Monitoramento de Falhas em Maquinas Elétricas Rotativas Usando Sinais de Vibração e Machine Learning

Resumo

Rotating electrical machines are essential components of modern production systems. Due to their
structural characteristics, as, small gaps between fixed and moving parts, and by operating at high rotational speeds,
a local incipient failure can spread out through the entire equipment, leading to system and production shutdown. In
view of this, an efficient maintenance strategy becomes necessary to ensure the availability and safety of equipment,
facilities, and operators. This work presents the development stages of a system to detect and diagnose failures
in rotating equipment, through vibration signals analysis from the data set Mafaulda. For failure detection and
diagnosis, a fusion of descriptors in both time and time/frequency domains is used. In the classification stage, the
following Machine Learning algorithms are applied: Support Vector Machine (SVM), Artificial Neural Networks
(ANN) and Random Forest (RF). The obtained results, considering the accuracy (98,63%) with RF, in diagnosing
failures such as shaft misalignments, structural unbalances, as well as bearings failures, indicate that the use of
intelligent systems in detecting and diagnosing failures in rotating machines, in fact, improves the availability and
reliability of production systems, reducing untimely stops for corrective maintenance.