At Lawrence Livermore National Laboratory, researchers use direct ink writing to make flexible cushions and pads from soft or paste-like material. A nozzle lays down strands that can be a fraction of a millimeter thick, and the strands’ dimensions and arrangement affect how the finished part performs.
Small gaps, broken strands, or changes in filament diameter can compromise that performance. Traditionally, researchers finish a print, remove it from the machine, then inspect it with X-ray imaging, mechanical testing, or another offline method. Those tests may reveal a problem only after the part has been printed.
LLNL’s camera-based inspection system moves measurement into the build. It monitors printed structures layer by layer, using artificial intelligence and machine learning to measure variations and potentially identify problems before a part leaves the printer.
Cameras and segmentation map deposited material
Direct ink writing, or DIW, deposits soft or paste-like materials through a nozzle in thin, precisely arranged strands. The LLNL system combines cameras mounted on a 3D printer with machine-learning image segmentation and computer-vision tools. It converts thousands of images captured during printing into measurements and spatial maps of the deposited material.
That gives the images a manufacturing purpose: software can measure the strands as they are laid down, while the object is still being built. Brian Weston, the project’s technical lead and an AI/ML lead for digital twins at LLNL, described the system as having “kind of a brain behind the eyes.” He added: “Our system can now see as we're printing, and we can make data-informed decisions going forward.”
The research is described in a paper in npj Advanced Manufacturing. LLNL says the approach could also help examine large, intricate parts that may be difficult to inspect in full with conventional X-ray computed tomography. That is a specific advantage of watching the build: the system records deposited material layer by layer instead of relying only on an examination of the completed object.
The project team includes data scientist Michael Zelinski, engineer Hamed Ziad Ammar, and principal investigator Brian Giera, according to Tech Briefs. The lab describes the system as a way to identify small variations and potentially find problems before fabrication is complete. The available reporting does not establish that it replaces X-ray imaging or mechanical testing.
The 100,000x figure is a reported pipeline comparison
Tech Briefs reports that the automated pipeline analyzes printing data up to 100,000 times faster than manual human inspection. That is a comparison between software analysis and a person reviewing images. It does not mean the printer runs faster, nor does the report establish a 100,000-fold advantage over X-ray CT.
The speed claim matters because DIW inspection involves large sets of images and fine strand measurements. Yet faster analysis alone does not qualify a part. Manufacturers would also need to know what measurements are repeatable, which defects the system can detect, and how those results relate to the part’s required performance. The supplied reporting does not answer those validation questions.
Layer records could add evidence to part qualification
The strongest case for in-process inspection is the evidence it can collect while a part is being made. A camera system that maps deposited material can show where strands were laid down and identify variations for review. That record could support a broader production history for a component, alongside later tests and inspection results.
The distinction matters for DIW because defects can be embedded in the arrangement of thin strands. Offline testing examines a completed object; layer-by-layer imaging observes the process that produced it. A record of the build could help connect a detected defect to the layer where it appeared. Whether that evidence is enough for a particular qualification decision depends on validation and acceptance criteria not established in the reporting.
This is a measured step toward process-aware inspection, rather than proof that LLNL has created a certification system. The lab says its system can measure and map material during printing. The sources supplied here do not show deployment across commercial printers, independent validation across different materials and parts, or a procurement requirement for in-situ inspection.
Those are the tests that would determine how far the approach travels. Repeatable measurements across machines and materials would make the maps more useful beyond the research setup. Buyers would then have to decide how those records fit with existing inspection and qualification practices. Until that work is documented, the 100,000x comparison is a speed claim about image analysis, not evidence that a printed part is automatically certified.
For engineers, the immediate change is where inspection begins. LLNL’s approach puts cameras and image analysis at the printer, where deposited strands can be measured during fabrication. Human review and offline tests may still be needed, but the system points toward a workflow in which problems can be examined before the finished part leaves the machine.
That is the promise of AI 3D printing inspection: measure the layer as it is deposited, preserve a map of the build, and give engineers evidence to assess before they reach for the completed part.
