Modeling and Detecting Anomalous Safety Events in Approach Flights Using ADS-B Data

Abstract

This thesis shows that it is possible to produce safety knowledge by mining Automatic Dependent Surveillance-Broadcast (ADS-B) data. The methodology combines exceedance detection and anomaly detection techniques to identify anomalous safety events in approach flights. One of these events is unstable approaches, which are identified with a rule-based algorithm and a Gaussian Mixture Model (GMM). The first model relies on the idea that an aircraft during the final approach needs to be flying within a certain horizontal area. The second one extracts the energy characteristics of the aircraft using ADS-B data, and later trains the GMM which is used for anomaly detection. Also, go-arounds are detected in the data using fuzzy logic with four S-functions to model the dynamics of a go-around. After identifying these events, indicators are constructed by aggregating the results to monitor safety performance. These models are applied to the ADS-B data from 2018 of the Schiphol Airport area in Amsterdam. Thus, it is possible to derive insights for runways and months, and it is possible to combine these indicators with extra variables such as meteorological data.