Researchers at IMDEA Materials Institute in Madrid, working with Lawrence Berkeley National Laboratory in the US, have developed an algorithm that detects hidden behavioral differences between supposedly identical 3D printers and automatically decides how to correct for them.
The work, published in Advanced Engineering Informatics, targets a problem that grows worse as manufacturing scales: even machines of the same brand, model, and near-identical serial number rarely behave exactly alike, and those small deviations compound into real defects across a production run.
Rather than applying one blanket optimization strategy across an entire fleet, the new system profiles each machine individually first. It runs a diagnostic assessment to build a performance signature for every printer, then uses statistical analysis to quantify how much the machines actually differ from one another. Based on that analysis, it makes a routing decision: if the machines are similar enough, it applies a shared optimization strategy to maximize overall efficiency; if it detects meaningful divergence, it switches to optimizing each machine individually, prioritizing accuracy over uniformity.
Testing the Theory on Three “Identical” Printers
To validate the approach, the team ran a case study on three theoretically identical 3D printers, expecting near-identical performance given matching hardware. Statistical divergence measures across the printed output, density estimates of printed pellets and pairwise divergence values between machines, showed each printer was operating in its own distinct output regime, reinforcing the case for individual rather than joint optimization in this instance.
The payoff, according to the researchers, was faster convergence and a substantial cut in errors in the weight of printed parts, compared with treating every machine as equivalent and letting individual biases go uncorrected. “Even mass-produced machines may have their own operational ‘personality,'” the team explained, noting that the system determines whether it’s more efficient to treat machines as a team or as individuals.
A Tool for More Than Just Printers
The methodology extends well beyond 3D printing. It applies to any high-throughput, parallelized experimentation that depends on nominally identical hardware behaving consistently, materials discovery, chemical synthesis, and sensor calibration among them, all fields where uncorrected equipment drift quietly undermines reproducibility. The researchers call this a step toward fully automated labs and factories that self-correct for hardware variability instead of requiring it be tuned out machine by machin
The study was carried out by Dr. Christina Schenk, who leads IMDEA Materials’ ML4Materials lab, along with Miguel Hernández del Valle, Luis Calero, and Dr. Maciej Haranczyk of IMDEA Materials, and Dr. Marcus Noack of Berkeley Lab.
The project was supported by the MAD2D-CM program on two-dimensional disruptive materials, funded by the Community of Madrid together with Spain’s Recovery, Transformation and Resilience Plan and the EU’s NextGenerationEU initiative. Noack’s contribution was backed by the US Department of Energy‘s Center for Advanced Mathematics for Energy Research Applications (CAMERA).
Closing the Reproducibility Gap in Automated Manufacturing
Most AI-for-manufacturing tools optimize what happens during a print but assume the machines are interchangeable. The IMDEA algorithm instead measures machine-to-machine variation itself and decides, per fleet, whether to optimize collectively or individually, a step earlier in the pipeline, aimed at scaled production rather than any single print.
That real-time detection once a print is already underway has been the industry’s main focus for years. Kansas State University researchers built an SVM-based system that used a camera and image processing to catch completion failures and geometric defects as they happened, while commercial tools like Printpal’s PrintWatch plugin monitor build areas in real time and can abort a job the moment a defect starts forming. Both treat each printer as its own isolated problem to solve, not as one node in a fleet with its own quirks relative to its neighbors.
Other recent work has gone the opposite direction, proving just how individual machines already are, without doing anything to correct for it. University of Illinois Urbana-Champaign researchers, working with SyBridge Technologies,trained a deep learning model that could trace a printed part back to its source printer with 98% accuracy, using surface textures invisible to the human eye. That study confirmed the same “personality” IMDEA’s team set out to correct for, it just used it for traceability instead of optimization.
The IMDEA and Berkeley Lab system sits between those two threads: it doesn’t watch for defects mid-print, and it doesn’t just fingerprint machines after the fact, it uses machine identity itself as an input to decide how a fleet should be run.
3D Printing Industry is inviting speakers for its 2026 Additive Manufacturing Applications (AMA) series, covering Energy, Healthcare, Automotive and Mobility, Aerospace, Space and Defense, and Software. Each online event focuses on real production deployments, qualification, and supply chain integration. Practitioners interested in contributing can complete the call for speakers form here.
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 LinkedIn.
Explore the full Future of 3D Printing and Executive Survey series from 3D Printing Industry, featuring perspectives from CEOs, engineers, and industry leaders on the industrialization of additive manufacturing, 3D printing industry trends 2026, qualification, supply chains, and additive manufacturing industry analysis.
Featured image shows Graphical abstract. Image via IMDEA Materials Institute.
