Process reliability in Laser Powder Bed Fusion (LPBF) is continually challenged by internal defects (e.g., porosities, inclusions). These discontinuities are tightly linked to instabilities in the melt pool dynamics, specifically resulting in the generation of spatters ejected from the melt pool. This work focuses on developing a highly effective quantitative tool using Deep Learning (DL) for precise spatter detection and trajectory tracking during LPBF. Experimental campaigns conducted on an instrumented LPBF bench at the PIMM laboratory with aluminum alloys Aheadd® CP1 and Aheadd® HP1 enabled the creation of a comprehensive spatter database using high-speed imaging. Detection with MSER and YOLOv11, combined with tracking algorithms such as Kalman filtering, allowed precise reconstruction of spatter trajectories, enabling detailed analysis of spatter behavior in relation to melt pool dynamics. These results demonstrate the potential of AI-based monitoring tools to improve understanding of process instabilities and support defect-lean metal additive manufacturing.

Keywords: Deep Learning; Laser Powder Bed Fusion (LPBF); Spatter; Trajectory Analysis; In-situ Monitoring