Implementation of a neural network based airspeed estimator applicable to an EH101 helicopter

This paper reports on the development of a novel neural network (NN) based airspeed estimator, and focuses on system design. In particular, a novelty detector, capable of preventing the system from producing an erroneous output during conditions in which it is known that a NN model will be extrapolating, is included within the design. A simulation model of a single main rotor helicopter was used to add a level of robustness to the design, which could not be gained from flight data alone. dSPACE, a rapid-prototyping tool, was used to simulate the system in real-time, and allowed an assessment of whether the airspeed system could be incorporated within an EH101´s management system.