Out-of-tolerance decisions
Inspectors check what to do with a part that’s outside tolerance, from your quality procedures.
Manufacturing · Quality
Camera inspection at line speed
A vision model on the gateway checks every part and flags defects without sending images out of the plant.
- Camera gateway
- Works offline
Camera gateway assistantOn this device · no internet needed
How many rejects on line 2 this shift?
14 rejects, 11 of them surface scratches on the left side. That’s twice the usual rate.
Why was this part rejected?
Surface scratch over the allowed 2 mm on the left edge, matched to the quality standard.
When did the scratches start?
Around 14:20, after the fixture change on station 3. Before that, scratch rejects were rare.
The problem
Manual inspection at line speed misses defects, especially late in a shift. Cloud vision services need images to leave the plant and add delay at every part. Many plants do not allow production images off site at all.
Who uses it
Quality engineers, line leaders and operators, with cameras on the line and a gateway beside it.
What you train it on
- Labelled defect images
- Quality standards
- Part drawings
- Reject reason codes
How it works on the device
- 1
Trained on your defect images
Examples of good parts and each defect type, labelled by your quality team.
- 2
Inspects on the gateway
The model runs on the gateway next to the camera and checks every part as it passes.
- 3
Explains each reject
Operators and engineers ask why a part was rejected and see the defect and the standard it broke.
Why it runs on the device
Every part checked
Inspection keeps pace with the line.
Images stay in the plant
Nothing is uploaded to inspect a part.
Reasons, not just rejects
Each reject comes with an explanation people can check.
Questions
No. Inspection runs on the gateway. You choose whether any images are kept and where.
Yes. Label examples of the new defect, and the next version reaches the gateway over the air.
Only if you connect it to your line controls. On its own it flags and explains rejects.
Related use cases
All use casesPut AI where your work happens.
Pick one team and one use case. We’ll have it running on your own devices in weeks.