If a strand breaks inside a printed cushion, when can the defect be noticed?
Lawrence Livermore National Laboratory (LLNL) developed a camera-based inspection system that measures deposited strands as a 3D print is built, layer by layer.
The useful distinction is between an early screen and a final determination that a part meets its requirements.
Why the strand geometry matters
The demonstration concerns direct ink writing, a 3D-printing process that extrudes soft or paste-like material through a nozzle as thin, arranged strands.
The dimensions and layout of those strands matter to the mechanical performance of flexible cushions and pads.
Gaps, broken strands, and changes in filament diameter can affect the finished part.
Conventional checks may happen only after the part is complete and removed from the printer, using X-ray imaging, mechanical tests, or other offline methods.
That sequence can reveal a problem after manufacturing is already finished.
LLNL's earlier screen captures images as each layer is deposited, identifies the newest strands, and calculates measurements such as filament diameter.
From pixels to a spatial map
Printer-mounted cameras supply images to a machine-learning segmentation model and computer-vision tools, which convert the images into measurements and spatial maps of deposited material.
The team trained the segmentation model on nearly 15,000 human-annotated images spanning several lattice geometries.
A computer-vision algorithm then traced the printed strands from the model output and measured their diameter.
In tests involving 55 parts, those automated measurements were typically within a few micrometers of human-derived measurements.
For a large image, the team estimated roughly 20 minutes to an hour for manual measurement, versus milliseconds for the automated analysis.
Its roughly 100,000-fold average speed comparison concerns image analysis, not the duration of printing a whole part or completing every inspection step.
What an average diameter would miss
To demonstrate the spatial view, the researchers mapped one layer of a cushion designed with an approximately 25-by-25-centimeter footprint, combining about 2,500 images.
Filament diameter changed gradually across that map.
The pattern pointed to a slight tilt of the printing platform relative to the nozzle, a problem that an average across the part could hide.
The location of a measurement therefore mattered in this demonstration, not just its average value.
X-ray CT can provide detailed three-dimensional information, but high-resolution coverage has a practical object-size limit.
A printer-mounted camera can inspect parts beyond practical CT coverage, while producing a different kind of information from a CT volume.
An early gate, with later decisions still ahead
LLNL describes the present role as a first-pass check before more expensive tests.
The team proposes that sufficiently defective parts could be scrapped before printing ends and that inspection results could help decide when post-build CT is warranted.
Using measurement data in simulations or digital twins to link structure with predicted performance is a longer-term vision.
A printer making its own accept-or-reject decision is also framed as a future possibility, not a capability established by this demonstration.
Sources:
Reader links:
Top comments (0)