UNIST and KRISS Develop AI Technology to Predict Earthquake Damage at Nuclear Power Plants with a Single Sensor

Development of Virtual Sensor AI to Reproduce Vibration Status in Nuclear Power Plant Auxiliary Buildings
Determining Priority for Power Equipment Inspection After Earthquake
Published in Computer-Aided Civil and Infrastructure Engineering

Electrical equipment such as switchboards and emergency generators, which are concentrated in the auxiliary buildings of nuclear power plants, are vulnerable to vibrations.

In fact, during the 2016 Gyeongju earthquake, the concrete buildings remained intact, but operations were halted to inspect the electrical equipment. Now, a new technology has been developed that can quickly identify which electrical equipment requires maintenance without the need for manual inspection.

On September 30, a team led by Professor Youngju Lee from the Department of Urban and Environmental Engineering at UNIST and a team led by Dr. Jaebeom Lee from the Non-Destructive Measurement Group of the Physical Measurement Division at the Korea Research Institute of Standards and Science announced that they have developed an artificial intelligence model capable of estimating the vibration status at 139 specific points within the auxiliary buildings of nuclear power plants.

Research team, (from left) Professor Youngju Lee of UNIST, Dr. Jaebeom Lee of Korea Research Institute of Standards and Science, Researcher Jingu Lee of UNIST (first author), Dr. Seungjun Lee of Korea Research Institute of Standards and Science. Provided by UNIST

The developed artificial intelligence model receives seismic data measured by a single sensor and calculates the seismic acceleration response at 139 points within the building in just 0.07 seconds.

Acceleration response is an indicator that shows how fast and strongly the equipment shook during an earthquake. By analyzing this data, it is possible to determine which equipment in which area should be prioritized for inspection.

To actually measure the seismic acceleration response at 139 points, hundreds of sensors would be required. However, the artificial intelligence acts as a virtual sensor, replacing those hundreds of physical sensors. Since there is no need to install actual sensors, maintenance costs can also be reduced.

The research team designed the artificial intelligence model with six-stage blocks, enabling it to learn a wide range of vibration patterns, from slow shaking to rapid trembling caused by seismic waves. Thanks to this, the model can accurately estimate not only the overall large movements of the auxiliary buildings but also the amplified vibrations around specific equipment.

Under noise-free conditions, the model achieved a prediction error of only 0.44 to 0.59 percent, and even in a 10dB artificially noisy environment, it maintained a low error rate of around 4 percent. Furthermore, when performance was verified using actual earthquake records (NGA-West 2), the model produced reliable estimates even under strong earthquake conditions that form the basis of nuclear power plant design safety standards in Korea and the United States.

The research team stated, "This technology dramatically reduces both the downtime caused by inspections and the burden of sensor maintenance at nuclear power plants. Especially in radiation-controlled areas like nuclear facilities, where installing and maintaining sensors is extremely limited and costly, this technology is expected to fundamentally solve these issues."

This research has also been internationally recognized for its excellence.

Existing physical sensor-based monitoring technology (left) and technology using artificial intelligence-based virtual sensors (right).

Jingu Lee, the first author and researcher, received an honorable mention in the Shitaba Award category for young researchers at the 28th International Conference on Structural Mechanics in Reactor Technology (SMiRT) for this achievement.

SMiRT (Structural Mechanics in Reactor Technology) is a prestigious academic conference in the field of reactor structures and seismic engineering. This year’s conference was held in Toronto, Canada, from August 10 to 15.

The research results were published online on September 1 in the international journal 'Computer-Aided Civil and Infrastructure Engineering,' which specializes in civil engineering.

The research was supported by the National Research Foundation of Korea under the Ministry of Science and ICT.

This content was produced with the assistance of AI translation services.