A new study has developed a method to identify which 3D printer produced a specific part.
Production parts are meant to be identical, so that they can be used without hesitation in real application scenarios. Each part should perform the same.
Well, the performance is not exactly the same: the performance must land within the specified tolerances. If so, it’s considered “good”.
To achieve this, most production operations will make the production parts in a consistent manner: the exact same type and model of 3D printer, the exact same material from the same supplier, using the exact same print parameters. This more or less guarantees that the same part is produced.
However, researchers at the University of Illinois Urbana-Champaign discovered something interesting: using an AI system, they could determine which specific 3D printer produced a part. Each machine was found to have a unique “signature” that results in slight variations between prints — even though they all fall within the production tolerances. The University of Illinois explains:
“While studying the repeatability of 3D printers, King’s research group noticed that the tolerances of part dimensions were correlated with individual machines. This inspired the researchers to examine photographs of the parts. It turned out that it is possible to determine the specific machine that made the part, the fabrication process, and the materials used – the production ‘fingerprint.’”
The University of Illinois Urbana-Champaign’s Bill King said:
“These manufacturing fingerprints have been hiding in plain sight. There are thousands of 3D printers in the world, and tens of millions of 3D printed parts used in airplanes, automobiles, medical devices, consumer products, and a host of other applications. Each one of these parts has a unique signature that can be detected using AI.”
Their AI model used images of over 9,000 3D printed parts made on 21 different 3D printers for training. Incredibly, the researchers found that the signatures could be detected with 98% accuracy from as little as a single square millimetre of surface image. They also found that usable AI training could be accomplished with only ten sample parts from the machines.
This technology will certainly be of interest to those working on quality control for 3D printed production parts. It would be possible to identify the specific machine that produced a failed part, for example.
It gets more interesting when considering how this technology could be used when obtaining parts from a third party. During setup, it would be possible to train an AI to recognize the machines that produced the parts. Then, later, if the supplier changed machines, it would be immediately detected by the buyer.
Another angle is to detect machine failures. If a machine is wearing out components, then it may slightly deviate from its original signature. Then, operators could be alerted to a possible failure or the need to retune the machine.
This is a fascinating development that will almost certainly be in wide use in the future.
