Intel and John Deere have announced a pilot program that leverages artificial intelligence and computer vision to spot defects in the welding process.

The problem

John Deere uses Gas Metal Arc Welding (GMAW) to produce machinery at over 52 factories worldwide. At this volume, the company needs to keep careful watch for any issues like porosity, a defect that occurs when trapped gas leaves cavities in the weld metal as it cools.

Porosity is a big problem in John Deere’s manufacturing. Currently, it requires highly skilled manual work to identify. On top of that, it can be quite expensive if not caught early enough. Flaws and errors early in the process can require expensive re-work.

The AI solution

John Deere and Intel have developed a technology stack that includes Intel processing hardware for edge computing and an industrial-grade machine vision platform. The center of the solution is a neural network trained on good welds and welds with porosity defects.

“To detect porosity defects in real time,” Intel explains in a white paper on the solution, they “put a camera where no human could go: on the welding gun, just 12 to 14 inches away from the actual weld.” Using frames from those cameras, the system examines the welds to detect porosity where humans can’t. When it detects a defect, it shuts down the welding robot.

Why it matters

Real-world, practical applications of a technology are a good indicator of its maturity, and its position on the hype cycle. This news shows that AI and CV—two important technologies for spatial computing applications—could be leaving the “trough of disillusionment” as it starts to deliver real value to businesses.