IntelliScreen
Computer vision quality inspection replacing manual QC
Duration
14 weeks
Team
3 engineers
Year
2024
An electronics manufacturer producing 50,000 PCBs per day relied on manual visual inspection for quality control. Human inspectors could only check 200 boards per hour, creating a bottleneck that slowed production and missed defects that led to a 2.3% return rate.
PCB defects are subtle — hairline cracks in solder joints, microscopic component misalignments, and variations in coating thickness. The model needed to detect defects as small as 50 micrometers while processing images fast enough to keep up with the production line running at 3 boards per second.
Built a custom dataset of 120,000 labeled PCB images, working with QC engineers to define 14 defect categories
Developed a YOLOv8-based detection model optimized for inference on NVIDIA Jetson edge devices
Created a real-time dashboard showing defect trends, yield rates, and automatic production line alerts
Implemented a feedback loop where engineer-verified corrections continuously improve model accuracy
The numbers tell the story.
Detection accuracy — higher than human inspectors at 94%
Inspection speed increased from 200 to 600+ boards per hour per line
Customer returns from quality issues dropped by 87%
“The system catches things our best inspectors miss. Our quality has never been this consistent.”
Ready to build something like this?
Every great product starts with a conversation. Let's talk about what you need.