Um estudo comparativo entre técnicas de aprendizagem de máquina e processamento de sinais para detectar barras quebradas em um motor de indução trifásico com rotor do tipo gaiola de esquilo

• Ano de defesa: 2019
• Autor(a) principal: Jesus, Rodrigo Cardozo de
• Orientador(a): Dias, Cleber Gustavo
• Banca de defesa: Dias, Cleber Gustavo , Di Santo, Silvio Giuseppe , Araújo, Sidnei Alves de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Nove de Julho
• Programa de Pós-Graduação: Programa de Pós-Graduação em Informática e Gestão do Conhecimento
• Departamento: Informática
• País: Brasil
• Palavras-chave em Português: motor de indução; barras quebradas; transformada rápida de Fourier; transformada de Hilbert; máquina de suporte vetorial; método dos vizinhos mais próximos; regressão logística; aprendizagem de máquina; processamento de sinais
• Palavras-chave em Inglês: induction motor; broken bars; fast Fourier transform; Hilbert transform; support vector machine; k-nearest neighbor; logistic regression; machine learning; signal processing
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://bibliotecatede.uninove.br/handle/tede/3052

Detection of failures in three-phase squirrel-cage induction motors has shown increasing interest in recent years, especially in the study and implementation of new diagnostic techniques, given their use in the industrial segment worldwide. Among the existing faults, stands out the rupture of one or more bars that are part of the machine's rotor cage. One of the most studied techniques in the literature, and still used today for the detection of broken rotor bars, is the electric current signature analysis of one of the motor phases, in order to nd certain harmonics that indicate the presence of the defect. This technique uses the Fast Fourier Transform for the analysis of harmonic components, and other more recent studies have employed, for example, the Hilbert Transform, in order to improve the frequency signal resolution, as well as the use of statistical measures to nd some parameters of the motor current signal, in the condition of a defective rotor. In addition, other researches have investigated the use of machine learning techniques to aid in the evaluation of machine conditions, based on the characteristics extracted in the time and frequency domains. Thus, the present work developed a comparative study between some signal processing techniques, used in the diagnosis of broken rotor bars, and the machine learning algorithms most used in the monitoring and diagnosis of failures in induction motors. These algorithms were parameterized in several conditions and it was possible to compare not only the accuracy of each model, but also the false positive and false negative rates in each case. The results showed that the use of the statistical characteristics with those extracted in frequency domains presented the best performance. Experiments with a 7:5 kW motor, for several load conditions, and especially at low load, would allow to evaluate the best combination of the aforementioned techniques for the detection and classi cation of the defect in the rotor cage .