Researchers from The Grainger College of Engineering at the University of Illinois Urbana-Champaign have developed a deep learning model that can determine which specific 3D printer created a part, using only a photograph.
Working with SyBridge Technologies, the method offers a new way for manufacturers to verify part authenticity, monitor supplier compliance, and detect process changes without needing access to factory records.
Led by Bill King, professor of mechanical science and engineering, the study published in Nature explains how parts made through AM carry subtle surface textures that vary between machines. Although these differences are invisible to the human eye, they are consistent enough for an artificial intelligence model to recognize and trace them back to the original printer with a reported accuracy of 98%.
“Our results suggest that the AI model can make accurate predictions when trained with as few as 10 parts,” King said. “Using just a few samples from a supplier, it’s possible to verify everything that they deliver after.”
Model training and performance
To train the model, the researchers produced 9,192 parts using 21 industrial 3D printers from Carbon, HP, Formlabs, and Stratasys. These included Carbon M2 and L2 systems for digital light synthesis (DLS), HP Jet Fusion 4200 and 5200 machines for multi jet fusion (MJF), Formlabs 3B+ printers for stereolithography (SLA), and Stratasys Fortus 450mc and 900mc systems for fused deposition modeling (FDM).
Parts were printed by multiple contract manufacturers using six materials: Carbon’s RPU 70, EPX, and UMA resins, Formlabs Black resin, HP’s PA12 (dyed black after printing), and ABS filament for FDM. All parts were fabricated in black to ensure consistent appearance and were scanned using a flatbed scanner at 5.3 µm resolution to capture fine surface details.
The AI model was based on the EfficientNetV2 architecture. It was trained to analyze small sections of each image and predict which machine had produced the part. These predictions were then aggregated using a voting method to determine the final result. Even when provided with just a few image patches, the system demonstrated high confidence in identifying the manufacturing source.
Among the four processes, DLS was explored in the most detail. For these parts, the model identified the correct supplier with 98.7% accuracy. It also predicted the part’s position on the build tray with over 98% accuracy when adjacent placements were included and matched parts to their production batch with nearly 87% accuracy. Even though the DLS materials were visually identical, the model correctly classified all three resins with 100% accuracy.
Performance was also strong across the other processes. For MJF and SLA, the model reached over 90% accuracy in identifying the printer once it had at least 1 mm of surface area to analyze. Naturally leaving more visible patterns, FDM required approximately 3 mm of image area to reach a similar level of accuracy. The model retained more than 95% accuracy for FDM parts even when the image resolution was reduced.
Out of the 21 machines tested, 12 were classified with zero errors. Most misclassifications occurred between printers of the same model and process, which suggests that the fingerprinting system can detect even minor differences between similar machines. Accuracy remained stable across different testing conditions, including randomized subsets of the data.
The study also explored how image resolution and sampling size affected the model’s performance. For DLS parts, just 200 µm of surface detail was enough for the system to make accurate predictions. MJF and SLA required about 1 mm of surface data, while FDM needed at least 2 mm to preserve key identifying features. Larger samples captured more area but lost clarity, while smaller ones failed to contain enough variation.
The researchers believe this fingerprinting approach could become a valuable tool for quality control and supply chain monitoring. It could help detect unauthorized changes in manufacturing processes, confirm the use of approved equipment and materials, and identify counterfeit parts in critical sectors such as aerospace and defense.
All data and code from the study have been made publicly available on GitHub, allowing others to test and adapt the model for different manufacturing environments.
Traceability efforts in AM
The Graigner team’s work is part of a growing effort across the industry to develop novel ways that ensure traceability, security, and authentication in additive manufacturing.
In 2021, University at Buffalo researchers developed a method to trace the origin of FDM 3D printed parts by analyzing the unique thermal behavior of each printer’s extruder. By measuring heat signatures during pre-heating, the team created “Thermotags,” thermal fingerprints that were embedded as hidden watermarks in printed parts.
Using 45 different extruders, the approach achieved 92% accuracy and allowed for binary encoding by altering layer thickness. Aimed at preventing IP theft and counterfeiting, the technique offered a passive, machine-specific alternative to traditional watermarking and is being refined for greater security through blind watermarking.
Elsewhere, New York-based additive manufacturing firm PrintParts shipped its first batch of SmartParts to select customers. Developed to support authentication and traceability, the SmartParts system uses programmable nanoparticles that can be scanned to verify the part’s material, supplier, and manufacturing method.
The data links each component to its digital twin on a browser-based platform, which also integrates with existing MES and ERP systems. Aimed at industries like aerospace and defense, the service offers a way to ensure parts meet strict quality and certification standards.
Take the 3DPI Reader Survey — shape the future of AM reporting in under 5 minutes.
What 3D printing trends should you watch out for in 2025?
How is the future of 3D printing shaping up?
To stay up to date with the latest 3D printing news, don’t forget to subscribe to the 3D Printing Industry newsletter or follow us on Twitter, or like our page on Facebook.
While you’re here, why not subscribe to our Youtube channel? Featuring discussion, debriefs, video shorts, and webinar replays.
Featured image shows scanned images of connector parts printed on eight different machines: four additive manufactured processes and two machines for each process. Image via The Grainger College of Engineering.
