LS ELECTRIC has cut the share of good circuit breakers wrongly flagged as defective by its robotic inspection system from 10% to zero since it introduced generative AI into product inspection at its Cheongju factory in South Korea in 2023, according to a March 2026 Dong-A Ilbo report. The plant's overall defect rate is now 7 parts per million.
"The probability of misreading a non-defective product as defective used to be 10%, but it is now 0%," a company official said.
The bottleneck was training data. A vision system has to learn what counts as a defect, and to create examples workers used to scratch or bend flawless products by hand, managing little more than 10 samples a day. Over-inspection, where good units are rejected, is the main source of waste in mass production.
In Building G of the Cheongju plant, a camera-equipped robotic arm photographs each finished circuit breaker from the top, bottom and sides, and AI analyzes the images in real time. Generative AI now takes a few real defect samples and produces thousands of varied synthetic defect images, which are used to train the computer vision model, improving both inspection accuracy and speed. The line runs on a data foundation in which automation equipment is linked to a manufacturing execution system that generates more than 500,000 data points per line per day; this data also keeps material inventory below 1.5 days of use.
The building is automated from parts supply through assembly, testing and packaging and produces 26 million industrial circuit breakers and 12 million contactors a year. More than 20 AGVs and AMRs move parts using programmed routes and AI-based real-time traffic analysis, and one or two employees per line monitor the process. LS ELECTRIC has also built a digital twin of the line by combining Microsoft's cloud, NVIDIA Omniverse and Sight Machine analytics, using ray tracing so robots can be trained in the virtual factory, and uses it to watch heat distribution inside a vacuum furnace that stays sealed for 16 hours. Linking production and energy data has cut the plant's energy use by more than 60%. "This is becoming a powerful tool for proactively responding to tightening ESG environmental regulations, particularly in the European Union (EU)," said Cho Uk-dong, head of the AP and EMEA Business Division. The company is developing an AI agent with Naver Cloud that applies AI models to power equipment diagnostics.