가스금속아크 용접에서 비드 형상 변화에 따른공정 변수 예측

Abstract

We attempted to implement a model and predict the changes in process variables that cause changes in specific parameters in a weld-bead shape during the gas metal arc welding process. For this purpose, the relationships between the process variables and bead shape parameters were obtained from a mathematical model to build a database, and a deep neural network (DNN) was then proposed with the bead shape parameter as an input and the process variable as the output through the learning of the database. A DNN with six hidden layers and 512 neurons was suitable for predicting the process variables, and a maximum error of 1.27% appeared in the test applied to it. To verify the validity of the proposed model, a set of process variables was entered into the mathematical model to calculate the bead shape, which was compared with that obtained using the set of process variables by the DNN. The DNN was applied to the case where each parameter increased from -10% to +10% in the reference bead shape, and it was possible to predict the process variables with an error margin generally within 5%.