Abstract

Laser directed energy deposition (LDED) is a promising metal additive manufacturing process, but printed parts’ quality highly depends on spatter formation and melt pool stability. In this work, a high-speed camera is employed to capture and record the complex interaction between the laser, fed powders, and fusion region under various LDED process conditions. A deep learning algorithm, the YOLOv7 model, is trained to automatically detect and track the location and motion of the melt pool and spatter particles. The well-trained YOLOv7 model achieves a precision of 0.94 and is then applied to extract information on spatter count, spatter size, and melt pool geometry. We find that an elevated laser power intensifies spatter formation due to augmented vapor recoil pressure, while a high scanning speed promotes spatter ejection through Plateau–Rayleigh capillary instability. A low powder feed rate further exacerbates spatter formation owing to high metal evaporation and hydrodynamic instability within the small melt pool. In addition, this work introduces a novel melt pool stability index for real-time process assessment based on the melt pool length change rate. A stable melt pool with a high stability index generates less spatter. Otherwise, more spatters are detected. These findings advance the mechanistic understanding of spatter dynamics in LDED, introduce a novel quantitative metric for real-time melt pool stability assessment, and establish a direct correlation between the detected spatter amount and the melt pool stability. This work provides a practical framework for spatter mitigation, melt pool stability enhancement, and in-process control in advanced manufacturing.

IPC Classification

Keywords