Prediction Model of Twin-Arc High Speed Submerged Arc Weld Shape Based on Improved BP Neural Network

Abstract

Twin-Arc high-speed submerged arc welding forming quality prediction model was developed by three layers BP (Back Propagation) neural network. In the model, twin arc current, twin arc voltage, welding speed and wire spacing are selected for the study factor, weld pool width and penetration depth are weld forming quality indicators. The adaptive learning rate and additional momentum term are introduced to improve BP algorithm. Experiments show that the network structure is reasonable of the nodes by inputting and outputting layers of 6 and 2 respectively, hidden layer nodes are 13. The developed neural network model can predict the weld geometry with high computing and predictive accuracy of maximum predictive validation error of weld width and penetration depth within 9.6%, 10.3% respectively, which can be used for real-time monitoring of the quality of welding in the twin arc high-speed submerged arc welding process.