Estimation of weld bead geometry of gas metal arc welding process using artificial neural network

Abstract

A single weld bead geometry has significant effects on the mechanical properties of the bead, layer thickness, quality of surface bead and dimensional accuracy of the metallic parts of the welding. This research presents the application of an artificial intelligence approach using artificial neural network (ANN) and conventional multiple regression analysis for predicting the weld bead geometry in gas metal arc welding (GMAW) in which galvanized steel was the material used for the experiment. The developed models for the study were based on the experimental data. The welding voltage, welding current, welding speed and wire feed rate have been considered as the input parameters and the bead width (W) and height (H) are the output parameters in developing the models. In order to demonstrate which method performs better in terms of higher accuracy and prediction, three performance measures related to the coefficient of determination (R2), root mean square error (RMSE) and mean absolute percentage error (MAPE) were applied to the models and later compared. The results from the analysis show that the ANN models are more accurate compared to multiple regression approach in predicting the weld bead geometry due to its great capacity in approximating the non-linear process of the system.