Aprendizado de máquina aplicado na classificação de defeitos de máquinas rotativas

Abstract

This paper addresses the application of machine learning techniques in the classification of defects in a rotor, a significant challenge in mechanic maintenance and ensuring operational efficiency in various important industrial sectors. With technological advancement and the increasing demand for more accurate and reliable condition monitoring systems, the use of machine learning algorithms has proven to be a promising approach for fault detection and classification, enabling more effective, faster interventions and reducing machine downtime. This study focuses on exploring different machine learning techniques, including Naive Bayes, decision tree, random forest, logistic regression, support vector machines, and k-nearest neighbors, to classify defects in rotating machinery based on vibration data. The adopted methodology involves the use of a database called Mafaulda (Machinery Fault Database), which was developed by the Polytechnic of the Federal University of Rio de Janeiro in 2017, extracted from a test bench. This database contains data with induced defects and normal machine data, requiring data preprocessing before applying the algorithms, after which the accuracy results of the different algorithms are compared. The results obtained demonstrate the effectiveness of machine learning techniques in classifying defects in rotating machinery, highlighting the importance of feature selection. Furthermore, the study reveals the potential of these techniques to improve mechanic maintenance strategies, contributing to the optimization of industrial processes and the reduction of operational costs.