Condition monitoring of rolling elements bearings is investigated in this paper. Recently [12, 13], we have developed a diagnosis procedure that combines a signal processing tool, i.e. Time Synchronous Averaging (TSA), and a pattern recognition method, i.e. Support Vector Machines (SVM), for bearing fault detection and prediction. As the generalization performance of the SVM-boundaries was strongly affected by the signal transmission path, this paper is then concerned with the integration of Spectral Kurtosis (SK) analysis in the diagnosis procedure to improve efficiency in such cases. We validate the use of both Time Synchronous Averaging and Spectral Kurtosis analysis, as signal processing tools that will automatically highlight bearing defect frequencies in the envelope spectrum. Twenty-one features (rms, peak, crest factor, band spectral energy, etc...) are extracted from the envelope of signals obtained from these two analyses and are used in the learning scheme. Results show that the generalization performance is less affected by the signal transmission path and the faulty bearing location, thus demonstrating that the modified diagnosis procedure can actually find some underlying patterns that are common to each type of bearing failure. The case-dependency of the support decision tool can therefore be reduced.

This content is only available via PDF.
You do not currently have access to this content.