LLNL uses AI and cameras to catch 3D printing defects mid-build

Lawrence Livermore National Laboratory developed a camera-based inspection system that uses machine learning to check direct ink writing parts while they are printed. On 55 test parts, automated strand diameter measurements were typically within a few micrometers of manual ones.

• Segmentation model trained on nearly 13,800 human-annotated images
• Measurements on 55 test parts within a few micrometers of manual ones
• About 2,420 images of a single layer for a 25x25 cm part
• Spatial map revealed a build platform tilt hidden by averages