PPGEE

Programa de Pós-graduação
em Engenharia Elétrica

UFRGS

Defesa Pública da Tese de Doutorado do Mestre em Engenharia Elétrica GUSTAVO RODRIGUES GONÇALVES DA SILVA

Data:  11/10/2019 - sexta-feira
Horário: 15h00min
Local: Sala 106 do Prédio Centenário da Escola de Engenharia da UFRGS (Av. Osvaldo Aranha, 9 - 1º andar - Campus Centro)

Banca Examinadora:
Prof. Dr. Reinaldo Martínez Palhares - PPGEE - UFMG (Relator)
Prof. Dr. Júlio Elias Normey Rico - DAS - UFSC
Prof. Dr. Luís Fernando Alves Pereira - PPGEE - UFRGS
Prof. Dr. João Manoel Gomes da Silva Jr. - PPGEE - UFRGS
Prof. Dr. Jeferson Vieira Flores - PPGEE - UFRGS
Orientador: Prof. Dr. Alexandre Sanfelici Bazanella - PPGEE - UFRGS
Coorientadora: Profa. Dra. Lucíola Campestrini - PPGEE - UFRGS

Título da tese: "MULTIVARIABLE DATA-DRIVEN CONTROL: NON-MINIMUM PHASE SYSTEMS, STATE-FEEDBACK AND CONTROLLER CERTIFICATION"

Resumo:
"This thesis addresses Data-Driven control methods for multivariable systems, with focus on Non-Minimum Phase systems and an approach for state-feedback in the Linear Quadratic Regulator framework. Since a fundamental assumption in Data-Driven control is that there is no model of the plant available, controllers obtained with these methods come with no guarantee to yield a stabilizing closed-loop, thus the controller certification problem is also tackled in a Data-Driven perspective. The goal is to achieve enhanced Multivariable Data-Driven control methods to cope with a variety of systems and that yield stabilizing controllers and performances similar to the ones specified. One approach is the direct extension of existent Data-Driven methods, which deal with the Model Reference control problem therefore considering transfer matrix systems representation. In this case, special attention is given to Non-Minimum Phase systems and in this work two Data-Driven methods are extended with a flexible criterion to cope with this issue and different structures of the reference model. Another option is to consider statespace systems representation and the corresponding state-feedback design. For this, an algorithm is proposed to cope with the infinite horizon Linear Quadratic Regulator problem, based on a previous predictive approach; in this case there is no issue when dealing
with Non-Minimum Phase systems. Finally, a one-shot purely Data-Driven estimation of the H1-norm is proposed and applied to the controller certification problem, along with the estimation byproduct – the system’s Markov parameters –, in order to solve the long-lasting stability guarantee issue in Data-Driven methods. Usage of these methods in simulation and experiments on actual systems shows the applicability and improvement of the proposed enhancements. Besides, the proposed non-parametric approach to the controller certification problem shows to outperform the system identification one. Therefore this work provides new tools and a certain level of polishing in the theory for multivariable Data-Driven control methods. Most importantly, all the proposed solutions require only one (maximum two in the noisy case) experiment on the true system.

Keywords: Data-driven control, non-minimum phase systems, controller certification, multivariable control."