Identification and compensator design using NARX models

• Ano de defesa: 2021
• Autor(a) principal: Petrus Emmanuel Oliveira Gomes Brant Abreu
• Orientador(a): Luis Antonio Aguirre
• Banca de defesa: Luiz Carlos Sandoval Góes, Claudio Garcia, Samir Ângelo Milani Martins, Luciano Antonio Frezzatto Santos
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica
• Departamento: ENG - DEPARTAMENTO DE ENGENHARIA ELETRÔNICA
• País: Brasil
• Palavras-chave em Português: Gray-box identification; Compensation of nonlinearities; NARX polynomial models; Hysteresis
• Link de acesso: http://hdl.handle.net/1843/40199

Gray-box identification techniques provide a promising way to build mathematical models that can be tailored to reproduce specific features of real systems and that have a suitable structure for their use in control or compensation schemes. In this thesis, some of the main concepts, definitions, and tools originally formulated in the black-box context to build Nonlinear polynomial AutoRegressive models with eXogenous inputs (NARX) and their necessary extensions to deal with the gray-box scenario are addressed. A more general NARX representation that fits the gray-box scenario is formalized when it is assumed that the auxiliary information can be converted as a new class of regressors that can optionally be included in the model. Some guidelines on how to determine a promising class of regressors from data are explored. Aiming at work on issues that have important implications for both the science and industry, the use of gray-box NARX models is studied for modeling and compensation of dynamic systems with hysteresis, a nonlinear behavior present in several applications. For a more consistent representation of this nonlinear behavior, some constraints on the structure and a specific one on the parameters of NARX polynomial models are proposed to be considered during the identification procedure. Such identified models are then able to describe not only the dynamic behavior, but also the static response which, despite being a very important feature for hysteretic systems, has been generally neglected in the literature. In addition, a more general framework is developed to explain how hysteresis occurs in such models. In the context of compensation, three approaches to design compensators are formulated for general dynamic systems and also for hysteretic systems. In short, the proposed approaches provide different ways to identify NARX models or rewrite the identified ones as a function of the compensation input signal and then use this function/compensator to calculate their values iteratively. Numerical and experimental examples are given throughout the text to enrich some discussions. Results obtained with a simulated piezoelectric actuator and an experimental pneumatic control valve demonstrate the efficiency of the identification and compensation proposals. Also, it has been found that the compensators based on gray-box techniques outperform those based on models identified through black-box techniques, and that any of the three proposed approaches significantly reduce the tracking error compared to the uncompensated system.