Using machine learning in on-board data processing

Our department has taken on the task of designing a system that will predict the aircraft structure load based on flight data and the movements of pilot controls in real time. The system must work in highly maneuverable and experimental aircrafts capable of maneuvering at a critical angle of attack. Considering this, along with the fact that no aerodynamic characteristics of the airplane wing or the airplane itself are available from the manufacturer, it is not possible to create a simulation model and perform calculations using it.This paper describes the design of suitable types of neural networks and their testing on real data. The selected solutions are compared and evaluated. It also discusses the possibilities of their implementation in on-board computing units.