural networks. While this approach works quite nicely if some rather restrictive conditions are satisfied, its limits are evident as soon as prediction requirements are stated.

In this talk we shall discuss two possible ways recently examined at the Institute of Design and Control of Mechatronical Systems of the Johannes Kepler University of Linz, namely special classes of universal approximators as well as direct structure identification. While the results of the first approach are essentially of theoretical nature and applications are just being started, the second method, which strongly relies on a search in the function space using genetic programming, is already yielding excellent results in the field of virtual sensors for engines.

Biographical Information

Luigi del Re is full professor and head of the Institute for Design and Control of Mechatronical Systems at the Johannes Kepler University Linz. He holds an MS in electrical engineering and a PhD at the mechanical faculty in automatic control of the ETH Zurich. He worked as project manager in South America, was then research associate at the Institute of Automatic Control of the ETH and later Head of Hybrid Control in the Swatch Group. His main interests concern applied control systems, in particular approximate methods, as well as identification and fault detection, with special applications in the field of engine and vehicle technology, hydraulics, process control and biomedicine.