The client monitored high-speed assembly lines where manual inspection of every part was impossible. Defects like scratches or misalignment were often missed, leading to costly recalls.

They needed a solution that could:

• Detect specific defect types (scratches, dents).
• Locate defects precisely using bounding boxes.
• Run at edge latency speeds.

We implemented Azure Custom Vision Object Detection.

Unlike simple classification which just labels an image, Object Detection provides bounding boxes, allowing the system to pinpoint exactly where the defect is located on the part.

The model was trained on a curated dataset of defect images and deployed to Azure IoT Edge containers to ensure millisecond-latency analysis directly on the factory floor.