Generalization of ANN-based aircraft dynamics identification techniques into the entire flight envelope

Abstract

In this paper, an approach has been proposed in order to extend the applicability of artificial neural network (ANN) techniques for flight dynamics identification into the entire flight envelope. In general, the aircraft flight dynamics is a nonlinear and coupled system whose modeling by ANNs is only possible to a limited degree around an operational point. Therefore, it cannot be expected that an ANN trained at a specific Mach and altitude will have satisfactory results in various flight conditions. Most recent studies on ANN-based identification and modeling of aircraft dynamics have been carried out primarily at specific Mach and altitudes. In this study, by introducing a new approach based on a multi-input multi-output ANN with multidimensional output, the identification process has been successfully extended into the entire flight envelope. The results show that the proposed network, trained with experimental flight test data, is able to model the dynamical behavior of a highly maneuverable aircraft with acceptable accuracy in all of its flight conditions.