Zero-Defect Production Line
a tier-one automotive parts supplier
defect detection rate
fewer customer warranty claims
less manual inspection time
Manual inspection caught defects inconsistently, and the ones that slipped through surfaced as warranty claims and chargebacks from OEM customers. Inspectors could not keep pace with line speed, so sampling replaced full coverage and marginal parts shipped. Each escaped defect risked a customer audit, and leadership needed inspection that scaled with production instead of slowing it down.
How we approached it
Instrumented three production lines with cameras and audited six months of defect and warranty data to build a labeled dataset.
Trained computer vision models on the line to detect surface and dimensional defects in real time, running inference on edge hardware to keep pace with cycle times.
Built an active learning loop where inspectors confirm or correct borderline calls, feeding every judgment back into retraining.
Integrated detections with the plant's MES so flagged parts divert automatically and quality engineers see defect trends by station and shift.
“The line does not slow down and defects do not get through. Our quality engineers went from arguing about escapes to preventing them.”
The models were benchmarked against a held-out set of known defective parts and audited weekly against inspector findings, so the rate reflects production conditions rather than lab tests.
No. Inference runs on edge hardware at each station and returns a pass or divert decision within the existing cycle time, so throughput was unchanged.
Borderline parts route to a human inspector, and every one of those judgments feeds back into retraining, so the uncertain zone shrinks with each release.
Want results like these?
Book a free consultation and we'll map the highest-leverage place to start for your team.