CASE 06 · CASE ATLAS · LEVEL B WORKSHOP SCENARIO

Blade surface inspection

SITUATION

Blade surface visual inspection is high-volume and fatigue-prone for human inspectors. The plant wants assistance flagging candidate surface defects for human review, without removing the inspector from the accept/reject decision.

EVIDENCE

Inspection images, the defect classification and severity criteria, and prior inspector dispositions on a sample of images.

AI APPROACH

An image-based detection model proposes candidate defect locations and a severity estimate, presented to the inspector as an overlay for confirmation, not as a pass/fail verdict.

KEY CONTROLS

The model's output never bypasses the inspector; a defined escalation boundary routes any low-confidence or borderline case to a second reviewer rather than defaulting to accept.

VALIDATION MEASURES

An attribute agreement study compares model flags against inspector consensus on known-defect and known-clean images, tracking missed defects and false flags separately since they carry different consequences.

HUMAN DECISION

The inspector accepts, rejects or escalates every flagged image; the model has no authority to disposition a part.

DISCUSSION QUESTIONS

  1. Which is worse here — a missed defect or a false flag — and does the tool's threshold reflect that?
  2. How would automation bias show up if inspectors start rubber-stamping the model's overlay?
  3. What triggers a re-evaluation of the model after a defect type it has never seen appears?

TRY IT LIVE

Run this scenario's approach through Prohibited List & Signature Test — What decisions must remain human and signed?

Open Prohibited List & Signature Test →