Additive manufacturing
In-situ monitoring for additive manufacturing
Per-layer vision monitoring that catches print defects early on long unattended builds.
- -60%scrapped builds
- per-layercapture + scoring
- ~30 sto a flagged anomaly
Stack
- Machine-vision camera
- Layer-sync trigger
- Anomaly detection (autoencoder)
- Auto-pause control hook
- MQTT to MES
The problem
A production additive shop lost machine-hours to long prints that failed late, warping, under-extrusion, detached parts, discovered only when someone walked the floor. They wanted early detection, an automatic pause, and a per-layer record for qualification.
Approach
- 01 Triggered a camera capture on each layer-change event from the printer's firmware.
- 02 Scored every layer against a learned model of a healthy build and flagged anomalies.
- 03 Paused the machine and alerted the cell on a sustained defect rather than a single noisy frame.
- 04 Kept a per-build image log so qualified parts carried their own process evidence.
Outcome
Most failing builds were now caught in the first layers instead of at the end, cutting scrapped material and freeing machine time. The per-layer log became part of how the shop qualified parts, not just a debugging aid. Tuning the alert threshold to ignore single-frame noise was what made it trusted.
