AI visual inspection that catches defects on the line

This work was led by GSWE's founder as senior AI engineer at an automotive electronics manufacturer. Client details are confidential.

The problem

Traditional machine vision was expensive, slow to set up and inconsistent on parts that vary slightly from unit to unit. Defects that slipped through became scrap, rework or customer returns.

What we built

• Screw inspection: a YOLOv5 object detection model that confirms every mounting screw is in place and flags loose screws left on the board
• Screw height inspection: detects screws that are not fully seated
• Component checks: confirms the right capacitors and connectors are present, in the right place and the right color
• Coating detection: checks that protective coating covers the areas it should and stays off the areas it shouldn't, as a low-cost alternative to dedicated coating inspection machines
• Siamese network defect detection: compares each part with known good parts, so new defect types can be caught with very few example images
• Optical character recognition to read serial numbers and link every result to the unit

The result

Each check runs in about 200 milliseconds without slowing the line. Together the projects prevented six-figure annual scrap and returns costs, and new checks could be trained and deployed in minutes instead of weeks.