An electronics manufacturer deployed edge-based computer vision across 6 SMT lines — analyzing high-resolution images on the conveyor in 14 milliseconds, flagging defects with 99.2% accuracy and eliminating inspection bottlenecks entirely.
BOARDS / MIN
14
DEFECTS / HR
3
ACCURACY
99.2%
True positive rate across 12 defect classes
Faster than manual AOI inspection
Line speed no longer limited by inspection
End-to-end inference on edge device
The manufacturer's 6 SMT lines produced 12,400 PCBs daily. Legacy AOI systems were slow, generated excessive false positives, and required operators to re-inspect 30% of boards — creating a bottleneck that capped throughput.
AOI false positive rate: 18%
Operators wasted hours re-checking good boards
Missed micro-defects
0201 components and hidden solder joints escaped detection
Line speed capped at 80%
Inspection was the slowest step in the entire workflow
$3.8M annual warranty claims
From defects that shipped to customers
AOI false positive rate
Boards re-inspected manually
Legacy inspection time / board
Annual warranty claims
Three layers — high-res capture, edge inference, and instant action — replace legacy AOI with a system that's 10x faster, more accurate, and learns continuously.
5MP global-shutter cameras with ring lighting capture 2048×1536 images at 60 fps. Multi-angle illumination reveals hidden solder joints and micro-defects.
Custom YOLOv8 model, quantized to INT8 and deployed on NVIDIA Jetson Orin. Runs 12-class defect detection locally in 14ms — no cloud round-trip.
Defective boards are auto-rejected via PLC signal. Alerts stream to MES with defect type, location, and image. Continuous learning from operator feedback.
From micro-solder bridges to missing 0201 capacitors — every defect type the legacy system missed.
Solder Bridge
99.4%
Missing Part
99.1%
Tombstoning
98.8%
Polarity Flip
99.6%
Insufficient Solder
98.5%
Micro-Crack
97.9%
5MP camera captures board at 60fps with multi-angle lighting.
Resize, normalize, augment on GPU. 2ms from raw to tensor.
INT8-quantized model runs on Jetson Orin in 10ms.
NMS, confidence thresholding, defect classification.
PLC rejects board. Image + metadata sent to MES.
Line speed limited by inspection · 30% boards re-checked manually
Inspection no longer the constraint · Line runs at full speed
"We used to slow down the line because inspection couldn't keep up. Now the vision system is the fastest step in our workflow. We run at full speed, catch more defects, and haven't had a warranty claim from an escaped defect in 9 months."
Akiko Kimura
Director of Manufacturing, NexCirc Electronics
Warranty savings
Throughput increase
ROI payback
3-year ROI multiple
Zero escaped defects
9 consecutive months
First time in company's 22-year history
See how edge computer vision can eliminate defects, remove bottlenecks, and unlock full line speed on your SMT lines.