Machine Learning for Prediction of Arc Length for Seam Tracking in Tandem Welding

Abstract

A tandem welding process is frequently used in the welding of large structures with the aim of improving productivity and deposition rate. However, in this process, considerable thermal deformation of the base material occurs owing to two arc heat sources, thereby deteriorating the welding quality. Seam tracking in tandem welding is essential to improve the welding quality. However, in tandem welding, there are several issues in selecting a welding line tracking signal owing to arc interaction. Therefore, in this study, machine learning was applied to select signals for seam tracking in tandem welding. The existing seam tracking signal selection method predicts the contact tip to work distance(CTWD) by measuring the welding current and voltage. The CTWD, varies depending on the welding current and voltage signal; in this study, it was predicted to perform regression analysis and signal-to-noise(SN) ratio analysis. Through the SN ratio analysis, the highest SN ratio was selected as the seam tracking signal. Subsequently, machine learning linear models and regression tree models were trained. As a result of comparing the difference between the machine learning model and the regression analysis, the prediction equation of the regression tree model is appropriate as the welding line tracking equation. Key words: Tandem welding, Arc sensing, Seam tracking, Signal analysis, Machine learning