Helicopter Transmission Diagnostics Using Vibration Signature Analysis
Abstract
Fault diagnostics represent a vital task in the monitoring of mission critical systems, as well as for condition-based maintenance of machinery in general. The focus of this report is on the early detection, and subsequent classification, of small changes in the behavior of mechanical systems. Such changes, known as incipient faults, portend the development of more serious failures. Physical models of machinery processes, which are useful for model-based fault detection and isolation, are not generally available in most applications. Instead, the approach to fault detection considered in this study involves the application of statistical change detection. Statistical change detection is essentially the problem of homogeneity testing within a time series. In particular, statistical change detection algorithms seek to detect situations in which a given model that describes the initial behavior of a time series, eventually fails to describe that time series accurately. The performance of non-likelihood-ratio techniques are evaluated on a CH-47D helicopter combiner transmission (non-seeded) fault; results indicate that the fault is detected in its incipient stage. The approach to fault isolation (i.e., classification) discussed herein is based on the use of minimum-logistic-loss polynomial neural networks (PNNs). The fault isolation capabilities of PNN classification networks are investigated using seeded-fault data taken from CH-46E helicopter combiner transmissions. Perfect fault classification results are achieved.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1996
- Accession Number
- ADP010197
Entities
People
- B. E. Parker Jr.
- H. V. Poor
- M. J. Szabo
- M. P. Carley
- R. J. Ryan