DGLR-Publikationsdatenbank - Detailansicht
M. Winter, C. Breitsamter
An essential task in design and certification of modern aircraft is the accurate prediction of unsteady flow-induced loads. The established techniques related to aeroelastic simulations are usually based on potential flow theory. However, in the transonic flight regime this methodology is limited in its fidelity due to distinct aerodynamic nonlinearities. Recent advancements are achieved by using computational fluid dynamics (CFD) approaches to address fluid-structure-interaction problems. Besides the accuracy improvements, the computational cost increases dramatically. Hence, the development of reduced-order models (ROMs) for aeroelastic analyses becomes a research area of increasing interest. The aim of ROM methods is to efficiently describe the dominant static and dynamic characteristics of the underlying system. Therefore, a limited set of CFD-based data is exploited to calibrate the ROM. Subsequently, the obtained model can be supplied with new inputs and ideally responds equivalent to the considered system. In this paper, a ROM approach is presented that employs radial basis function neural networks (RBF-NN) to train the dynamic relationship between the structural motion and the resulting flow-induced loads. For selecting an optimal set of basis functions, the orthogonal least squares (OLS) training technique is utilized. Since the recurrent RBF-NN is based on nonlinear system identification principles, it is suited to describe nonlinear aerodynamic effects with sufficient accuracy. Preliminary numerical investigations on the NLR 7301 supercritical airfoil show good correlation between the results obtained by the ROM methodology in comparison to the full-order CFD solution.
Deutscher Luft- und Raumfahrtkongress 2014, Augsburg
Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V., Bonn, 2014
21,0 x 29,7 cm, 10 Seiten
Stichworte zum Inhalt:
Aeroelastik, Modelle reduzierter Ordnung