There’s quite an announcement from 3DPrinterOS today: they intend on automatically detecting 3D-printed gun parts.
3DPrinterOS is the well-known cloud service that allows organizations to easily manage large fleets of 3D printers. The service can store files, dispatch jobs, monitor operations, render statistics, and much more.
Today they announced a partnership with Montclair State University’s Mix Lab to jointly develop an algorithm to automatically identify whether a given 3D model is intended to be part of a gun. They explain:
“The surge in accessibility of 3D printing technology has opened new avenues for creativity and manufacturing, but it has also raised significant concerns regarding the potential production of untraceable firearms and firearm components. In response to these challenges, 3DPrinterOS and Montclair State University are committed to leveraging advanced algorithms to detect and classify 3D-printed gun parts, facilitating better tracking and management.”
This is undoubtedly true; such weapons are being detected by police literally every day in every major city. If you check the news archive for your city, you’ll no doubt run into a story about this topic within the past few weeks. And that’s just what’s being publicly reported.
The Mix Lab is a unit that’s focused on prototyping and digital fabrication and could be an ideal partner for the development of the algorithm.
3DPrinterOS VP of Global Sales Rene-Oscar Ariko said:
“We believe that this collaboration will pave the way for safer 3D printing practices. By working closely with the MIX Lab at Montclair State University, we can harness academic expertise to create a robust solution that addresses a critical issue in our society.”
They don’t say much about how this would work, aside from this:
“The collaboration will utilize the Mix Lab at Montclair State University’s cutting-edge research facilities and expertise in computer science and 3D printing design to develop an algorithm capable of accurately identifying 3D printed firearm components based on unique design signatures.”
My guess is that they will have a collection of gun part 3D models, and then train an AI on them. The AI would then be able to recognize those parts and many of similar design and function. The key will be to prepare the input training data carefully by labeling the features of these designs, rather than simply training on the entire 3D model. That approach would enable detection of the features even if hidden among other structures.
In the end, it will all come down to the detection rates. If the system is able to detect, say, 90% of the gun parts in a stream of random 3D models, that’s probably pretty good. On the other hand, there is the false positive rate. You don’t want that to be high or even above zero.
If they are successful, then the question becomes, “how do you implement it?”
In the case of 3DPrinterOS, it should be straightforward: integrate the algorithm into the cloud service to inspect each 3D model uploaded to the system.
The more interesting case is whether the algorithm might be licensed to other parties. For example, a 3D printer manufacturer might want to integrate it into their machine as a safety feature. That might actually be a salable feature, as certain organizations might prioritize buying equipment with such filters — whether they work or not.
On the other hand, 3DPrinterOS might retain exclusive use of the feature to attract more customers to their service in the same manner.
Either way, this is an interesting development that we’ll be following.
Via 3DPrinterOS
