Wavelet Features for Failure Detection and Identification in Vibration Systems
Abstract
The result of this effort is an extremely flexible and powerful methodology for failure detection and identification (FDI) in vibrating systems. The essential elements of this methodology are: (1) an off-line set of techniques to identify high-energy, statistically significant features in the continuous wavelet transform (CWT); (2) a CWT-based preprocessor to extract the most useful features from the sensor signal; and (3) simple artificial neural networks (incorporating a mechanism to defer any decision if the current feature sample is determined to be ambiguous) for the subsequent classification task. For the helicopter intermediate gearbox test-stand data and centrifugal and fire pump shipboard (mild operating condition) data used in this study, the algorithms designed using this method achieved perfect detection performance (1. 000 probability of detection, and 0.000 false alarm probability), with a probability < .04 that a decision would be deferred-based on only 500 milliseconds of data from each sample case. While this effort shows the exceptional promise of our wavelet-based method for FDI in vibrating systems, more demanding applications, which also have other sources of high-energy vibration, raise additional technical issues that could provide the focus for a Phase II effort.... Failure Detection, Failure Identification, Wavelet Transforms, Signal Processing, Neural Networks, Vibration Analysis.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1992
- Accession Number
- ADA259427
Entities
People
- Alan S. Willsky
- Alonso E. Rhenals
- James C. Deckert
- Robert R. Tenney