Identifying A 3D Printer From A 3D Print

A TV crime show I saw recently centered on the ability of forensic scientists to identify a plastic bag as coming from a particular roll: it’s all down to the striations, apparently. This development isn’t fiction, though: researchers at the University of Buffalo have figured out how to identify the individual 3D printer that produced a particular print. The development, called PrinTracker, uses unique differences in the way a printer lays down print material to identify a printer with a claimed 94 percent accuracy.

The researcher behind PrinTracker is [Wenyao Xu], associate professor in the Department of Computer Science and Engineering at the University of Buffalo. He presented the details at a conference in October of 2018 (PDF link) and described how it works. In the study, they printed keys on 14 different printers, then examined each print with a high-powered microscope. In particular, they scanned the print to examine the banding and attachment textures of the print. Banding is the texture on the surface of the model created by the filament being laid down, which is created by the feed rate, temperature, and shape of the outlet of the printing head. The attachment texture is the way that layers are attached to each other. When they examined the sample prints, they found that, after some image analysis and number crunching, they could identify which printer had produced the print with a decent level of statistical accuracy.

It’s an interesting read for those who want to understand how this sort of analysis works. [Xu] speculates that it might be useful for law enforcement in firearms and counterfeiting cases, and the paper also looks at if techniques like scratching or heating a print might obfuscate the identification.

Thanks for the tip, [Sascho]!

36 thoughts on “Identifying A 3D Printer From A 3D Print

  1. There are a few flaws I see with this approach, for one the nozzle on most 3D printers are brass and they slowly wear changing shape and are replaced with generally third party options, two the temperature of a print is something people (at least i) change all the time even with the same filament and finally the feed rate is also variable and I adjust that also fairly commonly when I use different materials and some slicers even let you do that dynamically during the print
    Not to mention my printer like many others change with common matinence and replacing parts such as the heater or tightening belts
    I’m not very sure they can really fingerprint any one 3D printer, maybe they can tell the difference between the ones in their lab if they don’t touch them

  2. Interesting research. But from the 14 used 3d printers, 10 were FDM and 4 SLA. The 10 FDM printers consisted of 5 models. So to prove that you can reliably pinpoint a single source from a print, i assume a test with (more than) 100 of the same model, unmodified, printers would be needed, right?

      1. Rarely does this happen! I’ve only customized the following on my CR-10:
        E3D Titan Aero extruder
        Duet 2 WiFi board (amazing 1/256th stepper motor drivers, easy firmware changes)
        Custom hearing bed surface
        A/C-powered 800w bed heater

        I’m 94% sure these forensics would not identify my prints printed from a CR-10.

        1. The point isn’t “we found this part, and we think it’s from the printer HeyRob bought two years ago so let’s go ask him.” The point is “okay, we seized HeyRob’s current printer and the nozzle he discarded last week, and tests show significant likelihood it was HeyRob’s printer which made the weapon used in the crime.”

          1. In the US (beyond a reasonable doubt) the problems are the same. If there’s a reasonable chance that it could have come from someone else’s printer, then it’s not as useful as evidence. If they can’t tell one printer out of a few hundred thousand, then there’s too much doubt.

            The standards for DNA evidence and fingerprints are pretty ridiculously stringent — this isn’t even in the same ballpark. Maybe not even the same leauge / sport.

  3. Ridiculous. Almost every consumer-level FDM printer uses a consumable brass nozzle, as you run more filament through it, the thing wears down and the “fingerprint” of the print will change.

    It would be like trying to do ballistics with a gun if every time the suspect shoots somebody, he swaps out the barrel for a new one.

  4. In the study, they were able to recognize a given printer out of 18 known printers. If there are 18 million 3D printers in the world, the problem is _a bit_ trickier.

    Plus, as above in the comments, wear and nozzle-swapping, and frame modifications, and …

  5. Wenyao Xu, is just trying to make a name for himself. His hopes are the people he is going to try to sell this concept to is not as smart at the people who use these machines, and know this is a fools dream

  6. “Banding is the texture on the surface of the model created by the filament being laid down, which is created by the feed rate, temperature, and shape of the outlet of the printing head.”

    Given the variables that create the banding, someone should be taking a look at why this person is an associate professor. all of those are extremely environmentally dependent so two of the same printers in two different locations are going to print in different patterns given the same object. While this seems to be the premise of the report, the report seems to simply gloss over the environmental impact of these variables, This is something i would expect from an undergrad paper not one from phd’s

    Say the cops suspect me of a crime involving 3d printing, what are they going to do, come to my apartment and print something on my printer? having more than one person in here definitely changes the thermal profile of the place (its a small apartment), take it to a different location and the thermals wont line up either.

    I skimmed through the paper and its hilarious, first the examples include a 3d printed grenade, tssa security keys, hand cuff keys and magazines for firearms. The keys are probably the only thing that real criminals would 3d print, no criminal is going to 3d print a gun as all it takes to beat ballistic forensics is to change the barrel, which is easy enough to acquire and do with a minimum of tools.

    The one thing that did stand out to me is that the method relies on scanning the outside texture, which generally means that it wont work if you do any post printing surface finishing.

    then there are gems like this: “Even if the outer layer is destroyed, it can be removedto analyze the characteristics of the inner layer.” he clearly doesn’t understand how printers work, how do you remove one layer with out disturbing the layer that it is attached to if a printer is printing properly with proper layer adhesion.

    There are other places where the term 100% accuracy comes up, any time a paper claims 100% accuracy anywhere its a bullshit paper. Nothing is ever 100% accurate

    1. You are more right than you realize. Maryland had a ballistics fingerprint database at the cost of some $2 million per year. Over the nearly 15 years it was running, it was used in one case, and even then, the evidence was thrown out for being circumstantial.

