Interpretable tracking and detection of unstable approaches using tunnel Gaussian process

Abstract

Approach and landing are phases of flight with the highest accident risk. Advanced instruments and procedures have been developed to provide precise navigation for a stabilized approach and landing. With the proliferation of sensing techniques, real-time 4D trajectories can be captured at higher spatial-temporal resolution and enable data-driven decision-making for air traffic controllers (ATCO). This research attempts to augment the existing rule-based stable approach criteria using data-driven and interpretable tunnel Gaussian process (TGP) models to probabilistically characterize the 4D approach and landing parameters. The TGP explicitly and continuously models the underlying distribution of approach and landing parameters and their interrelations. In addition, it provides a comprehensible probabilistic description of anomalies in approach and landing parameters. Based on the trained TGP, we infer the landing parameters of go-around tracks recorded by the advanced surface movement guidance and control system (A-SMGCS) and analyze their adherence to stabilized approach criteria. Empirical results show that anomalous scores were in line with the factors (as reported to ATCOs) in all go-around data in the test dataset, between 0.5 NM (missed approach point) to 7.6 NM from the touchdown threshold, and provides better probabilistic insights of non-compliance, comparing to existing work. Hence, the proposed TGP can provide a ground-based safety net for the compliance of stable approaches. Furthermore, the proposed TGP-based anomaly tracking methods can be directly applied to other types of landing systems (e.g., GNSS landing system and RNAV approaches).