Traceability of welding quality through monitoring of operations

This study aims to establish traceability of welding quality in railway vehicle manufacturing by developing a monitoring system using the YOLO model. Railway vehicles require extremely high welding quality due to safety requirements, but maintaining consistent quality is challenging due to the high dependence on skilled workers and complexity of the welding process. In this research, we developed a system that detects "workers," "welding light," and "clamps" in real-time using the YOLOv11-X model trained on custom datasets specific to railway vehicle manufacturing. The detection results are visualized through Gantt charts showing temporal changes in work activities and transition matrices representing work patterns. Through filtering that considers continuous welding light for more than 3 seconds, we successfully extracted actual welding operations while eliminating noise. The analysis revealed correlations between welding duration and deformation, as well as characteristic work patterns in high-quality vehicles, such as movement from central areas to both ends and the use of clamps on opposite sides during edge welding. By integrating welding light and clamp information into a unified matrix, we enabled comprehensive analysis of the welding process. Comparison of work patterns among four vehicles with different quality levels demonstrated that this method can track welding operations and evaluate quality from multiple perspectives. This approach transforms welding work management, which traditionally relied on experienced workers' tacit knowledge, into an objective and reproducible method, contributing to work standardization and quality improvement in railway vehicle manufacturing.