Processamento de emissões acústicas aplicado a detecção de falhas em máquinas elétricas

Abstract

This work presents an approach using two acoustic emission sensors and machine learning methods to detect and classify faults in line-fed three-phase induction motors operating in steady state and subject to different shaft loads. An investigation is performed on frequency characteristics of acoustic emission and vibration signals of three kinds of faults: broken rotor bars, damaged outer race bearing and stator windings short-circuit. The method is based in frequency peaks and correlations between the microphones signals. The data are analyzed for each of the mechanical configurations and compared to the healthy situation. The results show the features behavior by means of Kohonen Self Organizing Maps and the classification is done by a combination of Support Vector Machines with accuracy rates over 99.9 %.