A prediction of welding process parameters by prediction of back-bead geometry

In this paper, results with regard to the geometry prediction of the back-bead in gas metal arc welding where a gap exists were compared. Methods using in geometry prediction were employed multiple regression analysis and artificial neural network. According to geometry prediction results, these geometry prediction methods showed low error enough to be applied to real welding. With these results, prediction system of welding process parameters was formulated in order to obtain the desired back-bead geometry. In geometry prediction error by multiple regression analysis, the gap had the largest geometry prediction error, followed by welding speed, arc voltage and welding current. Therefore, it is concluded that gap is the most difficult parameter in comprising prediction system of welding process in order to obtain the desired back-bead geometry in butt-welding.