Zero-defect ambition on the line
An automotive supplier was done relying on human eyes and sampling. They wanted every part checked, in real time.
An automotive supplier was done relying on human eyes and sampling. They wanted every part checked, in real time.
Critical components were inspected by operators and a small sample went to the lab. Defects still reached customers sometimes; when they did, the recall and reputation cost was huge. The plant had cameras and PLCs but no unified vision system that could classify defects, log them, and tie back to batch and machine.
We designed a vision stack that fits into their existing line. Cameras capture parts at key stations; images are sent to a local inference service that runs custom-trained models for their defect types — scratches, porosity, misalignment. Good/defect decisions and confidence scores are written to their MES and trigger alerts or automatic reject. The system runs on their OT network with strict access control; no images leave the factory. We trained and iterated on their historical defect data and a few weeks of new captures.
They moved from sampling to 100% inspection on two critical lines. Escaped defects dropped; when a new defect type appeared, they could retrain and redeploy in days. Quality and production now share one source of truth. The same pattern is being rolled out to another plant.