Sistema inteligente para diagnóstico de falhas em rolamentos de motores de indução trifásicos via análise sonora

Abstract

Three-phase induction motors are part of practically every industrial process, they are considered the core of the modern industry, however, like any other equipment, they are susceptible to faults, which can compromise the production and cause accidents if they are not rightfully monitored. The conventional techniques to detect faults in induction motors, for example, vibration and motor signature current analyses, have some degree of invasiveness. The purpose of this work is to develop a method totally non-invasive in order to detect and to diagnose faults in three-phase induction motor bearings through sound analysis from the motor in operation, using artificial neural network and Wavelet processing. The proposed methodology is based on the acquisition of the sound emission by an electronic dispositive, the decomposition of this acquired signal in details and approximation using the Multiresolution Wavelet Analysis (MWA), and finally, the evaluation of the statistical results of these details that will count as data input for a system of artificial networks of the Multilayer Perceptron type (MLP) and Radial Basis Function Network (RBFN). The proposed technique was tested, validated and experimentally trained using a bench test. As an outcome, the MLP presented an average of 97% of matching results in the detection, and the RFB presented an average of 90%.