Medidas de informação e sistemas inteligentes aplicados no diagnóstico de curto-circuito do estator de motores de indução trifásicos

Abstract

This work proposes the study and development of an alternative approach to diagnose stator short-circuit faults in induction motors driven directly from a supply line. In order to reduce the size and complexity in these types of systems, signal processing techniques of extraction and feature selection are used. In the extraction step, the mutual information of the delayed phases of current signals, ia and ib, are computed and in the selection procedure, the algorithms C4.5 decision tree and multilayer perceptron neural network are employed in order to obtain an effective diagnostic of stator short-circuit faults. To assess the classification accuracy across the various levels of stator short-circuit fault severity (from 1% to 20%), offline and online experimental tests also considered a wide range of load conditions and voltage unbalance in the power supply. The obtained results indicate that this approach can be employed to classify stator short-circuit faults.