A comparison of model-based and black-box methods for speed estimation in aircraft

The anti-skid control system in aircraft is confined to the landing-gear subsystem, and, for safety reasons, it must rely on local signals only. Therefore, it can use only two measurements: the wheel speed and the pilot braking pressure request. Therefore, the antiskid control logics are generally wheel deceleration-based, as the slip cannot be computed since the aircraft speed is not available. The vehicle speed estimation is commonly done in automotive systems, made possible also by the presence of additional sensors usually coming from an Inertial Measurement Unit (IMU). This work explores how the aircraft speed can be estimated using only the landing gear signals, and if the resulting estimate can be accurate enough to be used in closed-loop with a slip-based anti-skid controller. To do so, two estimation approaches are considered: a sliding-mode model-based one, and a black-box approach grounded on recurrent neural networks. Experimental results are shown, witnessing the potential of black-box approaches.