NVIDIA AI inferencing helps Foxconn enhance manufacturing operations

The Foxconn Technology Group (Hon Hai) is using NVIDIA technology to inspect the surfaces of parts for defects.

While the Foxconn Automatic Optical Inspection (AOI) process still takes three to six seconds, accuracy has jumped to 99% and the manpower required has more than halved, said a booth representative at NVIDIA's GPU Technology Conference (GTC).

Machine learning and deep learning have had advantages for Foxconn as the quality of human inspection can be affected by factors such as fatigue, whereas artificial intelligence (AI) delivers consistent quality all the time, the booth representative noted.

"The AOI picks up defects consistently where a human can't do it," he said.

In a demonstration of the AOI in action, parts go through a machine and are flagged if they are imperfect. The imperfections can be miniscule, such as a millimetre-long scuff or a single pixel-sized dot of paint missing.

"We have customers (who) care about this," the booth representative explained.

The AOI uses the NVIDIA T4 Tensor Core GPU for AI inference together with the NVIDIA TensorRT platform for high-performance deep learning inference. The technology is implemented in selected factories.

The Foxconn AOI system.

Putting a component through the AOI system.

A screen at the other end of the conveyor belt displays alerts when a part fails the inspection.

Hashtag: #GTC19

*NVIDIA sponsored transport and accommodation for GTC.