Automotive supplier
AI vision inspection on a high-speed stamping line
- Scope
- Vision station, edge inference, PLC reject integration, operator HMI
- Timeframe
- Pending confirmation
- Customer
- Not named — approval pending
The challenge
Surface defects on stamped parts were being found at final inspection, after several further operations had already added cost to parts that would be scrapped. Manual sampling could not cover every part at line speed, and intermittent faults passed through between samples.
What we did
- Industrial cameras and controlled lighting mounted at the forming station
- A model trained on images of the line’s own good production and its real defect types
- Inference on an edge device beside the line, with the verdict returned to the PLC within the cycle
- Reject handling wired through the existing PLC so safety interlocks were untouched
- Operator HMI showing the reason for each rejection, with a route to flag incorrect calls
Results
Measured results for this project are not published yet. Customer name and measured results awaiting written approval. Rather than print a number we cannot attribute, this section stays empty until the customer signs off on the figures and how they were measured.
Have a similar problem?
A plant assessment starts with walking your line, not with a proposal.