Allied Vision EoSens Camera Captures the Physics of Metal 3D Printing at 20,000 Frames per Second
KU Leuven researchers use high-speed imaging to unlock defect prediction in Laser Powder Bed Fusion, paving the way for zero-defect manufacturing
Gilching, Germany, May 6, 2026 — Researchers have demonstrated that the Allied Vision EoSens 3CL high-speed camera can reliably detect process signatures linked to subsurface defects in Laser Powder Bed Fusion (LPBF) — entirely in real time, with no modification to the laser path or build machine.
LPBF is a additive manufacturing process that uses a high-powered laser to selectively melt and fuse layers of metal powder, building complex parts layer by layer. LPBF's promise of zero-defect 3D printed parts has long been constrained by an inconvenient truth: most defects originate below the layer being printed, invisible to conventional monitoring, often resulting from splatter. A new study from KU Leuven (Katholieke Universiteit Leuven, Belgium) changes that equation by correlating in-situ optical signatures captured at 20,000 fps with subsurface porosity confirmed by X-ray computed tomography.
Why Splatter Matters
Spatter is the ejection of molten metal droplets from the melt pool during laser fusion of metal powder, and it is far more than a cosmetic blight. Once airborne, spatter particles oxidize in flight, landing back on the build surface as inclusions. Cooling and resolidifying at rates mismatched with the surrounding material, it leaves behind a cascade of defects: microstructural anomalies, elevated surface roughness, irregular layer thickness, and process-induced porosity that compromise part integrity from the inside out. Spatter count, velocity, and directionality are measurable, making them primary physics-based signatures for real-time defect detection in LPBF.
Resolving Spatter Events, Frame by Frame
Mounted off-axis at 25° to the build plate, the EoSens 3CL camera required no alterations to the LPBF machine's optical path. Operating at a 30 µs exposure time through a 975 nm short-wave-pass filter, the camera captured the full visible-to-NIR emission spectrum (350–975 nm) of 316L stainless steel single-track fusions across a matrix of laser power and scan speed test conditions. Each 120 × 120 pixel frame covered a 12 mm × 12 mm field of view at ~100 µm/pixel resolution, which was sufficient to resolve individual spatter events against the melt pool in fine detail.
From the raw video stream, the KU Leuven team extracted five classes of physics-based process signatures: process zone length-to-width ratio, process zone area, process zone mean intensity, spatter speed, and spatter count. Scan speed emerged as the dominant driver of process instability with the highest zone area and aspect ratio observed under Plateau-Rayleigh instability conditions — a defect regime associated with high laser power and scan speed combinations. Droplet spatter velocities spanned 0.4 m/s to 7.8 m/s across the tested parameter space.
The EoSens 3CL's Camera Link interface, paired with an NI PXI acquisition system and synchronized with MCP controller data at 100 kHz, delivered the temporal and spatial fidelity needed to resolve individual spatter events frame-by-frame, at production scan speeds. The result is a monitoring solution that is machine-agnostic, non-invasive, and ready to serve as the sensing backbone for the next generation of autonomous, closed-loop LPBF control systems.
The headline finding: a strong correlation between spatter count from real-time EoSens 3CL monitoring and keyhole porosity count from post-process X-CT measurements. This result establishes a direct, quantifiable link between an optically observable surface signature and a volumetric subsurface defect without touching the workpiece. Surface topography measurements corroborated the camera-derived spatter data through cross-correlation template matching, and a validated finite element model accurately predicted melt pool geometry across multiple parameter sets.
Learn more about Allied Vision cameras at www.alliedvision.com.