Industrial vision: verify first, then deploy
Visual inspection on a factory floor is not a demo: for every image judged defective, it must be possible to replay why it was judged so, who reviewed it, and which work order it entered. Black-box scores do not belong on a factory floor. Negency's approach is to state the method first, then prove it with a controlled pilot — pilot data decides whether the scheme continues.
The current status, stated plainly: this page is a technical scheme and a pilot method, not delivered capability. The first offline validation run is complete (2026-11-13, DeepPCB public benchmark: AUROC 0.627, detection rate 11.8%, false-positive rate 6.6% — weak numbers, a measured floor, not a capability claim); no real-device testing has taken place, and no edge-device pilot or site acceptance has begun. The verification record is published in the industrial vision sample in the Engineering Lab — what has been measured, and what is still missing, is written there.
The six-step loop
Image capture (industrial cameras and sensors) → edge inference (models running locally on edge compute modules) → defect analysis (defect-type judgement + confidence + evidence image) → evidence archiving (original image + inference result + review record, kept together) → human review (a mandatory step — the machine never clears a defect alone) → work order (confirmed defects enter the work-order and maintenance system). Deployment shape follows the client site: public cloud, client private cloud, on-premises, or edge devices can all be assessed; data can stay inside the plant. The reference architecture is diagrammed below.
Offline validation first
Before a model reaches an edge device, it is validated offline on the test set: detection accuracy, false-positive rate, and miss rate are measured, with thresholds set by acceptance metrics, not by demo effect. Datasets are licensed public datasets or client-authorised data only; defect categories and labelling rules are fixed before model selection; train, validation, and test sets are separated, and the test set plays no part in tuning before acceptance. Only after offline validation passes does the model move to an edge-device pilot.
Pilot path: four gates
Dataset selection → offline validation → edge-device pilot → site acceptance. Each step gates the next; pilot results decide whether site acceptance proceeds. Acceptance metrics (detection accuracy, false-positive rate, miss rate, per-image inference latency, continuous-run stability) and their thresholds are agreed with the client before the pilot starts and go into the acceptance document; a measurement that misses reworks the scheme instead of lowering the bar. The pilot plan and acceptance gates are tabulated below.
Capability boundaries
No promises on detection accuracy, false-positive rate, or deployment schedule — those await pilot measurement. An inspection verdict never triggers equipment action directly; no defect verdict reaches a work order without human confirmation; safety-related judgements are never left to the model alone. Equipment certification, functional safety, cybersecurity, and import/export compliance must be checked per product; no device is promoted as "fully compliant with EU and US regulations".
