Desenvolvimento de rede neural de multicamadas para predição de parâmetros de soldagem

Abstract

On this present work is presented the development of an Artificial Neural Network (ANN), whose finality will be get the necessary parameters to model mathematically of the curve of an heat affected zone in welding operation. Is maked a theorical referencial about the ANNs with focus on multilayer perceptron, where this is a one of structure's main that uses the supervised learning. Then it's treated about the acquisition of the database for helper on trainning of the ANN, as well about the algorithim of learning of the network, where this will modifer the synaptic weigths of the RNA. Right after is made a analyse of the percentual error of the network, as well the comparation enter the network developed on this work and other network already developed as the MLP Regressor from library Scikit Learn and the Neural Network Toolbox from the software Matlab. Keywords: