Aplicação de rede neural MLP na identificação de falhas em motores de indução trifásicos utilizando base de dados gerada por modelo computacional

Abstract

This study addresses the application of Multilayer Perceptron (MLP) Neural Networks in identifying faults in three-phase induction motors. Using a database generated by a computational model, the neural network's performance in classifying different operational conditions was explored. The adopted methodology included generating the database through a computational model that simulated various operational conditions and potential faults in three-phase induction motors. The choice of MLP Neural Networks was motivated by these models' ability to learn complex patterns and deal with nonlinear relationships in the data. Detailed analysis of performance metrics revealed effective learning by the network, particularly notable performance in various fault classes. However, difficulties were identified in differentiating between the "Normal" and "Overload" classes. Generalization tests confirmed the network's effectiveness in detecting various faults while highlighting limitations in the model's sensitivity to overloads. The confusion matrix and ROC curve reinforced this difficulty, indicating similar patterns between these classes. The positive applicability of MLP Neural Networks in identifying faults in three-phase induction motors is emphasized, with significant potential for industrial monitoring. This study contributes to advancing the understanding of the capabilities and challenges associated with the application of neural networks in industrial electrical systems, with important practical implications