Diagnóstico de falhas localizadas de rolamento em motores de indução trifásicos a partir de um modelo matemático

• Ano de defesa: 2023
• Autor(a) principal: Bianchini, Carolina Alves
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Motores elétricos de indução; Rolamentos; Padrões de produção; Electric motors, Induction; Bearings (Machinery); Production standards; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/31011

The wide use of three-phase induction motors (TIM) in various industry sectors encourages the search for techniques of monitoring and diagnosing defects in these machines to avoid production downtime and unforeseen maintenance. Among the TIM’s components with more occurrences of failures is the bearing, whose fault detection can be performed by analyzing the behavior of some motor variables, such as vibration and/or stator current, and its identification, using artificial intelligence. However, few studies on the modeling and simulation of these faults present a viable resource to cover several engines and specific operating conditions, such as different levels of loads and supplies. Therefore, this work aims to develop a model for TIM’s bearing failure simulations that provides current data of a stator phase to analyze this statistical features in the time and frequency domains. From information extracted from simulated and experimental current datasets of 1 and 2 HP motors, samples with 23 statistical attributes were submitted to unsupervised (Kohonen self-organizing network) and supervised classifications (Random Forest, k-Nearest Neighbors, Multilayer Perceptron and Support Vector Machine). The high accuracy rates and 1-score obtained in the classifications between healthy and faulty engines can validate the use of this artificial database linked to experimental observations in order to make the diagnosis of bearing failures currently performed more robust. In addition, the generated topological maps allow verifying the proximity between simulated and experimental samples when clusterized, given the similarity between their characteristics. To check the generalization of the proposed model, the supervised algorithms were also trained and tested with simulated data from different motor powers, along with the experimental samples of 1 and 2 HP engines, obtaining rates above 94,00%.