NASCIMENTO, Manoel Henrique Reis2024-07-252024-07-252021-07SILVA, Ítalo Rodrigo Soares. Modelo de previsão do KPI Confiabilidade em um grupo de Máquinas de Combustão Interna utilizando Técnicas de Redes Neurais Artificiais em Usinas Termoelétricas. 2021. p. 114. Dissertação do Programa de Pós-Graduação em Engenharia, Gestão de Processos, Sistemas e Ambiental (EGPSA), Instituto de Tecnologia e Educação Galileo da Amazônia (ITEGAM), Manaus, 2021.https://rigalileo.itegam.org.br/handle/123456789/20The unavailability of equipment in thermoelectric plants for any reason becomes a risk of the entrepreneur, who as a consequence bears even greater losses with a high cost of machines stopped, in addition to the penalties sanctioned and provided by law, based on this assumption the maintenance programs are methodologies that aim to contribute with techniques and tools to mitigate this problem, however, However, only the use of maintenance programs are not enough, thus, this research aims to develop an Engine Reliability Prediction Algorithm, capable of predicting the Reliability Key Performance Indicator, with the purpose of indicating the probability of the equipment to operate in a pre-defined space of time, as the object of study has a group of internal combustion machines of Thermoelectric Power Plants. In view of this, the research meets the objectives of cataloging the significant variables for the prediction model; analyze twelve ANN training algorithms, considering the supervised learning approach, where the number of neurons, hidden layers, and activation functions are performance requirements of the network; To develop the prediction model for the reliability of the motor group, where the training algorithms are validated using the best model stopping criterion; to find the best network performance based on Mean Squared Error (MSE), Root Mean Square Error (RMSE), Linear Regression, and Best Model stopping criterion; and finally, to simulate the cataloged failure data in order to analyze the technical state of the motor group with the best model. The innovation of the research is characterized by the computational methods of data processing by using: optimization methods, iterative and heuristic, characterizing the use of artificial intelligence techniques to predict the reliability in days and months, in addition, it is used predictive maintenance indicators as: Mean Time Between Failures (MTBF), Mean Time To Repair (MTTR), Availability and Reliability. To analyze the results of this research, a set of twenty load generation units was used as parameters for investigating the frequency of failures, the twelve training algorithms were applied, with a combination between the activation functions: Sigmoid, Linear and Hyperbolic Tangent, the research results show that the techniques of Levenberg-Marquardt and Bayesian Regularization showed 100% correlation between the output and simulated variables, characterizing the efficiency in predicting in days and months.pdfUsinas termoelétricasDissertação3.04.00.00-7 - ENGENHARIA ELÉTRICA