A Study on the Prediction of Real-Time Bead Width Using a DNN Algorithm in Tandem GMA Welding

The tandem welding method employs multiple wires to enhance productivity and increase deposition rates in arc welding. This study aimed to develop and validate a deep neural network (DNN) algorithm for predicting bead geometry. The algorithm processes real-time data and bead geometry measurements obtained from tandem gas metal arc (GMA) welding. A tandem GMA welding experiment with SS400 plates was performed, collecting current and voltage waveforms in real-time via a monitoring system. Furthermore, post-experiment bead width and height data were precisely captured with a 3D scanner. The acquired data served as training data to develop the DNN algorithm. Backpropagation was employed in the DNN for bead geometry prediction; its accuracy was evaluated using the Predictive Ability of Model (PAM). The DNN algorithm achieved over 96% accuracy in predicting bead width and height, suggesting its applicability in industrial settings like shipyards and automotive plants to improve weld quality and efficiency. KCI Citation Count: 0