You're invited to attend
(Advisor: Prof. Dimitri Mavris)
“A Methodology For The Prediction and Analysis
of Precursors to Flight Adverse Event”
Monday, April 12
12:00 p.m. (EST)
Air transportation is known to be the safest mean of transportation nowadays. The drastic improvements in aviation safety since its gain in popularity are undeniably a factor in the industry's growth over the last several decades. This growth brought social and economic benefits throughout the world and was expected to keep its momentum pre-COVID-19. Stakeholders such as the National Aeronautics and Space Administration (NASA), the Federal Aviation Administration (FAA), the National Transportation Safety Board (NTSB), aircraft manufactures, and airlines have developed systems, techniques, and technologies that are to thank for today's overall safety improvements and the reduction of accidents. The industry's maintained growth is welcomed, but current safety performances have been observed to stagnate instead of declining. With safety initiatives such as the (Flight Operational Quality Assurance) FOQA program and the growing number of aviation data, many of the previous techniques used to understand the causes of accidents are not scalable. These reasons led to the development of novel methods leveraging advanced analytical tools such as Machine Learning and Deep Learning. However, current use cases have focused mainly on anomaly detection and system health monitoring, which does not bring enough reaction time to deal with an imminent event.
This research proposes the improvement of aviation safety through precursor mining. Precursors are defined as events that are highly correlated to the adverse event that they precede. Therefore, they provide predictive capabilities and can be used to explain pre-defined events. This thesis uses publicly available flight data to 1) develop a novel Deep Learning method to identify and rank precursors of multiple adverse events, 2) use unsupervised learning algorithms to group flights based on their precursors to identify potential causes for these events at a fleet-level, and finally 3) detect novelty to ensure that the developed precursor models operate within their limits and that new non pre-defined adverse events could be detected.
- Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)
- Prof. Duen Horng Chau – College of Computing
- Dr. Tejas Puranik – School of Aerospace Engineering
- Mr. Bryan Matthews – KBR, Inc., NASA Ames Research Center