Wenxin Zhang
(Advisor: Prof. Dimitri N. Mavris]

will defend a doctoral thesis entitled,

A Data-driven Methodology for Aircraft Trajectory Analysis to Reduce Mid-air Collision Risk in Terminal Airspace

On

Monday, April 1 at 12:00 p.m. EDT

and on 

https://teams.microsoft.com/l/meetup-join/

Abstract
A mid-air collision refers to an aviation accident that occurs when two aircraft make contact with each other while in flight, posing a significant safety concern in present-day operations. Air Traffic Control (ATC) is crucial for ensuring safe separation between aircraft to prevent mid-air collisions. However, the growing aviation traffic poses challenges to ATC operators, known as Air Traffic Controllers (ATCOs), who handle heavy workloads. To address this, ATC is introducing automated systems to assist ATCOs and enhance safety. This study is motivated by the need for advanced analysis and automated decision support in terminal airspace. It proposes a novel data-driven methodology to analyze aircraft trajectories to reduce mid-air collision risk in terminal airspace. The methodology involves three main steps: (1) traffic flow identification and recognition, (2) trajectory prediction, and (3) conflict detection.

The first step requires identifying air traffic flows in terminal airspace and recognizing the traffic flow of individual flights. For the identification task, Ordering Points to Identify the Clustering Structure (OPTICS) is proposed as an alternative clustering algorithm to Density-Based Spatial Clustering of Applications with Noise (DBSCAN) commonly used in existing methods, with Weighted Euclidean Distance as the distance metric to achieve more effective traffic flow identification. For the recognition task, ensemble models like Random Forest and Extreme Gradient Boosting (XGBoost) are proposed as they offer rapid training and high accuracy. In the second step, existing trajectory prediction methods rely on an encoder-decoder architecture with Long-short Term Memory (LSTM) trained on entire trajectory sets. To address the learning challenge posed by diverse input trajectories, this study proposes training multiple predictors subsets with distinct traffic flows to improve prediction accuracy. To address the prolonged training caused by LSTM's sequential nature, Transformer is proposed as an alternative to LSTM, due to its parallelization capability and attention mechanisms, which can lead to reduced training time and comparable accuracy. In the last step, conflict detection in terminal airspace is needed while accounting for uncertainty. A Weighted Kernel Density Estimation (KDE) method is proposed to estimate aircraft positions by integrating outputs from traffic recognition and trajectory prediction, subsequently facilitating conflict detection in terminal airspace through the intersection of flight Probability Density Functions (PDFs) to achieve reliable detection of potential conflicts between aircraft.

Experiments are designed and executed to validate the effectiveness of the individual steps above. Several real flight scenarios are selected to demonstrate the efficacy of the overall data-driven methodology in analyzing aircraft trajectories to reduce mid-air collision risk in terminal airspace. This research could evolve into a real-time decision support tool in ATC, although practical application may require real flight tests and Real-time Assurance (RTA) mechanisms.

Committee

·         Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)

·         Prof. Daniel Schrage – School of Aerospace Engineering

·         Prof. Duen Horng (Polo) Chau – School of Computational Science and Engineering

·         Prof. Kyriakos Vamvoudakis – School of Aerospace Engineering

·         Dr. Tejas Puranik – Boeing Commercial Airplanes

·         Dr. Alexia Payan – School of Aerospace Engineering