UNF Wins NSF Grant to Build Real-Time Defect Correction for Metal AM

Researchers at the University of North Florida have secured a National Science Foundation grant to develop an automated system that detects and corrects defects in laser powder bed fusion (LPBF) metal additive manufacturing as each layer is printed.

The problem the team is tackling is specific and expensive. In LPBF, the powder-spreading mechanism sometimes disturbs the metal bed mid-print, creating streaks that compromise part integrity. Manufacturers typically discover these defects only after a build completes, forcing them to scrap parts or restart entire jobs. The result is wasted powder, wasted energy, and a parts-per-pass rate that makes LPBF harder to justify for high-volume production.

Dr. Longfei Zhou, assistant professor of advanced manufacturing engineering, is leading the project. His team will use high-speed cameras to capture layer-by-layer process data, then train machine learning algorithms to identify streaking and other anomalies in real time. The system will not just flag defects—it will apply targeted corrections immediately, keeping the print running rather than halting it for post-build inspection.

I think the distinction between detection and correction matters here. Sigma Labs' PrintRite3D platform, deployed on DMG MORI LASERTEC machines, does solid work tracking melt pool dynamics and flagging problems during printing. Academic groups like the researchers at Aalen University's LaserApplicationCentre have made real progress using high-speed imaging to analyze spatter and smoke patterns. But most of these systems stop at detection. Zhou's team is building toward active correction, which is a harder engineering problem and a more valuable one if they can pull it off.

The project includes undergraduate researchers Maria Fernanda Ocrospoma Figueroa and Tessa Baur. Both hold leadership positions in UNF's SAMPE chapter and placed at the global Additive Manufacturing Competition at SAMPE 2026 Seattle, with Baur finishing first and Ocrospoma second. That competitive experience suggests these students understand real manufacturing constraints, not just lab-scale theory.

The open-source commitment is worth noting. Zhou's team plans to release datasets, trained models, and digital-twin software publicly. That's a deliberate choice to avoid locking the technology behind licensing deals. If the algorithms prove robust, making them freely available could accelerate adoption across the industry faster than a proprietary product would.

There are legitimate questions about scalability. A monitoring system that works in a research lab under controlled conditions may face different challenges on production floors with varying powder lots, machine wear, and environmental factors. The team hasn't published performance data yet, so claims about defect reduction remain theoretical until testing begins.

The project launches this fall. Beyond the technical work, the team will develop course modules introducing students to data-driven quality control in manufacturing. That educational component matters for building a workforce that can actually deploy these tools.

Industry Context

LPBF has become critical for aerospace, medical, and energy components where geometries are too complex for conventional machining. The process works, but yield rates often disappoint when defects go undetected until final inspection. If Zhou's system delivers on real-time correction, the economics of LPBF improve for mid-volume production runs where post-build inspection costs hurt margins.

M4S TAKE

My take: AI claims need scrutiny. The useful implementations reduce cycle time or defect rates in measurable ways. Vague promises about 'optimization' without specific metrics are usually marketing.

Simon McLoughlin