Análise da viabilidade da utilização de uma rede neural de 3 camadas para a estimação da velocidade de flutter de uma seção típica 2D

Abstract

Aeroelastic effects have had a major influence on aircraft design and performance since be fore the Wright brothers’ first flight. In particular, flutter is an extremely important problem in aeronautical design, due to its destructive capacity. Therefore, the study of methods to estimate the speed at which flutter occurs is essential to guarantee a safe flight envelope. Among these methods, the Multilayer Perceptron (MLP) neural network is a possible alter native. (MLP) neural network is a possible alternative, as it has the property of being a universal function approximator. In this sense, this work aims to use a simple MLP neural network to estimate the flutter velocity of a typical section subjected to a two-degree of freedom movement. With this in mind, the equation and implementation in MATLAB of the k method for a typical 2D section was developed and used to create the training, validation and test data for the neural network. Next, MATLAB was used to create and train MLP neural networks with a hidden layer with different numbers of neurons. Finally, the accuracy of these networks and the validity of the aeroelastic model produced by the neural network with the highest number of neurons were evaluated.