VIBRATION ANALYSIS OF ROTATING PLANTS USING ADVANCED SPECTRAL TECHNIQUES AND NUERAL NETS

A problem of current interest to the Air Force is the automatic analysis, detection, and classification of potential critical component failures in rotating machinery. Current processing for health monitoring of parts uses relatively simple metrics or features to measure and characterize changes in vibration data recorded from several sensors. An alternative solution is to use neural nets coupled with appropriate feature extractors. Neural nets have been used extensively and successfully at ORINCON to solve multisensor and multifeature detection and classification problems in mechanical systems, as well as for sonar and radar problems. Neural networks assume no underlying model for the events of interest. Rather, the neural network "learns" to detect and classify faults by examination of features from training data which have known fault conditions. In Phase I a demonstration system that uses multiple feature extractors coupled with a hierarchy of neural nets to solve the problem will be developed. The detection and classification system will be extended in Phase II to perform fault prognosis as well. The system will be demonstrated processing real data supplied by the Air Force.