Monitoramento em tempo real da qualidade de sinais de vibrações, utilizando inteligência artificial

Abstract

This work has the objective to evaluate, in real time, the signals of vibrations acquired fç>r monitoring purpose. An experimental setup compound by an electric motor and five b^ll bearings, with a load applied in the central bearing. The support bearings are self-aligning b^ll bearings and the central three are rigid bearings of single career. Was built techniques of spectral analysis and Frequency Response Functions have been applied to characterize the vibratory behavior of the studied System. Five data sets of signal condition were acquired, as: good signal, sensors in wrong position, cable problems, transient events and turned ç>ff machine. Only the self-aligning bali bearings were monitored. A Null Hypothesis Test for average comparison and a Boxplot graphics analysis were used to filter the 22 chosen vibration parameters in order to select the best sensitivity of the signals set. After identifying of the five more sensible parameters for each bali bearing, they have been used to training a Neural Probabilistic Network and into a Fuzzy Inference System. The classification tools showed good results close to 100 % of success with a test set. As one of the bali bearings presented a cage defect during the operation, it was possible to evaluate the best indicative parameters, of the studied ones, to detect this kind of defect. In this case, the global RMS value and the peak values of envelopes in the frequency range 50 Hz to 1 kHz and 500 Hz to 8 kHz, respectively.