Ph.D. received on: 8/7/1997

E-mail: savaresi@elet.polimi.it

Tutor: Prof. S. Bittanti, Politecnico di Milano
 
 

 ___________________________________________________________________________________________________________

Parametric identification and control of non-linear systems   ___________________________________________________________________________________________________________

Advisor:

Prof. G. Guardabassi, Politecnico di Milano

Referee:

Prof. A. Isidori, Università “La Sapienza” di Roma

Summary:

The identification and controller design of non-linear plants play a crucial role in the modern control system theory, since, in practice, most of the real plants show an intrinsic non-linear behavior.
Among the various methods for the modelling and control of non-linear plants, those based on simple I/O measurements received a great deal of interest in the last decade. Such techniques typically make use of multi-purpose non-linear parametric functions (polynomials, neural networks, fuzzy sets, wavelets, etc.), which are designed by optimization of suitable data-based performance indices.
The main goal of the work is the development and the analysis of new techniques for the design of non-linear parametric controllers. All the techniques presented in the work are based on the model-reference paradigm; in particular, they have been used for the solution (in an approximate fashion) of classical problems of feedback linearization.
The analysis of the proposed techniques is based both on theoretical analysis and on their application to classical control problems.
In the work it is also presented and analyzed a new class of parametric non-linear functions; their most appealing feature is the fact that they can be easily constrained to be invertible (with respect to one input), by solving a simple Quadratic Programming problem. Thanks to this feature, such a class of non-linear approximators is well-suited for data-based non-linear control system design

 
MINORS

Identification of hybrid filters for energy estimation in nuclear spectroscopy

Advisor: Prof. E. Gatti, Politecnico di Milano

In many application of nuclear spectroscopy (e.g. the LCH experiment at CERN) an accurate energy measurement is of paramount importance; this is classically done by using analog filters of hybrid, but can also be done by analog-digital (hybrid) filters. In the latter case, a crucial problem is given by the accurate estimation of the poles position of the analog part of the filter, in order to obtain an accurate shaping of the overall impulse response. In this work the problem of the identification of the singularities of an analog filter for nuclear spectroscopy, starting from a single noisy measure of its impulse response, has been studied. In particular, an innovative technique based on State-Space Subspace System Identification is proposed. Using such a technique, encouraging results have been obtained.
 
On the design of a ride control system for the attenuation of the cobblestone effect on Surface Effect Ships

Advisor: Prof. Sergio Bittanti, Politecnico di Milano.

Surface Effect Ships (high speed air cushion catamarans) are characterized by a peculiar vibration problem, called “cobblestone effect”, which is due to the interaction (resonance) between the air cushion and the incoming waves. Ironically, this phenomenon is more evident at low sea states.
In this work many different aspects and sub-problems have been analyzed. In particular, a sophisticated simulator of the ship has been built, and an innovative strategy for the efficient employment of the actuators (fans and discharging valves) have been proposed. Moreover, the final control law we’ve proposed is based on simple proportional controllers, but is characterized by a non-trivial strategy for the decoupling of the vertical and longitudinal resonance modes.

 _______________________________________