Salazar, Andres OrtizAraújo, Valbério Gonzaga de2025-02-132025-02-132024-12-05ARAÚJO, Valbério Gonzaga de. Monitoramento e diagnóstico de falhas em motores de indução trifásicos utilizando rede neural NARX. Orientador: Dr. Andres Ortiz Salazar. 2024. 107f. Tese (Doutorado em Engenharia Elétrica e de Computação) - Centro de Tecnologia, Universidade Federal do Rio Grande do Norte, Natal, 2024.https://repositorio.ufrn.br/handle/123456789/62691Three-phase induction motors play a crucial role in industrial operations. However, failures in these machines can lead to significant operational issues, affecting both productivity and safety. Traditionally, fault detection in induction motors has been carried out using conventional techniques, such as time and frequency domain analysis, utilizing characteristic signatures from vibration and current. Although these approaches can be effective in some cases, they face limitations in terms of accuracy and their ability to handle large volumes of complex data, particularly under variable operational conditions. In light of these challenges, this study proposes the application of an Artificial Intelligence (AI) technique for fault diagnosis and classification, aiming to identify issues at early stages more efficiently and robustly, overcoming the limitations of traditional methods. The analysis was conducted using current, temperature, and vibration signals. Experiments were carried out on a test bench, simulating real operational conditions, including stator phase unbalance, bearing damage, and shaft misalignment. For fault classification, an Auto-Regressive Neural Network with Exogenous Inputs (NARX) was developed, a predictive architecture tailored for classification tasks. The optimal network configuration was determined through a selection process using a scanning method with multiple training iterations, followed by the introduction of new data to validate its efficiency. Tests conducted with the additional data demonstrated the high performance of the neural network, showcasing its ability to generalize across all evaluated conditions. The accuracy rates ranged from 94.2% to 98%, depending on the machine’s operational state.Acesso AbertoInteligência ArtificialClassificação de FalhasMotor de InduçãoRede Neural NARXdoctoralThesisCNPQ::ENGENHARIAS::ENGENHARIA ELETRICA