Artificial neural network modeling of weld joint strength prediction of a pulsed metal inert gas welding process using arc signals

This paper addresses the weld joint strength monitoring in pulsed metal inert gas welding (PMIGW) process. Response surface methodology is applied to perform welding experiments. A multilayer neural network model has been developed to predict the ultimate tensile stress (UTS) of welded plates. Six process parameters, namely pulse voltage, back-ground voltage, pulse duration, pulse frequency, wire feed rate and the welding speed, and the two measurements, namely root mean square (RMS) values of welding current and voltage, are used as input variables of the model and the UTS of the welded plate is considered as the output variable. Furthermore, output obtained through multiple regression analysis is used to compare with the developed artificial neural network (ANN) model output. It was found that the welding strength predicted by the developed ANN model is better than that based on multiple regression analysis.