Vision Station Daily Checks

The question a check routine has to answer is whether the station still sees what it saw when it was accepted. Gradual failures preserve the appearance of health for a long time: parts are imaged, verdicts are issued, nothing is flagged. By the time output visibly degrades, a quantity of product has gone through the gap. Designing the check around imaging the same object and comparing the result converts that class of failure from something reconstructed afterwards into something caught early.

One reference sample beats a shelf of instruments

Keep one part, confirmed acceptable, at the station. Run it through the normal process at the start of each shift and record the result. Because the part does not change, any change in its result can only have come from the system: the light, the window, the camera position, the fixture. A light meter tells you the illumination has fallen; it does not tell you what that did to the image. The reference sample measures the thing that actually matters, which is whether the criteria still hold.

Choosing the sample needs some care. Do not pick one whose features are so obvious that any setting will pass it — such a part is insensitive to a small shift, so by the time it starts to fail the shift is already large. Pick one that sits near the edge of acceptable while being confirmed stable by a person. Its result responds first, which puts the sensitivity where it is wanted. If the station measures dimensions, choose a part whose dimension sits within the tolerance band but towards one side. If it judges appearance, choose one with typical features, a rich grey level distribution and visible sensitivity to illumination.

The limits of cleaning

Cleaning optical surfaces looks like the least skilled task at the station and is in fact one of the easiest ways to cause lasting harm. The wrong cloth leaves fine scratches that scatter light permanently. A solvent attacks coatings or the plastic of a housing. Pressing hard enough to shift a stubborn mark can nudge the camera bracket, which discards the calibration. Cleaning should be a defined procedure carried out with defined materials, and it should always be followed by a reference sample run — not to confirm the glass is clean, but to confirm that nothing else shifted while it was being cleaned.

There is a related point about lighting that surprises people: illumination often degrades unevenly. In a row of emitters, the ones nearest the supply or in the worse thermal position fade fastest, so the distribution across the field of view changes rather than the overall level. A reference sample occupying one part of the view will not show this. Stations sensitive to lighting should periodically image a uniform white or grey card filling the field and compare the grey level distribution across it. The cost is small and it covers what a single sample cannot.