Teste de hipóteses sobre o espectro de frequência, aplicado na manutenção preditiva de motores de indução

Abstract

The goal of this project is to improve techniques and develop a predictive maintenance tool for the diagnosis of electrical machines. Based on the nonintrusive Motor Current Signature Analysis MCSA approach for broken bars detection in rotor of threephase in- duction motors, a spectral analysis method was developed which enables an increase in accuracy and reliability when compared to a simple application of the Fast Fourier Trans- form - FFT. Upon an analysis of the FFT stochastic properties and its deficiencies, here presented, a Maximum Likelihood Estimator MLE was developed that allows Hypothesis Test implementation for Gaussian random variable from the signals amplitude, phase and continuous frequency estimates. To check its viability this method was implemented on the analysis of broken bars and achieved results as efficient as Traditional Analysis. In addition, the method returns a suitable diagnostic variable for fault progress analysis, customized for each motors specifications by means of determining the confidence intervals for the diagnosis. Analysis under common industrial environmental disturbances (undervoltages, overvoltages, unbalanced phases) and other sort of mechanical and electrical faults (axis unbalance, axis misalignment, single phase interturns short-circuit) tests revealed conditions in which Traditional Analysis variable (usually present in artificial intelligence methods) may suffer changes leading to jeopardizing the diagnostic performance.