To create the AI model, the team fed it 9,192 photographs of parts. These parts were printed on 21 machines, which were collectively built by six brands and with four different fabrication processes. The team then gave the AI an image of a square millimeter of a printed part, and it managed to pin the print to the machine with 98% accuracy.
I wonder what the accuracy is like with a lot more machines, all of the same brand, model, and fabrication process.
Even 98% accuracy is not very good if we start talking about using this for evidence against someone in court.
Furthermore, this (probably) only works if you setup the printer in an at least similar way. You can re-level the bed, tighten or loosen any belts, change the nozzle (wear level, material, diameter), change slicer settings, use an entirely different slicer, change the room temperature, humidity, use different filaments, shake/vibrate the printer during the process etc.
At least I find it very unlikely that a 3D printer’s fingerprint wouldn’t be impacted by any of these.
I wonder what the accuracy is like with a lot more machines, all of the same brand, model, and fabrication process.
Even 98% accuracy is not very good if we start talking about using this for evidence against someone in court.
Furthermore, this (probably) only works if you setup the printer in an at least similar way. You can re-level the bed, tighten or loosen any belts, change the nozzle (wear level, material, diameter), change slicer settings, use an entirely different slicer, change the room temperature, humidity, use different filaments, shake/vibrate the printer during the process etc.
At least I find it very unlikely that a 3D printer’s fingerprint wouldn’t be impacted by any of these.