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.
- 71
Power, mounting and thermals: fitting an inspection unit to a real cell
The mechanical and electrical envelope a cell designer needs: universal 90-240 VAC at 60 W, under 9 kg, 320x240x180 mm, fanless, 0-45 C (0-65 C on the H100).
July 2026 · 8 min read
- 72
Questions to ask any machine vision vendor before you buy
The buyer questions that expose what a demo hides: false-negative rate, training images and time, data location, protocols and I/O, and the enclosure IP rating.
July 2026 · 7 min read
- 73
Machine vision vs computer vision vs AI vision: the terms, untangled
A factory-floor glossary of three terms buyers use interchangeably, computer vision, machine vision and AI vision, and where a hybrid edge-AI unit fits.
July 2026 · 6 min read
- 74
Modbus TCP as the neutral integration path for AI inspection
Why Modbus TCP is the vendor-neutral way to carry an AI inspection result to almost any PLC: a small register map for result, confidence, part-count and status.
July 2026 · 8 min read
- 75
Model drift and retraining: keeping an inspection model accurate over time
Why an inspection model drifts as lots, tools and lighting age, and the recapture-retrain-redeploy loop that keeps false rejects from creeping up, all on the edge.
July 2026 · 8 min read
- 76
Natural variation vs true defects: teaching a model what to ignore
Build natural variation into the good set so a model ignores normal spread and flags only true defects. Fewer false rejects from about 20 good images.
July 2026 · 8 min read
- 77
OCR, barcode and label verification with AI vision
How hybrid vision reads barcodes and OCR text and judges label and print quality, what Adente Vision confirms, and why to verify specific support on the datasheet.
July 2026 · 6 min read
- 78
OEM-embedding an edge-AI inspection unit into your own machine
Ship inspection as a standard feature by embedding one fanless edge-AI unit: 320x240x180 mm under 9 kg, triggering, 4 I/O at 24V and five protocols.
July 2026 · 6 min read
- 79
Fewer moving parts: why one edge box beats a camera-plus-server stack
A distributed vision rig adds a grabber, a PC, a GPU and cabling, each a failure point. Why one 320x240x180 mm enclosure under 9 kg is the reliability play.
July 2026 · 7 min read
- 80
OPC UA for inspection: when a semantic protocol beats raw I/O
When an inspection result must reach MES as named, self-describing data, not one pass/fail bit, OPC UA is the carrier. Per-part confidence and millimeter measurements, published from the edge unit itself.
July 2026 · 6 min read