ESTIMATION OF MECHANICAL PROPERTIES OF WELDED S355J2+N STEEL VIA THE ARTIFICIAL NEURAL NETWORK

Abstract

A new estimation study on the material features for the welding processes is reported. The method bases on the artificial neural network (ANN) for the estimation of material features after in the gas-metal arc welding process. Since the welding is a very common process in many engineering areas, this method would certainly assist the technicians and engineers to estimate the material features related to the welding parameters before any welding operation. In the proposed method, the input parameters of welding are defined as various shielding gas mixtures of Ar, O2 and CO2. As the resulting feature, the estimation is made on the mechanical properties such as tensile strength, impact test, elongation and weld metal hardness following ANN. The controller is trained with the scaled conjugate gradient method. It is proven that some estimated values are consistent with the experimental data, whereas some others have relatively higher errors. Thus, this method can be used to estimate especially the yield strength and elongation values, when the shielding gas proportions are ascertained before the welding, thereby the method helps to ascertain the welding gas selection in a very short time for engineers and assists to decrease the welding costs.