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

Not all flight data anomalies correspond to operational safety concerns. But anomalous safety events can be linked to anomalies in flight data. During the final phases of a flight, two significant safety events are unstable approach and go-around. In this paper, using Automatic Dependent Surveillance-Broadcast (ADS-B) data, we develop several exceedance and anomaly detection techniques to identify these events. Rule-based algorithms and data-driven Gaussian Mixture Models (GMM) are proposed to identify unstable approaches. A fuzzy logic approach is developed to model and to identify go-arounds. We extend our analysis combining runway information and meteorological reports to provide deeper insights on flight safety during the approach. These identification models are also applied to the ADS-B data from the Schiphol Airport area in Amsterdam in 2018. By using a reference report provided by the Dutch transportation regulatory agency, the chosen GMM model can identify 25% to 30% of reported unstable approaches, and the go-around detection model can identify 98% of go-arounds.