Korea Research Institute of Standards and Science (KRISS) said on the 26th it developed a Deep Learning technology that predicts, in real time, shaking at multiple points inside a nuclear power plant using the signal from a single seismometer and quantifies the risk.
When an earthquake occurs, operations can be halted for a long period during safety inspections even if facilities suffer no major damage. In fact, during the 2016 Gyeongju earthquake, Wolseong Nuclear Power Plant units 1–4 resumed operations about 80 days later after performance tests and detailed inspections.
However, it is not easy to determine which equipment to inspect first right after an earthquake because the information available is limited. Installing sensors at key points in a nuclear power plant faces constraints such as wiring, licensing, and maintenance, and computational analysis takes a long time to produce results.
To address this, the KRISS research team developed a virtual sensing technology that analyzes seismic waves measured by a single seismometer with artificial intelligence (AI) to estimate the vibration responses at 139 points inside a nuclear power plant where sensors are not installed. They also verified predictive performance using real earthquake records not included in training.
The technology developed this time also calculates, from 0% to 100%, the likelihood that vibrations at each point will exceed a preset risk threshold. Using this, one can not only distinguish between safe and risky but also refer to it to set inspection order starting from locations with higher risk likelihood.
The researchers also proposed a method to design the AI model architecture based on the natural frequencies of the structure. They reduced the number of parameters significantly compared with conventional large Deep Learning models while securing similar accuracy, lowering computational burden.
The team said the technology could be applied not only to nuclear power plants but also to earthquake‑sensitive industrial facilities such as semiconductor factories, data centers, and plants.
Senior researcher Lee Jae-beom of KRISS said, "In safety AI, it is important not only to predict accurately but also to recognize the limits of judgment," and added, "Based on this, we will develop trustworthy AI technology that informs uncertain situations and helps human decision-making."
The results were published in international journals including "Reliability Engineering & System Safety."
References
Computer-Aided Civil and Infrastructure Engineering (2025), DOI: https://doi.org/10.1111/mice.70051
Reliability Engineering & System Safety (2026), DOI: https://doi.org/10.1016/j.ress.2026.112582
Nuclear Engineering and Design (2026), DOI: https://doi.org/10.1016/j.nucengdes.2026.115114