Test In Operative Environment Of An Artificial Neural Network For Aerodynamic Angles Estimation

Angelo Lerro, AeroSmart s.r.l.
Manuela Battipede, Politecnico di Torino
Alberto Brandl, Politecnico di Torino
Piero Gili, Politecnico di Torino
Alberto Rolando, Politecnico di Milano
Lorenzo Trainelli, Politecnico di Milano

Abstract

The development of an avionic system passes through several steps, which aim is to obtain a reliable equipment able to confirm the initial theoretical design. All these steps can be influenced by a lot of factors coming both from the sensors applied and from practical considerations. Thus, the identification of a valid Flight Test Instrumentation (FTI) has a key role in the design of innovative equipment. After being validated in simulated environment, an ANN (Artificial Neural Network)-based equipment started the operative environment test phase.

This innovative patented virtual sensor named Smart-ADAHRS (Air Data, Attitude and Heading Reference System) is able to provide a complete suite of inertial and air data measurements using on-board sensors only plus a single external source of dynamic pressure. In contrast to the state-of-the-art ADS (Air Data System), Smart-ADAHRS is capable of estimate the aerodynamic angles by means of a non-linear regression that has been proven to meet, at least in simulated environment, the requirements given by typical aircraft control systems and safety operations.

A successful training procedure has been conducted thanks to a partnership between academic and industrial entities. A prototype of Smart-ADAHRS has been placed on a ULM (Ultra-Light Machine) fully equipped by an ULM-dedicated low cost, low intrusive and reliable FTI capable of synchronous acquisition of more than 40 parameters. After a comprehensive description of Mnemosine Mk-V and Smart-ADAHRS, an analysis of some results obtained by flight test is presented. Typical issues around the flight test phase, as well as future improvements are discussed in the final part of this work.