Reducing failures in machinery and equipment is the main objective of maintenance processes to increase production efficiency. Therefore, predictive maintenance processes seek to predict failures by detecting defective components. The bearing is the central component of studies for the development of failure detection systems, as bearing defects represent the main cause of failures in rotating machines. Diagnosing the presence of a defect and determining when to replace the bearing requires the installation of supervisory systems that, from sensors, collect data used to predict a failure. The diagnosis of defects in bearings is based on the analysis of the vibration signal, since mechanical defects change the vibration level. The High-Frequency Resonance (HFR) technique is one of the most widespread for diagnosing defects in bearings based on the analysis of the vibration signal, capable of identifying frequencies that characterize a failure. However, this technique is capable of diagnosing the defect only in vibration signals collected with constant rotational speed, being unable to detect defects on signals with variable rotational speed conditions, known as nonstationary signals. To solve this problem, this research proposes a strategy based on Generalized Demodulation, which cancels the effect of the nonstationary signal velocity variation. After applying this technique, it is possible to identify frequencies that characterize failures by applying the HFR technique to the demodulated signal. The experiments were performed on real and simulated data from defective bearings. The results induced that, knowing the speed variation curve of a signal in a nonstationary regime, it was possible to demodulate the signal and identify frequencies that characterize bearing failures.
|
