タイトル | A comparative robustness evaluation of feedforward neurofilters |
本文(外部サイト) | http://hdl.handle.net/2060/19940017269 |
著者(英) | Merrill, Walter; Troudet, Terry |
著者所属(英) | NASA Lewis Research Center |
発行日 | 1993-12-01 |
言語 | eng |
内容記述 | A comparative performance and robustness analysis is provided for feedforward neurofilters trained with back propagation to filter additive white noise. The signals used in this analysis are simulated pitch rate responses to typical pilot command inputs for a modern fighter aircraft model. Various configurations of nonlinear and linear neurofilters are trained to estimate exact signal values from input sequences of noisy sampled signal values. In this application, nonlinear neurofiltering is found to be more efficient than linear neurofiltering in removing the noise from responses of the nominal vehicle model, whereas linear neurofiltering is found to be more robust in the presence of changes in the vehicle dynamics. The possibility of enhancing neurofiltering through hybrid architectures based on linear and nonlinear neuroprocessing is therefore suggested as a way of taking advantage of the robustness of linear neurofiltering, while maintaining the nominal performance advantage of nonlinear neurofiltering. |
NASA分類 | AIRCRAFT STABILITY AND CONTROL |
レポートNO | 94N21742 NASA-TM-106440 E-8282 NAS 1.15:106440 AIAA PAPER 94-0397 |
権利 | No Copyright |