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.
- 51
What is a golden sample in visual inspection?
A golden sample is the reference that defines a good part. In few-shot anomaly inspection it is a set of about 20 good images, not one master part.
July 2026 · 7 min read
- 52
High-temperature inspection for foundries, glass and hot stamping: the AV-H100
Near furnaces and presses, ambient air, not part temperature, limits the camera. The AV-H100 raises the range to 0-65 C, IP54 and fanless, with the same optics and AI.
July 2026 · 8 min read
- 53
What is hybrid machine vision? Classical CV plus deep learning
How hybrid vision pairs classical CV for measurement with an AI layer for judgment, an auditable, non black-box design, and which checks belong to each half.
July 2026 · 7 min read
- 54
In-house camera vs third-party camera in an AI inspection unit
What you gain when the camera, lens, lighting and AI are co-designed in-house rather than a third-party camera bolted onto separate software.
July 2026 · 8 min read
- 55
Pilot to production: acceptance-testing an inspection system before you trust it
A defensible acceptance test for an inspection pilot: run seeded defects, measure F1 and false-negative rate on your own parts, and add a gauge R&R check before sign-off.
July 2026 · 7 min read
- 56
PROFINET vs EtherNet/IP vs Modbus TCP: choosing a fieldbus for an inspection cell
Pick the fieldbus that carries inspection pass/fail into your PLC: PROFINET on Siemens, EtherNet/IP on Rockwell, Modbus TCP neutral, matched to the controller you already run.
July 2026 · 6 min read
- 57
Product changeover and seasonal switchover on an inspection station
Save a configuration per part, geometry, lighting, threshold and trigger, recall it to switch in minutes, validate the first parts, and retrain from 20 images only when a variant is new.
July 2026 · 8 min read
- 58
The cycle-time budget: fitting inspection into a fast index line
Budget inspection cycle time against takt: capture, inference and I/O settle per part, and how a measured ~30 ms fits under a 100+ parts-per-minute line.
July 2026 · 7 min read
- 59
Keeping inspection images inside the factory: a data-residency guide
A raw part image is sensitive IP. On-device inference classifies each frame on the line and emits a pass or fail, so images never ship to a vendor cloud.
July 2026 · 7 min read
- 60
What you need on the line before the inspection unit arrives
A site-prep checklist for an edge-AI inspection unit: power, mounting, trigger, network and enclosure variant, so a 30-minute install stays on schedule.
July 2026 · 8 min read