Blog · AI visual inspection
Notes from the inspection line.
Anomaly detection, few-shot training, edge deployment and the metrics that matter: practical writing for integrators, quality managers and automation engineers.
- 61
Who owns the unit after install: maintenance and keeping accuracy stable
What drifts on an inspection station over months, a light maintenance cadence for a fanless edge unit, and the operator-integrator-Adente ownership split that holds launch accuracy.
July 2026 · 7 min read
- 62
Measuring inspection uptime: the availability metric behind your OEE
Inspection uptime is availability for your vision station, separate from line uptime and OEE. Define it as a KPI and see how a fanless edge unit protects it.
July 2026 · 8 min read
- 63
Inspection uptime when the network goes down: why edge keeps running
Network dependence is an availability risk. A self-contained edge unit captures, decides pass/fail and logs on-device, so an IT outage never ships uninspected parts.
July 2026 · 7 min read
- 64
IP54 vs IP65 vs IP66: choosing the right enclosure rating for your line
What the two IP digits mean per IEC 60529 and how IP54, IP65 and IP66 map to the four Adente Vision variants. Same optics and AI inside, only the enclosure changes.
July 2026 · 7 min read
- 65
Fanless Jetson-class edge compute: doing the inference at the camera
Running the model on a fanless Jetson-class board at the camera is what makes on-device inference of about 30 ms per part possible with no PC and no cloud.
July 2026 · 8 min read
- 66
Why line-speed anomaly detection cannot live in the cloud
Anomaly detection runs on every part under a tight per-part budget, and a cloud round-trip adds unbounded jitter. Why the decision has to stay on the edge.
July 2026 · 7 min read
- 67
One enclosure, one software stack, four environments
Four variants (AV-S100/W100/H100/X100) share identical optics, compute and AI modes and differ only in the enclosure, so one integration covers four environments.
July 2026 · 7 min read
- 68
Diffuse, directional or coaxial: lighting geometry for AI inspection
Lighting geometry decides whether a defect appears in the image at all. How diffuse, directional and coaxial light reveal different flaws, all configurable in one 24V unit.
July 2026 · 8 min read
- 69
Machine vision market trends for 2026: edge AI, few-shot and the integrator channel
Where industrial vision is heading in 2026: on-device inference, few-shot and anomaly training from about 20 images, and turnkey units delivered through integrators, tied to verifiable capabilities.
July 2026 · 7 min read
- 70
Triggering an inspection: encoder pulse vs photoelectric sensor vs fixed interval
How you trigger an inspection decides if the part is in frame: encoder for conveyors, photoelectric for part-present, fixed interval for indexed lines, with a 12 MP global shutter to freeze motion.
July 2026 · 6 min read