A Machine Learning Approach for Prediction of Surface Temperature of the Weld Region in A-TIG Welding

• Published in: Transactions of the Indian Institute of Metals 2024, Vol.77 (3), p.907-917
• Main Authors: Chandra, Mukesh, Kumar, Shatrudhan, Ankit, Kumar, Rajak, Sonu Kumar, Rajak, Sonu
• Format: Article
• Language: English
• Subjects: Algorithms; Chemistry and Materials Science; Corrosion and Coatings; Decision trees; Fluxes; Gas tungsten arc welding; Machine learning; Materials Science; Metallic Materials; Original Article; Prediction models; Predictions; Rare gases; Software; Supervised learning; Surface temperature; Tribology; Welded joints; Welding fluxes
• ISSN: 0972-2815; 0975-1645
• DOI: 10.1007/s12666-023-03187-7

Activated tungsten inert gas (A-TIG) welding is an advanced version of TIG welding that uses activated flux to improve the penetration of weld joints. A-TIG is finding an increasing interest in the welding application due to the development of new fluxes. In recent years, machine learning has emerged as the most efficient computational tool for many manufacturing industries. In this study, a thermal camera equipped with an A-TIG welding setup was used to record the surface temperature of the weld region of the bead deposited on 304L steel. Five supervised machine learning algorithms were used to develop predictive models for the prediction of the surface temperature of the weld region in A-TIG welding for different activated fluxes. Decision tree, random forest (RF) and XGBoost are three machine learning models that were successfully implemented to predict the surface temperature with an acceptable error. Root mean square and mean square error using the RF achieved the least value equal to 16.45 and 19.922, respectively.