The useful search phrase is still machine vision inspection
Most factories do not search for abstract AI. They search for machine vision inspection, camera quality control, missing part detection, barcode reading, OCR, object detection, or visual inspection automation. The practical system is a camera, lighting, trigger, inspection logic, operator screen, pass/fail output, and a record of what happened.
Where modern AI helps
Classical image processing is still excellent for measurement, alignment, clean barcode reading, and controlled presence checks. AI helps when parts vary, defects are visual, products are organic, or the system needs object recognition, classification, or tolerance to messy real-world images.
A sensible first pilot
Start with one station, one camera angle, one defect category, and a clear OK/NOK rule. Capture real images, measure false rejects and false accepts, then decide whether the solution needs OpenCV rules, a custom model, better lighting, a fixture, or a human review step.