Comparison of frequency domain methods in feature extraction

摘要

Frequency spectrum is widely used to determine frequencies present in a signal, but it fails to extract information regarding deviations from Gaussianity and presence of phase relations. Higher order spectra, on the other hand, contain such information and have shown large potential in numerous disciplines including vibration based diagnostics. In this paper, the frequency spectrum and five approaches to the estimation of bicoherence are used to extract condition-specific features from vibration signals. The laboratory experiments were done using a test rig which includes an electric motor coupled to a water pump via a bevel gear. Acceleration was measured from the pump in different conditions which include normal operation, mechanical looseness, misalignment and cavitation. The feasibility of the tested approaches to the feature extraction problem is discussed.