Manufacturing: Quality Control with Computer Vision

A mid-size automotive supplier produces over 50,000 plastic parts daily. Defective parts must be reliably identified — previously through manual visual inspection, with the associated error rate and personnel overhead.

The Challenge

The production hall has no stable internet connection. Cloud AI was therefore not an option — not just for data protection reasons, but also due to latency requirements. A quality system must decide in under 200ms.

SoverIQ Box at the Line End

The SoverIQ Box — a hardened edge device — was installed directly at the inspection station. A locally running computer vision model analyses camera images in real time for:

• Surface defects and scratches
• Dimensional deviations
• Missing connection points

No image leaves the production facility. The model runs completely offline.

Result

The defect detection rate rose from 76% to 98%. Scrap was reduced by 40%. The Box paid for itself within four months — through saved warranty cases and reduced inspection overhead.