I've paused threshold tuning on our faceplate plan. The brightest streak shifts with presentation, but the example labels call it a scratch. Need quality's physical references and a controlled imaging comparison before this becomes commissioning evidence.
Can you link the views to individual retained faceplates? I'd start with a small set you can actually put on the bench. A hundred impressive pictures won't help if nobody knows whether they show one plate or twenty.
Four plates retained and numbered. Two views of one accepted-looking plate have different scratch labels, but accepted-looking is my description, not a quality decision.
Keep that wording clear until quality judges it. Need the actual appearance requirement, viewing method and physical decision attached to the sample, not just another round of votes on the image.
Can the inspection team also record how the plates were seated for each view, so maintenance has something concrete to reproduce when the presentation changes later?
Could the label annotation itself hide the feature people are meant to compare? I would keep an unannotated view available beside any marked defect location.
Naomi, seating wasn't recorded in the old examples. Eli, one caption overlaps the bright streak. I've retained the raw image and stopped using that captioned version as a teaching reference.
A moving highlight doesn't acquit the whole faceplate, either; you can have reflection and a genuine scratch in the same image, so keep the physical judgement separate from deciding which bright feature moved.
We used one disputed plate as the entire training exercise and just rehearsed the same argument, so I'd add known acceptable finish variation and a confirmed reject before asking anybody to tune this.
Quality has judged the four retained plates now. One reject has a physical scratch away from the brightest moving stripe. Nora, that was exactly the distinction we were missing.
That sounds like a useful comparison piece. I'd ask the vision person to vary presentation and lighting separately around it, with the actual defect location retained. Then you can see which condition hides the scratch rather than just improves the photograph.
Our inspection lead will own the reference set with maintenance holding the fixture and lighting revisions. Controlled imaging work hasn't finished, so we still have no threshold set to approve.
That should also help the trainer show an unambiguous comparison. I would keep the physical defect marker separate from captions explaining why another bright stripe is not the decision.