We're planning an FR5 bezel inspection station. Small presentation changes move the bright streaks, but the captions keep calling them scratches. I want the physical reference decisions and camera conditions sorted before we tune anything. What should I separate so the image folder stops acting like an inspection authority?
Separate the physical bezel judgement from the image observation, with identified retained samples linking the two. Then compare repeat seating and controlled presentation. I mean actual seating, not simply returning the robot to the same nominal pose; those can differ.
Quality examined the retained bezels and accepted two broad streaked regions, but rejected a small edge nick on another. The reference links are corrected. Our nominal camera pose hides that nick, so the imaging specialist is assessing view coverage before anybody starts arguing about a threshold.