Fault Diagnosis of Rotating Machinery Using Multiscale Entropy and Principal Component Analysis Based on Motor Current Signature Analysis

Abstract

In this experimental paper, we propose a method for identifying misalignment and bearing faults, as well as their combined fault, from phase currents measured on a variable-speed induction motor. For this purpose motor current signature analysis (MCSA) technique successfully distinguishes healthy states from faulty states. In order to optimise the identification of correlations in the signals on several time scales, we propose a new fault detection tool based on multiscale entropy (MSE) algorithm SampEn applied to extract detailed features, dimensionality and isolating fault relevant variations are reduced through principal component analysis (PCA) and the comparison of the measured values with those predicted by our model built from the nominally healthy data. Finally, we used the squared prediction error (SPE) analysis to improve the fault detection capability and ensure a more comprehensive analysis of the induction machine condition, the combination of these techniques aims to provide early and accurate fault diagnosis. The experimental results obtained show that the proposed method is efficient, accurate and reliable for extracting characteristic frequencies from the current signal, and that it provides a powerful tool for fault diagnosis and condition monitoring.