Novel Neural Architecture for Air Data Angle Estimation

This paper presents a novel architecture for air-data angle estimation. It represents an effective low-cost low-weight solution to be implemented in small, mini and micro Unmanned Aerial Vehicles (UAVs). It can be used as a simplex sensor or as a voter in a dual-redundant sensor systems, to detect inconsistencies of the main sensors and accommodate the failures. The estimator acts as a virtual sensor processing data derived from an Attitude Heading Reference System (AHRS) coupled with a dynamic pressure sensor. This novel architecture is based on the synergy of a neural network and of an ANFIS filter which acts on the noise-corrupted data, cancelling the noise contribution without interfering with the turbulence frequencies, which must be preserved as key information for the AFCS activity