Detecção automática de falhas em motores de combustão interna utilizando rede neural

• Ano de defesa: 1999
• Autor(a) principal: Mancuzo, Mechelangelo Viana
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Engenharia Mecânica; Engenharias; UFU
• Palavras-chave em Português: Diagnóstico automático de falhas; Rede neural probabilística; Análise de vibrações; Motor de combustão interna; Automatic fault detection; Probabilistic neural network; Vibrations analysis; Spark engine; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/14810

This work presents the implementation of an automatic diagnosis tool of some faults founded in spark engine. Firstly, a cinematic study of the fault generating mechanisms is presented, and a partially modal analysis of an engine is illustrated, objectifying to provide information on how the vibratory energy spreads of these mechanisms to the sensor that captures the vibration signature. In a second moment, the measurement instrumentation used in the engine data aquisition is specified, and the methodology employed in the pre-processing of engines data is characterized, disposing them for the phase of characteristics extraction. In the sequence, using the magnitude of envelope spectrum, in characteristic frequencies, certain patterns associated to the defects more frequently found during the tests in the assembly line were defined. Groups of these patterns were presented to a neural network objectifying its training. The type of neural network adopted, was the probabilístic neural network (PNN), which presents desirable characteristics to solve the problem in focus: the reliability of the statistical model in which it is based, the addition or removal easiness of patterns in data set of training and validation, and the reduced computacional time expended in its training. The work finishes showing a validation procedure, in which the capacity of the trained net in recognizing patterns that are different from those used in its training, was tested. The PNN was capable to automate the fault diagnosis, establishing, like this, an objective approach for the control quality of spark engines operation condition.