Análise de redes neurais artificiais no comportamento da vida em fadiga de junta soldada do aço LN-700

Abstract

Knowledge of the behavior of a material when required under different conditions of loading becomes one of the main steps to understand its usefulness, because enables its correct sizing for various applications, so that it does not fail and have an appropriate service life. When a material is subjected to stress cyclical, undesirable effects can occur, such as the onset and propagation of cracks and failures by fatigue and with it the probability of catastrophic failure. The present work consists of the numerical evaluation of the fatigue test of a welded joint of LN-700 steel via neural networks artificial structures divided into two steps: the first is to compare the fatigue test of the base material welded with the unwelded and the second is to define which welding energy and add-on material was more effective to be used on LN-700 base material compared to the properties mechanics of unwelded LN-700 steel. With the results of the work, it was possible to validate the model of artificial neural networks used, correlating the analyzes found with the literature reviews. The artificial neural network was accurately trained, obtaining in this case 100% of the correct data in its prediction, defining the filler metal ER90S-D2 and the energy of welding of 0.44 (KJ/mm) as the ones that best suited the fatigue test to the properties mechanics of the base metal.