Neural network-based dynamic model and gust identification system for the jetstream G-NFLA

Abstract

Artificial neural networks are an established technique for constructing non-linear models of multi-input–multi-output systems based on sets of observations. In terms of aerospace vehicle modeling, however, these are currently restricted to either unmanned applications or simulations, despite the fact that large amounts of flight data are typically recorded and kept for reasons of safety and maintenance. In this paper, a methodology for constructing practical models of aerospace vehicles based on available flight data recordings from the vehicles’ operational use is proposed and applied on the Jetstream G-NFLA aircraft. This includes a data analysis procedure to assess the suitability of the available flight databases and a neural network-based approach for modeling. In this context, a database of recorded landings of the Jetstream G-NFLA, normally kept as part of a routine maintenance procedure, is used to form training datasets for two separate applications. A neural network-based longitudinal dynamic model and gust identification system are constructed and tested against real flight data. Results indicate that in both cases, the resulting models’ predictions achieve a level of accuracy that allows them to be used as a basis for practical real-world applications.