Mining Aircraft Telemetry Data With Evolutionary Algorithms

<p>The Ganged Phased Array Radar - Risk Mitigation System (GPAR-RMS) was a</p> <p>mobile ground-based sense-and-avoid system for Unmanned Aircraft System (UAS)</p> <p>operations developed by the University of North Dakota. GPAR-RMS detected proximate</p> <p>aircraft with various sensor systems, including a 2D radar and an Automatic Dependent</p> <p>Surveillance - Broadcast (ADS-B) receiver. Information about those aircraft was then</p> <p>displayed to UAS operators via visualization software developed by the University of</p> <p>North Dakota. The Risk Mitigation (RM) subsystem for GPAR-RMS was designed to</p> <p>estimate the current risk of midair collision, between the Unmanned Aircraft (UA) and a</p> <p>General Aviation (GA) aircraft flying under Visual Flight Rules (VFR) in the surrounding</p> <p>airspace, for UAS operations in Class E airspace (i.e. below 18,000 feet MSL). However,</p> <p>accurate probabilistic models for the behavior of pilots of GA aircraft flying under VFR</p> <p>in Class E airspace were needed before the RM subsystem could be implemented.</p> <p>In this dissertation the author presents the results of data mining an aircraft</p> <p>telemetry data set from a consecutive nine month period in 2011. This aircraft telemetry</p> <p>data set consisted of Flight Data Monitoring (FDM) data obtained from Garmin G1000</p> <p>devices onboard every Cessna 172 in the University of North Dakota's training fleet.</p> <p>Data from aircraft which were potentially within the controlled airspace surrounding</p> <p>controlled airports were excluded. Also, GA aircraft in the FDM data flying in Class E</p> <p>airspace were assumed to be flying under VFR, which is usually a valid assumption.</p> <p>Complex subpaths were discovered from the aircraft telemetry data set using a novel</p> <p>application of an ant colony algorithm. Then, probabilistic models were data mined from</p> <p>those subpaths using extensions of the Genetic K-Means (GKA) and Expectation-</p> <p>Maximization (EM) algorithms.</p> <p>The results obtained from the subpath discovery and data mining suggest a pilot</p> <p>flying a GA aircraft near to an uncontrolled airport will perform different maneuvers than</p> <p>a pilot flying a GA aircraft far from an uncontrolled airport, irrespective of the altitude of</p> <p>the GA aircraft. However, since only aircraft telemetry data from the University of North</p> <p>Dakota's training fleet were data mined, these results are not likely to be applicable to GA</p> <p>aircraft operating in a non-training environment.</p>

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Year

2012

Author dc:creator

• Ogaard, Kirk Anders

Contributors dc:contributor

• Ronald A. Marsh

• ant colony algorithms, collision avoidance, data mining, genetic algorithms, K-means algorithm, unmanned aircraft

Repository record dc:identifier

https://commons.und.edu/theses/1262

OAI identifier oai:identifier

oai:commons.und.edu:theses-2263