BayesiaLab Blog

The BayesiaLab Digest - January 26, 2015:
Modeling Aircraft Encounters with Bayesian Networks

Written by Stefan Conrady | 26 January 2015

Here is today's citation of new and interesting applied research with Bayesian networks:

Griffith, J. D., Matthew W. Edwards, Raymond M. Miraflor, and Andrew Weinert.
Due Regard Encounter Model Version 1.0.
Project Report. Lexington, Massachusetts: Massachusetts Institute of Technology Lincoln Laboratory, August 19, 2013.
https://www.ll.mit.edu/mission/aviation/publications/publication-files/atc-reports/Griffith_2013_RP-4327_WW-27478.pdf

Abstract

Airspace encounter models describe encounter situations that may occur between aircraft in the airspace and are a critical component of safety assessment of sense and avoid (SAA) systems for Unmanned Aircraft Systems (UASs). Some UAS will fly in international airspace under due regard and may encounter other aircraft during these operations. In these types of encounters, the intruder aircraft is likely receiving air traffic control (ATC) services, but the UAS is not. Thus, there is a need for a due regard encounter model that can be used to generate these types of encounters. This report describes the development of a due regard encounter model. In order to build the model, Lincoln Laboratory collected data for aircraft flying in international airspace using the Enhanced Traffic Management System (ETMS) data feed that was provided by the Volpe Center. Lincoln processed these data, and extracted important features to construct the model. The model is based on Bayesian networks that represent the probabilistic relationship between variables that describe how aircraft behave. The model is used to construct random aircraft trajectories that are statistically similar to those observed in the airspace. A large collection of encounters generated from an airspace encounter model can be used to evaluate the performance of a SAA system against encounter situations representative of those expected to actually occur in the airspace.

Lincoln Laboratory has previously developed several other encounter models. There is an uncorrelated encounter model that is used to generate encounters with an intruder that does not have a transponder, or between two aircraft using a Mode A code of 1200 (VFR). There is also a correlated encounter model that is used when both aircraft have a transponder and at least one aircraft is in contact with ATC. Both of these models were built from radar data collected from the National Airspace System (NAS). There is also an unconventional encounter model that is used to generate encounters with unconventional intruders such as gliders, balloons, and airships—these vehicles have different flight characteristics than conventional aircraft. The framework used to construct the due regard encounter model described in this paper is similar to the prior models. The primary difference is that a different data feed is used and the model covers encounters in international flight where the aircraft of interest is flying due regard, which were not within the scope of prior models.