On the use of time synchronous averaging, independent component analysis and support vector machines for bearing fault diagnosis
N.C. Komgon, Njuki W. MUREITHI, Aouni A. Lakis, Marc Thomas · 2007
Condition monitoring of rolling elements bearings is investigated in this paper. Recently [11], we have shown that Time Synchronous Averaging combined with Support Vector Machines can lead to efficient bearing fault diagnosis. But the generalization performance of the SVMboundaries was strongly affected by the transmission path of the signals. This paper is then concerned with the integration of Independent Component Analysis (ICA) in this diagnosis procedure to improve its efficiency in such cases. First, we validate the use of TSA as a signal processing tool that will automatically highlight bearing defect frequencies if they are present in the envelope spectrum. Next, twenty classical features (rms, peak, crest factor…) are extracted from the envelope of the TSA-signal. To study the influence of Independent Component Analysis on the generalization performance of SVMboundaries, the twenty dimensional feature vectors are projected in their independent components space. The generalization performance of SVM-boundaries and the influence of signal transmission path as well as the faulty bearing location are then analyzed using these independent components.