SYSIDENT

SYSIDENT - KIT124

Keywords: non-linear modelling, non-linear systems, neural networks, vibration analysis, loud-speakers, embedded systems, VLSI design, parallelisation

Start Date: 1 January 95 / Duration: 36 months

[ participants / contact]

Objectivies and Approach

In the area of non-linear system identification, research suggests that instead of the conventional method of searching for appropriate parameter dependant representations of the system, and devising algorithms to estimate relevant parameter values, on-line model structure generation methods based on collective structures, such as neural networks, show considerable promise. In this project, the group will study the feasibility of on-line structure generation, as well as its implementation in the form of parallel VLSI processors, as an approach to system identification. This methodology will be evaluated in the context of industrial prototype problems such as non-linear modelling of loudspeakers and non-linear vibration analysis.

Progress and Results

Preliminary research has been conducted on neural network modelling with radial basic functions, wavelet neural networks, and fuzzy neural networks. The results of this work, which have been presented at conferences or in scientific journals, can be obtained from the above mentioned contact point.

Coordinator

Katholieke Universiteit Leuven
Electrical Engineering Department
Kardinaal Mercierlaan 94
B-3001 Leuven, B

EU Partners

Katholieke Universiteit Leuven, B
Delft University of Technology, NL

Non-EU Partners

National University of Singapore, SG

CONTACT POINT

Prof. J. Wanderwalle
Tel: +32 16 32 17 09
Fax: +32 16 32 19 86
E-mail: joos.vandewalle@esat.kuleuven.ac.be