Incipient Failure Detection Using Wavelets
Abstract
This project demonstrated the feasibility of incipient fault detection for vibrating systems not only for bench test conditions (helicopter gearbox) but also for mild operating conditions (condensate and fire pumps). We employed only one of the several available data channels from the multiple accelerometers mounted on the case of each equipment type. We used a continuous wavelet transform (CWT), based on the Kiang wavelet, to select features for an artificial neural network (ANN) classifier. The CWT provided enough visibility into the fault signals to allow us to reduce the size of the feature set to 10 - 15 features. We used a low-dimensional, conventional ANN classifier with rejection of ambiguous classifications. We achieved 0.000 probability of false alarm, 0.000 probability of missed detection, and < 0.04 probability of deferral (to a subsequent feature vector) for all three data sets. The major product of this work is a single methodology to identify robust features that lead to these performance levels.... Failure detection, Failure classification, Wavelet transforms, Signal processing, Neural networks, Vibration analysis.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1992
- Accession Number
- ADA259423
Entities
People
- Alan S. Willsky
- Alonso E. Rhenals
- James C. Deckert
- Robert R. Tenney