Machine learning-based optimal value calculation for welding variables in AR training

Currently, the shipbuilding industry is experiencing a surge in orders due to the rising demand for eco-friendly ships, necessitating the optimal use of available resources for production. However, the production workforce has not fully recovered to the level required to meet these increased orders following large-scale industry restructuring. In particular, there is a shortage of highly skilled welders, and concerns are growing about the transfer of expertise due to an aging workforce and a lack of younger workers. Shipbuilders worldwide face similar challenges and are exploring various methods to transfer the tacit knowledge of skilled welders to less experienced workers, which has introduced unforeseen challenges. In this study, we develop a machine learning algorithm that suggests the optimal values of key welding variables for an AR-based welding training system designed to assist less skilled workers. We collected welding data from highly skilled workers using the FCAW (Flux-Cored Arc Welding) technique, which is commonly employed in the shipbuilding process. The welding variables that represent tacit knowledge were identified and trained using the Extra Trees Regressor model. Subsequently, a welding AR training system was implemented, allowing the trained model to guide users on the optimal values for welding variables. Finally, the effectiveness of this system in training welders was verified at a shipyard technical training center.