Funding entity MIT (Massachusetts Institute of Technology)
Automatic Detection of potential collisions between two airplanes is greatly improved by computing aircraft-position predictions or separation-distance prediction. Neural Networks produce accurate estimation in systems that are non-linear or may behave in a broad variety of working conditions, consequently making this technique a good candidate for airplane prediction in airports and nearby regions. In addition, the same neural network model can be easily tuned to different airport using training algorithms that automatically learn specific behaviors in such airports.
Current project integrates neural network prediction models into AMASS (Airport Movement Area Safety System) to improve estimation of the separation distance between each pair of airplanes detected by any of the two radar systems used in major airports. Better prediction allows the system to detect potential collisions earlier, hence giving more time to air traffic controllers and aircraft pilots for avoiding the incidence.