Detecção de falhas em rolamentos de motores de indução trifásicos por meio da análise de sinais de vibração usando a decomposição em modo empírico

Abstract

Three-phase induction motors play a central role in industry, absorbing a significant portion of the electrical capacity of an industrialized nation. Innovations such as advanced materials and control algorithms are shaping the sector, driving efficiency and reliability. The electric motor market is constantly expanding, expected to reach 169 billion dollars by 2026. Although recognized for their reliability, electric motors face challenges such as bearing failures, which makes maintenance essential to ensure efficiency and longevity. In this context, classifying potential failures in the bearings of three-phase induction motors becomes fundamental for industrial efficiency and sustainability. This research proposes an approach for detecting bearing failures, using vibration analysis combined with Empirical Mode Decomposition (EMD) to extract features from the vibration signal and apply it to Machine Learning classifiers. To perform the analysis, the study considers two main databases: one composed of signals from healthy bearings, and another composed of signals from defective bearings. The importance of fault detection is highlighted as essential to avoid unscheduled shutdowns and reduce operating costs. The study seeks to answer the question of how to improve vibration analysis for detecting anomalies in bearings, contributing to reducing downtime costs, in addition to improving the operational reliability of electric motors and advancing fault analysis, offering an integrated solution that ranges from model development and training to evaluating their performance in comparison with existing methods. The study uses a vibration signal analysis technique to identify faults in electric motor bearings. The signals are obtained by means of sensors under different conditions and subjected to Empirical Mode Decomposition (EMD), which allows separating the signal into intrinsic modes of different frequency components. For each extracted mode, descriptive statistics are calculated, such as mean, standard deviation, asymmetry, variance, kurtosis and zero crossings, with the aim of characterizing the vibrational behavior. The parameters are used to compare the performance of bearings in healthy and defective conditions, considering different rotation speeds.