Flight simulator modeling using neural networks with spin flight test data

Abstract

Introductionpilot training effectiveness in a flight simulator depends on the fidelity of the simulation. While parameter estimation techniques provide a high level of fidelity for most flight regimes, aerodynarmc modeling for highly non-linear regimes is di5cult. In this paper, a recurrent neural network is shown to be capable of providing an adequate simulation for aircraft dynamic response during spinning flight. A method is developed to train the network with flight test data measuTed during actual spins in a light single engine aircraft, and is shown to be f%ri rl y capable of simulafing responses to control inputs for other spins not used in the training An added advantage ofthis approach is its simplicity and ease of implementation. d a k P Q R T V X

A review and historical development of analytical techniques to predict aircraft spin and recovery characteristics

“…Although system identification has been used since late 1960s to identify aircraft models in the normal flight regime, its application to identify spin models is relatively recent (Klein and Morelli, 2016). Several modeling approaches, including neural networks (Cho and Nagati, 1995), local model networks (Dias, 2013), locally weighted regression (Bunge and Kroo, 2018) and empirical mode decomposition (Mokhtari and Sabzehparvar, 2019) have been used to identify aircraft spin models. As the flight data is collected from the actual spinning aircraft, there are no scale errors pertaining to dynamic similitude requirements.…”