About the program: Automated Prognostics and Health Management (PHM) is
now a requirement for advanced military aircraft. PHM is the key to achieving
true condition-based maintenance. PHM processing strategies include modules for
the detection, diagnosis and prognosis of known fault conditions. However in
real operations there will also occur faults and other
off-nominal operations that were never anticipated nor ever encountered before.
We call these events anomalies. Missing the presence of an anomaly could
potentially be catastrophic with the loss of the pilot and aircraft.
Presented are approaches for performing anomaly detection (AD). These include
simple snapshot statistics to neural network time series modeling. The
approaches are generic and can be used for all anomaly detection problems and
for fusion with other detectors with excellent results. Application to
detection of anomalies in helicopter vibration data in a Web based application
developed for the US Army and in turbine engine gas path sensor fault detection
and isolation for NASA will be discussed. In the Army Web application, as
new data enters the system, the AD is run automatically weekly to flag
engineers of potential impending faults that would otherwise go undetected.
That information can then be used for upgrade of both on-board and ground PHM
systems. In the NASA application, the anomaly detector runs continuously and is
used to detect and differentiate sensor faults from engine component
deteriorations..
About the Speaker: Tom Brotherton
received his B.S. degree from