D.M. Pitt, D.P. Haudrich

This paper presents the investigation of an artificial neural network (ANN) to predict flutter speeds of a generic fighter with multiple external stores for the use in the selection of the "Most Critical" or lowest flutter speed. The new technique utilized an Artificial Neural Aeroelastic Network (AN2) that was trained on flutter data to predict flutter speeds as the external weight/pitch inertia configurations were changed. The AN^2 was evaluated for flutter speeds that it was trained with, as well as an additional set of flutter speeds that the network was not trained with. These flutter speeds were also used to evaluate the spline method, which is the current method to produce the contour plot. The number of test cases required to train the neural network was also evaluated, with the error between the input and output used as a metric to determine how well the neural networks generalized the problem. This new technique was shown to be faster and more efficient than the current techniques to produce the flutter speed contour plots.

International Forum on Aeroelasticity and Structural Dynamics, 2005, München

Conference Paper

englisch

21,0 x 29,7 cm, 12 Seiten

DGLR-Bericht, 2005, 2005-04, International Forum on Aeroelasticity and Structural Dynamics 2005; S.1-12; 2005; Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V., Bonn

NA

in getr. Zählung;

artificial neural networks, flutter, external stores

Bibliothek

2005