Symbolic Analysis of Hilbert-Huang spectrum of Vibration data for condition monitoring of rotating machines

Abstract

This article presents a novel data-driven method of abnormal pattern identification for real-time monitoring of faults that are evolving within a rotating machine. The concept is built upon partitioning of Hilbert spectrum that are generated on intrinsic mode functions (Hilbert-Huang spectrum) of a measured vibration signal. The partitioning is further used for feature (pattern) extraction based on symbolic dynamic filtering (SDF). The statistical patterns are generated and used for identifying any possible damage in the rotating machinery. The proposed method is not only capable of detecting small anomalies (i.e. deviations from the nominal condition) in the vibration pattern, but also quantifies the extent of anomalies within an observed signal, thereby generating early warnings on damage initiation. The proposed realtime fault monitoring algorithm has been validated on vibration data recorded on a bearing test-bed, where system behavior gradually changes because of the accruing damage in bearing components.