Domestic researchers develop a technique that quickly predicts the internal damage state of nuclear power plant buildings after an earthquake using a single sensor. The photo shows the devastating damage from the Great East Japan Earthquake in 2011 that led to the Fukushima nuclear accident. /Courtesy of Greenpeace

A domestic research team has developed a technology that quickly predicts the damage status inside a nuclear power plant building with just a single sensor when an earthquake occurs.

Ulsan National Institute of Science and Technology (UNIST) said on the 30th that a team led by Professor Lee Young-ju of the Department of Earth Environmental and Urban Construction Engineering, together with a doctoral team led by Lee Jae-beom of the Nondestructive Measurement Group, Physical Measurement Division, at the Korea Research Institute of Standards and Science (KRISS), developed an artificial intelligence (AI) model that estimates the vibration status at 139 detailed points inside an auxiliary building at a nuclear power plant.

Important electrical equipment such as switchboards and emergency generators are concentrated in the auxiliary building, and these are particularly vulnerable to earthquakes. In fact, during the 2016 Gyeongju earthquake, the building itself remained intact, but operation of the nuclear plant was suspended due to inspections of electrical equipment.

The newly developed AI model calculates the building's overall vibration status in 0.07 seconds based on earthquake data measured by a single sensor. This "acceleration response" information indicates how fast and strongly the equipment shook, allowing inspectors to quickly single out areas with a high likelihood of damage for checks. There is no need to install sensors at each point as before, greatly reducing maintenance and repair expense.

Conventional physical sensor-based monitoring technology (left) and artificial intelligence-based virtual sensor technology (right). Research illustration. /Courtesy of UNIST

The researchers designed the AI model in six-stage blocks so it could learn a variety of vibration patterns, from slow swaying to fast trembling. In performance tests, the prediction error was only 0.44%–0.59% in a noise-free environment, and it stayed around 4% even when artificial noise was added. Experiments applying actual earthquake records also produced reliable results that meet nuclear power plant design standards in Korea and the United States.

The research team said, "With this technology, we can significantly cut downtime from nuclear plant inspections and the burden of sensor maintenance," adding, "In particular, in radiation-controlled areas, installing and maintaining sensors is difficult and costs a lot of expense, and this can fundamentally solve that."

This study was published in the latest issue of the international journal in civil engineering, "Computer-Aided Civil and Infrastructure Engineering." Researcher Lee Jin-gu, the first author, was selected in the young researcher award category at the 28th International Conference on Structural Mechanics in Reactor Technology (SMiRT) with this achievement. SMiRT, a world-renowned society in reactor structures and seismic design, was held in Toronto, Canada, in Aug. this year.

Lee Young-ju (from left) Professor at UNIST, Lee Jae-beom Dr. at Korea Research Institute of Standards and Science (KRISS), Lee Jin-gu UNIST researcher (first author), and Lee Seung-jun Dr. at Korea Research Institute of Standards and Science (KRISS) develop the earthquake damage prediction AI model. /Courtesy of UNIST

References

Computer-Aided Civil and Infrastructure Engineering (2025), DOI: https://doi.org/10.1111/mice.70051

※ This article has been translated by AI. Share your feedback here.