      When ballistic fingerprinting first started gaining traction as the new crime fighting buzzword, the FBI did some testing. See, it turns it that while the nuances and fine machining marks might make impressions on the bulletin, basically anything else will, too. Like say, lead residue or unburnt gunpowder. In practice, they determined that it was impossible to reliably identify if two bullets were fired from the same barrel in a span of as little as 10 shots. Similarly, cleaning scrubs that residue and major scoring features from the barrel. Literally, just doing basic care and maintenance on your gun defeats ballistic fingerprinting. You don’t have to do something like drop $80 on a new barrel, a $5 Hoppe’s kit from the Walmart sporting goods section will do.

      This paper is just more “publish or perish” grant seeking drivel latching onto a popular fad. The dingus who wrote it will get another grant from some faceless national endowment and the world will keep spinning as usual.

      1. I can’t help but notice that this comment points out a $2M/year waste of public funds (in only one, small, state) with the possible ramifications of putting innocent people in jail…

        and concludes by talking about the publishing pressures of academia.

  7. Its real purpose is to secure lab funding and the reputation of the contributors and the institution – it’s unlikely that the method, similar to the long-discredited Bertillon method of identifying people, would stand up to any rigorous test or expert-witness challenge in court (I’ve been that witness on occasion).

    There’s a false equivalence happening in the article as well – the plastic bag example could match two bags coming from the same roll through equivalent striae, but would not be able to identify the machinery that produced it from that alone.

    The 3D printer identification claim is similarly flawed in that (as others have pointed out) the nozzles wear and change relatively with use, and the other features (layer thickness etc.) are user-selectable to the point that if a person was intending to do something underhanded it would be no problem at all to obfuscate the print. If it were part of a small production run of some sort (someone’s GroupGets project, say) the resin roll batch and other components would all add up to pretty good (though not conclusive) matching.

    The problem with such bloviated claims (from the conference paper):

    “To address the challenge of traditional forensics in identifying unlawful 3D printed products, we present PrinTracker, the 3D printer identification system, which can precisely trace the physical object to its source 3D printer based on its fingerprint.”

    is that they’re taken seriously in some quarters and may be used to convince a technically unquestioning jury or similar enforcement agency somewhere that an innocent person is guilty of something.

    1. The other thing here is that for papers to be accepted to conferences/journals they tend to need STRONG conclusions that “X works”. A paper which says “we find a method which gives a 50% chance of…” doesn’t get past review. Also Fig 13, seriously, if you think you’ve got an effect which gives differences within the 98% to 100% range then zoom the scale in, don’t bother to show the bars all the way down to 80%.

  8. Oh good. More pseudoscience to Harass and convict innocent people. Probably less reliable than field drug tests, and definitely less reliable than polygraphs, hair analysis, and fingerprinting.

  9. Seriously? What a waste of time…
    Maybe who doesn’t know anything about 3D printing thinks this is serious.
    Depending on the slicer you use, changing the parameters, each print of the same object can look very different from each other when looked closely using a microscope.

  10. wouldn’t it make more sense to look for (dust) particles embedded in the print that can be linked to the environment the printer was in.. or perhaps even DNA from skin flakes or hairs etc that made their way into a print..

  11. very much doubt the effectiveness of this. Now if they have the same model made on multiple printers with identical materials and settings they might be able to correlate, but:
    1.A printer’s way of laying down filament will vary strongly depending on the geometry of the model
    2.A different material, even just an older reel wth more atmospheric moisture absorption, or a newer one with less will change how the print goes
    3.If the printer has had things like calibration, especially of nozzle opening size, redone between prints
    4.The temperature of the ambient air during printing, atleast to some extent
    5.If the printer is using a different profile, mroe or less perimeters or infill, a faster or slower x-y motion speed for the head…

    Incidentally, fat chance that any of those print examples the give would in any way, you know, WORK. Keys, those all have a petty good chance of snapping in the lock if not first time then after several uses. Anyone with malicious intent is going to look around for the proper tools so they can make the items from the proper (metal) materials.

    Also, what a creepy line of argument the paper uses. proposing it would be ok to get all printers sold to have these characteristics meaured beforehand. One can guarantee that oppressive measures like that would be used, solely, to hunt down anyone who saves themselves some money (and reduces environmental pollution) by 3D printing a replacement part for something where a corrupt manufacturer has built in mechanical planned obsolescence.

  12. The research is incomplete and done by someone who understand very little about 3D printing. Is enough to change few parameters on the gcode or in the firmware or even on the slicer to have very different results.

  13. do you have a hitchy bearing anywhere?
    Does your filament feed drive leave any teeth marks or have any eccentricities in its works?
    These can leave repeatable variations in layers.
    Are you using threaded fastener rod as a feed screw anywhere?
    It’s not meant for repeated spinning of a nut and wears down and binds.
    Do you have any scratches on your bed? Not talking about gouges, just something 3~4 thousandths deep.
    (Think ballistics lab here folks.)
    Do you have any little hiccups in a radius or corner.
    Think of everything that is an issue with your printing and run the tangents because those are some of the things that will be claimed to narrow down the field of possible units.

    Lets not forget chemical analysis of the filament.
    Is there anything else in your building that can permeate the filament? Paint fumes, adhesives, brand of hair spray, etc.
    Ask a clean room worker what can contaminate a process. IF they know it, so does the forensics lab.
    It’s all going to be a numbers game to convince that judge or jury of the “reasonable doubt”

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.