Our FR10 camera proposal has confident scratch labels on bright streaks that move when a faceplate is presented differently. No agreed physical decisions behind them yet. How do I build a useful reference set without teaching the inspection to agree with our first guesses?
Keep each faceplate identified and have quality inspect its actual surface under an agreed method. Include accepted finish variation. Then the reference set can show why a bright region was accepted or rejected, not only where somebody drew an arrow.
The physical plates survived, so that is possible. Our current folder has several images of the same plate with different labels, but no record that quality made either decision.
Who is keeping the plate-to-image link? Give one person ownership. Otherwise a renamed file and a moved sample can undo the careful inspection without anyone noticing until training starts.
Beth's question matters for the next comparison. I would repeat one seat with fixed conditions first, then change presentation deliberately. It is easier to explain a result when the loader and lamp have not both changed.
Fiona, I own the sample register and image links. Beth, the angles were changed deliberately; the holder gives a repeatable seat in our existing placement check. The folder failed to preserve which angle produced each image.
Then keep those old images as observations with missing conditions. Don't retrospectively assign an angle because a highlight looks familiar. The new capture can be better without rewriting what you know about the old one.
Quality accepted the broad bright region and rejected a small edge gouge on another plate. We have captured identified views at the proposed inspection pose. The gouge is partly hidden there, which is a more useful problem than our original bright-streak argument.
Another view is being assessed; quality has not agreed to omit the edge. Thank you, Isabel, for insisting on accepted examples too. We now have physical decisions, but the camera plan still lacks demonstrated coverage of the actual rejected feature.