Solutions

AI surface defect inspection for electronics manufacturing

Detect scratches, cracks, burrs and print defects on metal and plastic surfaces inline, with 99.4% accuracy.

Why visual inspection is no longer enough

Surface defects are often smaller than 0.1 mm and only become visible under a specific lighting angle. Human inspection depends on the operator, is not repeatable, and cannot keep up with cycle times under two seconds per part.

System architecture

Image acquisition

12 - 24MP global shutter cameras with telecentric lenses remove perspective error. Dome, coaxial and low-angle lighting are switched per cycle so each defect class is revealed.

AI processing at the edge

An Edge AI computer runs a segmentation model trained on the customer's own images, grades defects by severity and returns an OK/NG signal to the PLC within 120 ms.

Traceability and improvement

Every NG image is stored with its defect label, feeding Pareto root-cause analysis and periodic model retraining.

Recommended hardware

  • 24MP 10GigE industrial camera (DEMO-CA-2450M)
  • 1/2.5" C-Mount telecentric lens
  • Dome and low-angle lighting with a 4-channel controller
  • Edge AI computer with a real-time inference GPU

Results

99.4% detection rate, under 0.5% false calls, fully replacing two manual inspection stations per shift.