Lab success is not production success. A system that works beautifully on a bench can fall apart on a line.
What production introduces
- Vibration
- Operators with different habits
- Maintenance (or the lack of it)
- Throughput constraints
- Continuous process variation
Design choices that survive production
- No adjustment without a scale - every degree of freedom an operator can touch will eventually be touched. Replace "adjust until it looks good" with hard stops, scales, or motorized positions with stored presets.
- Self-checks on a schedule - a reference target in the field of view, or one the system can move to, turns "is the camera still okay?" from an argument into a measurement. Five seconds per shift buys you a documented imaging health history.
- Lights age - LED output falls over years and dust accumulates faster than that. Either monitor brightness against a reference or schedule lighting checks like any other consumable.
- Fail loudly, never silently - if the camera drops a frame or the trigger misfires, the part must be flagged as "not inspected", never silently passed. The most dangerous AOI failure mode is one that produces green lights while inspecting nothing.
- Logs and image archives by default - when a dispute arrives ("your system rejected good parts all night"), the archived images settle it in minutes. Without them, the discussion runs on opinion, and the system loses.
The questions I ask before calling it done
- Can it run for months without babysitting?
- Can other people maintain it?
- Can it survive process variation it has never seen?
- Can a new operator break it in their first week? (They will try, accidentally.)
- When it fails at 2 AM, can the night shift get the line running again without calling anyone?
That last one shapes more design decisions than any optical requirement. A documented "safe restart" procedure, tested by someone who did not build the system, is part of the deliverable.
Lab success means very little. Production is the real test.