Trajectory Based Flight Phase Identification with Machine Learning for Digital Twins

Kurzfassung

Analysis of aircraft trajectory data is used in different applications of aviation research. Areas such as Maintenance, Repair and Overhaul (MRO) and Air Traffic Management (ATM) benefit from a more detailed understanding of the trajectory, thus requiring the trajectory to be divided into the different flight phases. Flight phases are mostly computed from the aircraft’s internal sensor parameters, which are very sensitive and have scarce availability to the public. This is why identification on publicly available data such as Automatic Dependent Surveillance Broadcast (ADS-B) trajectory data is essential. Some of the flight phases required for these applications are not covered by state-of-the-art flight phase identification on ADS-B trajectory data. This paper presents a novel machine learning approach for more detailed flight phase identification. We generate a training dataset with supervised simulation data obtained with the X-plane simulator. The model combines K-means clustering with a Long Short-Term Memory (LSTM) network, the former allows the segmentation to capture transitions between phases more closely, and the latter learns the dynamics of a flight. We are able to identify a larger variety of phases compared to state of the art and adhere to the International Civil Aviation Organisation (ICAO) standard.