Catching visual defects on a line with a fixed camera

A fixed camera photographs every unit as it passes; a model trained only on defect-free examples scores how far each image departs from them, and an operator adjudicates whatever it flags. The rig — lens, fixture, lighting — is part of the system, not a detail arranged around it.

Effort
More than a few weeks
Skill level
Some technical skill
Organisation size
Mid-market
Value
Quality, Cost saved

Tools named for this

  • A fixed machine-vision station with controlled lighting and a repeatable part fixture
  • An anomaly-detection model trained on defect-free images only
  • A review queue in which an operator adjudicates every flagged unit and labels the reason

What to check before you ship it in India

  • A camera pointed at a line photographs the people working on it, so the frames are personal data whatever the model is looking for. Sending them to a supplier's cloud makes that supplier a processor. Section 8(2) reaches a processor engaged for an activity related to offering goods or services to Data Principals, and your own operators are not that; the clause that bites is section 8(1), under which you stay responsible for compliance in respect of processing carried out on your behalf, irrespective of any agreement to the contrary.
  • Trained on defect-free images alone, the model can tell you something is unusual, never that it is a defect. A new supplier's finish, a changed label or a fresh batch of raw stock all read as unusual, and a line that halts for those will be switched off by the second shift.

Sources

Every claim on this page traces to one of these, on the date it was read.