AI vision quality control on high-speed electronics lines

The challenge

A manufacturer of precision smartphone components needed inspection that could run at full line speed. Human inspectors missed defects at that pace, while traditional machine vision struggled with new defect types, lighting changes and material variation, rejecting good parts and stopping lines.

Our approach

We deployed vision-transformer models on edge hardware beside each line, classifying defects such as micro-cracks, contamination and assembly errors frame by frame. An anomaly detector flags defects the model hasn’t seen before for human review, and verified rejects feed retraining without stopping production.

Inspection results go to the MES, which diverts defective parts and quarantines suspect supplier lots, and supplier scorecards push root-cause fixes upstream.