A Study on Real-Time Process Defect Monitoring Based on Object Detection in Coaxial LW-DED

Coaxial LW-DED (laser wire directed energy deposition) process offers advantages of high material efficiency and rapid production speeds. However, during multi-layer deposition, heat accumulation can cause excessive heat input, leading to dripping defects, which degrade deposit quality and cause process failure. In this study, feature engineering was performed based on prior knowledge of excessive heat input phenomena in the multi-layer deposition in the coaxial LW-DED process was introduced, and a YOLOv8 (You Only Look Once ver. 8)-based object detection model was developed for real-time process monitoring. To account for differences in heat accumulation characteristics, multi-layer deposition experiments were carried out using both single-pass and multi-pass deposition strategies. The melt pool and associated phenomena under conditions of excessive heat input were analyzed using a high-speed camera, confirming that fumes and droplets are primary indicators of dripping. Based on these findings, an object detection model was developed using melt pool images to diagnose dripping defects in real time. The developed model achieved classification accuracies of 99.02% and 99.50% for single-pass and multi-pass deposition processes, respectively. Furthermore, its suitability for real-time process monitoring was confirmed by an inference time of 9.5 ms.