Artificial neural networks for aerodynamic modeling

An ANN is an information or signal processing system consisting of a large number of simple processing elements. The elements are called artificial neurons or nodes. These neurons are interconnected by links with are characterised by the so called weights and cooperate to perform parallel distributed processing in order to solve a desired computational task. There are mainly two types of ANNs: Feedforward neural networks (FFNNs) and recurrent neural networks (RNNs). In this report four variants of the recurrent neural networks (RNNs) have been studied and their features and relations have been explicitly brought out. Two RNN schemes have been developed for parameter estimation of dynamic systems and validated using noisy simulated data with and without outliers and with practical flight mechanics models. Subsequently the schemes, in output error formulation, have been applied to flight data for parameter estimation. Many interesting features of these schemes as applied to simulated and real data have been illuminated. Also relationships between various parameter estimation schemes have been made explicit. Some problem areas have also been highlighted which sould be resolved before clear benefits from the RNN based schemes can be obtained for parameter estimation using flight data. Two training algorithms for feedforward neural networks (FFNNs) have been implemented and validated using simulated data. Some aspects of ANN applicability to aircraft data analysis have been summarized. Validation and application of FFNNs to aircraft aerodynamic modeling based on flight data have been taken up. (orig./AKF)SIGLEAvailable from TIB Hannover: RN 4165(1994,41) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekDEGerman