Spectral Kurtosis against SVM for best frequency selection in bearing diagnostics

Rolling bearing is probably the most widely used component in rotating mechanical equipments and its condition monitoring and fault diagnosis to prevent the occurrence of breakdown is growing in interest since many years. Vibration signal based methods are the most popular and have been adopted in many kinds of condition monitoring systems. Starting in the early 60, an immense range of different methods has been proposed on this basis, to perform diagnosis, fault identification and classification of bearing faults. Among the others, one typical approach consists in deep analysis of the most informative frequency range output of the system under test; the identification of this band is not straightforward because the fundamental task consists in finding out the band which is the most informative in contents which, in turn., might not be corresponding to that one of the maximum response, as claimed by some authors. In this paper, Spectral Kurtosis and Support Vector Machine are analysed and compared and it is shown that they typically reach similar results, in spite of their totally different approach. A brief description of both methods is given and laboratory data are analysed from a lab rig which uses spare parts of a full size power transmission gearbox, designed by AVIO. By taking advantage of these comparisons, the analyses are conducted using classical indicators applied to the specific bands suggested by previous analysis such as the RMS and other statistical quantities. Multi dimensional graphs are reported to show the reliability of the obtained results.