Aviation Graphs offer rich insights into the global air traffic. These are traces left by an aircraft on the ground and in the sky. While they are usually only visible via vapour trails for the human eye, aviation graphs consist of signals transmitted by aircraft to a satellite system covering the entire world. The signals containing information about the aircraft’s status are analysed as they provide essential insights to answer questions from aviation industry and economy. Machine learning techniques from particle physics are used to analyse those signals due to striking similarities between the reconstruction of flight paths and particle tracks.
Vapour trails in the air are the visible traces of an aircraft for most people. The trails describe the latest path – or track – of an aircraft through the sky during its journey. But these are only the visible traces. Aircraft leave their complete traces via broadcasted signals captured by satellites around the entire world. This satellite system, called ADS-B, sends signals to ground stations and has a world-wide coverage – except at the poles. The transmitted data contain not only information about the aircraft’s position and height, but also have a high resolution, which can be used to provide in-depth analyses. They are filtered into different subsets called “categories” – abbreviated by CAT and a number – serving different purposes.
The importance of data
Data of the CAT 21 contain commercial flight data and have been used to create the aviation graphs. Once accessed, these data provide answers to questions such as “What has the aircraft been doing?” and “Where is it flown?”. The richness of these data allows to shed a light on complex questions related to regulatory aspects, economic implications due to COVID-19 and other financial impacts. Physicists at gluoNNet find answers to these questions by using machine learning techniques they have learned during their time at research institutes and university.
Motorways in the air
After having visualised the data, flight paths – or Aviation Graphs– appear as individual motorways on the world map. At a first glance, the comparison to motorways seems only correct when looking at flight paths over the oceans, but clear patterns are visible all over the globe. These patterns are especially visible around airports, for instance when aircraft approach and take off.
Connecting science and industry
Aviation graphs connect in this way science and industry, because the machine learning techniques used in physics for reconstructing particle tracks are applied to the CAT 21 data. In particle physics, physicists connect different hits in the various detectors for reconstructing the passage of a particle through the detector. Finding the origin of this trace is their goal as this is the place at which these particles have been created due to a collision between other particles. These collisions happen very often in a very short amount of time. This is why physicists need high-level algorithms that are capable to deal with the huge amount of collected data. Collisions in physics would then correspond to a place where many aircraft depart or arrive, such as airports and a particle would then correspond in this analogy to an airplane.
The more data is available, the more complex is the analysis. Novel techniques such as (Quantum) Graph Neural Networks are currently explored in both aviation and particle physics. A graph neural network contains of points – nodes – and edges – lines connecting the nodes. After having carefully trained a graph neural network, it decides how likely the different nodes correspond to each other. In this language, nodes correspond either to particles having left a signal in the detector, or to aircraft having sent a signal to the satellite depending on the underlying data. As a result, the whole passage is visible. There are already promising approaches with Quantum Graph Neural Networks to address these two use cases.
Applications in aviation industry
When applied to aviation industry, machine learning techniques offer real insights of flight patterns of individual aircraft and fleets. Additionally, regions of interest are easier detectable, such as regions affected by weather conditions or crisis regions. Another vital aspect offers the prediction of flight paths based on the flown flight track, which is important for the prediction of emergency situations.
