Aviation graphs

Aviation, Big Data, Curated, Machine Learning, Quantum Computing, Visualisation


At gluoNNet, we have been investigating the processing and structuring of airplane positioning (ADS-B) data to track, analyse, and potentially manage flight tracks, aircraft behaviour, and flight efficiency. We are working with a national aviation authority to develop a proof-of-concept as part of the digitisation of the authority with the goal of allowing the aviation authority to operate in a more proactive manner.

We processed satellite data using a scalable cloud-based architecture and tailored the data processing to the analysis requirements. We automatically populate a graph-based contextual database that contains elements such as flight times and points of origin and destination. In collaboration with our customer, we are generating alerts for high priority use cases, centred around behaviours that are of most interest to the aviation authority. The data is visualised in a multi-panel, interactive UI, focussed on supporting human-initiated tasks which will later be encapsulated within automated alerting. We have visualised data for both multiple planes and single planes based on user selection criteria.

Planned work includes productionising the system whilst adding more sophisticated analyses, more regular alerting and processing, increasing the richness of the UI and the number of alerting conditions, as well as integrating the PoC into the aviation authorities’ digital ecosystem and that of some its airline operators.

In collaboration with CERN openlab, we explore the strong similarities between particle-track reconstruction and tracking of airplane routes. In both cases, there is a huge increase in complexity as the amount of data points increases. In this study, the reconstruction of airplanes trajectories has two major aims: understanding incidents and understanding triggers of a deviation of flight path from the optimal route utilising quantum-computing techniques.

An aviation authority traditionally operates in a reactive manner, often becoming aware of issues it needs to manage well after the events themselves have occurred. Flight datasets are large and complex and require sophisticated processing and analysis. Many challenges have to be tackled when moving towards a digitally integrated and proactive aviation authority, responding to events as it receives the data indicating events of interest.

In collaboration with the aviation authority, we created a set of tools which allow us to process, structure, analyse, visualise, and alert on the aviation data available to the authority. Our tailored processing functions are based on scalable cloud architecture that can easily scale based upon the size of data received. We analyse and structure the data contextually using machine learning, generating alerts for key focus-conditions. We produce visualisations presented within a configurable UI, allowing the aviation authority team to analyse the available data through a number of lenses.

The available sensor data is structured to allow for rapid searching and querying. Its contextualised nature allows the user to ask sophisticated context-based questions, as well as building a system to easily produce context-based alarms. Our work in quantum-computing research also leads to the possibility that in the future alarms can be given in real-time, altering behaviours in advance of incidents.