Modelagem pseudoinversa e Análise de efeitos dos parâmetros no desempenho da máquina de relutância variável 6/4 e 8/6

• Ano de defesa: 2021
• Autor(a) principal: Mamede, Ana Camila Ferreira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Máquina de relutância variável; Procedimentos de projeto; Análise de elementos finitos; Pseudoinversa de Moore-Penrose; Modelagem; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/31567; http://doi.org/10.14393/ufu.te.2021.164

The switched reluctance machine has gained much interest in industrial applications, wind power systems and electric vehicles. This happened because, its main disadvantages, such as the ripple in the torque, were overcome due to continuous research, and its advantages, such as simple and robust construction, ability to operate at high speeds and variable speeds, insensitivity to high temperatures and fault tolerance, have made the switched reluctance machine the right machine for many applications. However, some difficulties are faced to execute good designs, such as the inherent non-linearity of the machine and the modeling. This work explores the theory, operation and design procedures in the first chapters. In practical terms, performance sensitivity analysis and machine modeling is performed, both based on the variation of geometric parameters and the results obtained through finite element simulations. The purpose of this study is to provide consistent data on which dimensions must be changed for specific applications, in order to support choices made in the design and optimization stage. An analytical approach using the Moore-Penrose pseudo-inverse is proposed to model the characteristics of the switched reluctance machine. However, the proposed modeling can be generalized and used to model any physical system, as long as it is possible to obtain experimental/computational data from it. To organize the modeling data set and perform the sensitivity analysis, methods of design of experiments were used, which is a branch of applied statistics used to conduct scientific studies of a system, process or product. To certify the quality of the proposed modeling, a model using artificial neural network was trained with the same data set and its results for values prediction are compared to the pseudo-inverse model and the values obtained by finite element simulation. The results obtained prove the feasibility and quality of the proposed model.