Date of final exam: 11/02/1999

E-mail: previdi@elet.polimi.it

Tutor: Prof.  S. Bittanti, Politecnico di Milano

  ___________________________________________________________________________________________________________

Identification and Control with Local Linear Models   ___________________________________________________________________________________________________________

Advisor:

Prof.  S. Bittanti, Politecnico di Milano

Summary of the thesis

The problem of identification of NARX models using local ARX models is addressed. Also, the development of a nonlinear predictive control strategy, based on this model family, is considered. This models will be used in the context of black-box identification for control purposes; in particular, solutions will be proposed to the problem of identification and control of Functional Electrical Stimulation of muscles in the rehabilitation of paraplegic patients.
In this Thesis, the partition of the operating regime is done by means of Gaussian and super-Gaussian bells. The effect of normalization is investigated and the benefits introduced by the use of high order super-Gaussian bells has been pointed out. With regard to Multiple Model based control, a strategy based on local Generalized Minimum Variance design has been introduced, the links with classical gain scheduling control has been analyzed and the properties of the time-varying operators involved are finally outlined.
Briefly, the main contribution in this Thesis are the following:

• an analysis of the role played by the validity functions in Local Model Network based identification;
• the introduction of a new family of validity functions, the generalized Gaussian or super-Gaussian functions, and the analysis of their benefits on the identification procedure and the control design;
• a discrete-time velocity algorithm for gain scheduling control;
• a new algorithm for structural identification of NARX polynomial models;
• a nonlinear predictive control strategy based on Local Model Networks and the analysis of the time-varying properties of the system;
• a nonlinearly time-varying model family and a structured algorithm for the estimation of its parameters by data.
• The development of black-box models based on Local Model Network and of nonlinear controllers useful for the realization of closed loop neuroprostheses for paraplegic patients.

 
 

MINORS

Identification of urban drainage network rainfall-runoff black-box models

Advisor: Prof. A. Paoletti, Dipartimento di Ingeneria Idraulica, Ambientale e del Rilevamento, Politecnico di Milano

The calibration of conceptual models for the design of urban drainage networks is an important and well known problem in hydraulic engineering. The problem has been analysed and the use of black-box identification methods has been proposed and applied to experimental data. Both linear (ARX-OE PEM and subspace methods) and nonlinear (polynomial and neural NARX) models have been considered and their performance in the simulation and prediction of the network flow from rainfall measurements has been evaluated.
Moreover, squared coherency based nonlinearity tests have been used in order to classify the permeability of the catchments. Finally, an anambiguous method for the direct tuning of classical conceptual models from data has been given.
 

Yeast cell metabolism investigated by CO2 production and soft X-ray irradiation

Advisor: Prof. A. De Silvestri, Dipartimento di Fisica, Politecnico di Milano

Co-advisor: Prof. D. Batani, Dipartimento di Fisica "G. Occhialini", Universita' di Milano-Bicocca

Results obtained using a new technique for studying cell metabolism are presented. The technique, consisting in CO2 production monitoring, has been applied to Saccharomyces cerevisiae yeast cells. Also the cells were irradiated using the soft X-ray laser-plasma source at Rutherford Appleton Laboratory (UK) with the aim of producing a damage of metabolic processes at the wall level, responsible for fermentation, without great interference with respiration, taking place in mitochondria, and DNA activity. The source was calibrated with PIN diodes and X-ray spectrometers and used Teflon stripes as target, emitting X-rays at about 0.9 keV, with a very low penetration in biological material. X-ray doses delivered to the different cell compartments were calculated following a Lambert-Bouguet-Beer law. Immediately after irradiation, the damage to metabolic activity was measured again by monitoring CO2 production. Results showed a general reduction in gas production by irradiated samples, together with non-linear and non-monotone response to dose. There was also evidence of oscillations in cell metabolic activity and of X-ray induced changes in oscillation frequency.

  _______________________________________