Interpretação matemática da variação luminosa no processo de soldagem MIG/MAG pulsado para classificação do destacamento através de uma rede neural artificial

Abstract

The study of Neural Network demands a careful treatment concerning the data base to attain the best results during the creation and training phase. This treatment is carried out with several statistical methods that can easily describe the physical phenomenon analyzed. This work focus on the study of the best way to describe mathematically the signal geometric shape that is produced by optical sensor during the welding process. In this way, enough quantity of statistical parameters, like mean, inclination and polynomial regression, was used in each network. Some networks have taken to the final stage to classify the type of metal transfer in the welding process. The majority of the results was satisfactory for both answer velocity and efficiency.