Local Mean Decomposition, using an Empirical Optimal Envelope and a Log-Envelope, for Bearing Fault Detection

Abstract

The analysis of bearing vibration signals is presented in this study, using a proposed approach that combines several signal processing tools. Starting with the local mean decomposition using an empirical optimal envelope algorithm to decompose the signal into several components. We then select the relevant components related to defects, which include high energy pulses, using a new indicator called defect symptom. Then, a new signal is reconstructed by adding the previously selected effective components. Peaks can be observed at the fault frequency of component in the log-envelope autocorrelation spectrum of the new signal. After applying the approach to the signals available in the Case Western Reserve University and Paderborn University databases, peaks were observed at the respective inner and outer race fault frequencies.