Survey on a neural network for non linear estimation of aerodynamic angles

Unmanned Aerial Vehicles (UAV) design may involve issues on redundancy of the systems due to restricted available space and allowable weight. Virtual sensors offer great advantages from this point of view and several research projects carry out more or less complicated solutions in order to estimate a signal without applying a physical sensor. This approach brings to a reduction of the overall cost and to improve the Reliability, Availability, Maintainability and Safety (RAMS) performance. The patented technology named Smart-ADAHRS (Smart - Attitude and Heading Reference System) is a powerful technique presented during previous research for estimation of the aerodynamic angles. This algorithm is based on Artificial Neural Network (ANN) and receive inputs from on-board sensors only. Whereas previous studies considered also the signals coming from the Flight Control System (FCS), this work presents the important simplification of not considering them in the input vector. This paper resumes the previous results obtained in simulated environment with former neural network-based estimators. Then, a comparison of the results obtained by the new estimator, applying the reduced input vector in different environments, is carried out. Moreover, it re-discusses accuracy by means of a new test case that consider simulated realistic faults and noise. Eventually, a first analysis around performance in operative environment is conducted using data obtained from flight test campaigns. Results show how accuracy is preserved both in realistic situation and critical circumstances.