FAULT DIAGNOSIS BY WHITE NOISE TECHNIQUES.
Abstract
Topics investigated include: representation of nonlinear systems (the Wiener-Volterra functional power series and other forms), methods of defining faults, the effect of these definitions on the diagnosability of the system, determination of sufficient measurement data to characterize the system under test, and diagnostic procedures. Statistical decision theory is shown to provide an effective approach to fault diagnosis. Most efficient use is made of whatever measurements or a priori information is available about the system under test. Learning machine techniques have been used to determine the diagnostic decision rule by examining known faults. In one example, a single time sample of the output waveform is sufficient to detect faulty systems with 99% accuracy. A hardware implementation is capable of diagnosing any one of four fault classes in less than 2 seconds, including measurement and printout time. Only four time samples of the output are measured. The technique is also applicable whenever the properties of the data which characterize the various fault classes are unknown, e.g., testing via secondary effects, fault prediction. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1966
- Accession Number
- AD0483022
Entities
People
- Larry W. Christenson
- Lawrence D. Turner
- Victor S. Levadi
Organizations
- Honeywell International, Inc.