Análise de som para detectar desbalanceamento em um sistema de hélice usando redes neurais convolucionais / Alexandre Sampaio de Figueiredo

Abstract

Early fault diagnosis is traditionally performed using vibration analysis, acoustics, and visual inspection techniques; however, these methods have limitations when applied on a large scale or in environments with high noise levels. In this scenario, advances in artificial intelligence, particularly Convolutional Neural Networks CNNs, have enabled new approaches for the automatic detection of anomalies in signals of audio and images. This work presents the application of CNNs to the processing of audio to identify faults in rotating machinery. The experiment consisted of collecting data under two conditions: normal state and with imbalance in one of the propeller blades. The audio signals were processed and transformed into spectrographic representations, allowing them to be used in CNNs for automatic classification. The applied methodology aims to contribute to the advancement of predictive maintenance in machines by offering an artificial intelligence-based approach for diagnosing mechanical faults. Results indicate high model accuracy (100% accuracy, 100% f1- score, 100% precision, 100% recall), demonstrating the viability of the proposal.