Manifold Learning for Visualization of Vibrational States of a Rotating Machine
This paper describes a procedure based on the use of manifold learning algorithms to visualize periodic –or nearly periodic– time series produced by processes with different underlying dynamics. The proposed approach is done in two steps: a feature extraction stage, where a set of descriptors in the frequency domain is extracted, and a manifold learning stage that finds low dimensional structures in the feature space and obtains projections on a low dimensional space for visualization. This approach is applied on vibration data of an electromechanical rotating machine to visualize different vibration conditions under two kinds of asymmetries, using four state-of-the-art manifold learning algorithms for comparison purposes. In all cases, the methods yield consistent results and produce insightful visualizations, suggesting future developments and application in engineering problems.
Keywordsmanifold learning dimensionality reduction vibration analysis
Unable to display preview. Download preview PDF.
- 6.Hyvarinen, A.: Survey on independent component analysis. Neural Computing Surveys 2, 94–128 (1999), http://www.cis.hut.fi/aapo/pub.html
- 7.Jollife, I.: Principal component analysis (1986)Google Scholar
- 12.Schoen, R.R., Habetler, T.G., Kamran, F., Bartheld, R.G.: Motor bearing damage detection using stator current monitoring. IEEE Transactions on Industry Applications 31(6), 1224–1279 (1995)Google Scholar
- 13.Tavner, P.J., Penman, J.: Condition Monitoring of Electrical Machines. Research Studies Press Ltd., John Wiley and Sons Inc. (1987)Google Scholar
- 15.Torgerson, W.: Theory and Methods of Scaling. Wiley, Chichester (1958)Google Scholar
- 16.Zhang, T., Tao, D., Li, X., Yang, J.: Patch alignment for dimensionality reduction. IEEE Transactions on Knowledge and Data Engineering, 1299–1313 (2008)Google Scholar