Camera-based inspection system for 3D printing built by LLNL
Researchers produce a system that uses machine learning to flag print defects layer by layer, before a part ever leaves the printer
A team at Lawrence Livermore National Laboratory (LLNL) has built a camera-and-software system that checks 3D printed parts layer by layer as they print, rather than after, using artificial intelligence and machine learning to flag flaws in real time.
The system targets direct ink writing (DIW), and cameras on the printer feed images into ML-based segmentation software that maps the material as it’s deposited.
“We now have kind of a brain behind the eyes,” said Brian Weston, Engineer and AI/ML Lead for Digital Twins at LLNL and the project’s technical lead. “Our system can now see as we’re printing, and we can make data-informed decisions going forward.”
Conventionally, a part must be finished and removed before it can be checked, typically with X-ray imaging or mechanical testing, but these checks are costly and slow. The LLNL system instead captures images as each layer goes down and calculates measurements such as filament diameter in real time.
The team trained its image-segmentation model on close to 15,000 hand-annotated images. Across 55 test parts, results landed within a few micrometers of manual measurements, which could take up to an hour per image versus milliseconds for the software, about 100,000 times faster on average.
In a larger test, the team mapped roughly 2,500 images from one layer of a 25-by-25-centimeter cushion, revealing a slight platform tilt that an average measurement would have masked.
Brian Giera, LLNL’s Associate Program Director for Data Science, AI and Manufacturing, said: “On-machine inspection will be a huge unlock for decreasing costs, increasing throughput and providing more information on the things we build. It’s achieving a holy grail capability in the field and was done so in a way to spread to other important areas.”
