Bold of you to assume my prints come out the way they’re supposed to.
I have one of those generic chinese printers. Good luck printing anything on it.
On my printers, each print looks differently. The prints sometimes also fail for various reasons. Add a constant process of modifications and I doubt someone can find enough reproducible unique features on the prints, to recognise the correct printer.
Reading the article, it seems like the intent of this technology is much more geared toward manufacturing supply chains, rather than saying “this part came from John Doe’s Ender 3”. As many people have pointed out, consumer/ hobbyist grade 3D printers aren’t nearly consistent enough to produce anything resembling something as unique as a true “fingerprint”, and when you consider that most printers are modified in some way… There’s just zero possibility of it being used in that way.
The only way I could see it being used in that way is trying to prove that this printer printed this part; if they have the printed part, and it hasn’t been post-processed at all (sanded, treated, etc), they could reprint the same part on the printer in question and see if it’s “fingerprint” is the same. But I’d be pretty surprised if this tech could even reliably say, “this part came from an Ender, this part came from a Neptune, and this one from came from a P1”.
I wish my printer was consistent enough between two prints for proving that two parts came from the same printer.
Yeah this is basically just quality control geared towards mass 3d printed parts.
The technology could also be used to track the origins of illicit goods.
Does that mean ghost guns? That was my first thought when I heard of this tech.
My speculation says yes.
Those eyes are staring right into my soul
This seems like very standard ML. I’m not surprised it works, but also it likely takes a huge amount of training data (i.e. print samples) to recognize a specific machine.
I’ve done stuff like this. For instance I took a pre-trained model that could identify animals and used reinforcement learning to feed it thousands of annotated images of my cats. After this fine-tuning it could reliably tell the difference between them. Useful? Yes. Neat? Yes. But it’s not like it can identify a cat it’s never been trained on.
So it’s interesting and useful, but not as impressive or useful as the article makes it seem.
Also I’m sure something as simple as changing a nozzle or even what slicer is used would completely throw it off.