Jakarta – Japan is one of the countries most prone to earthquakes. This reality has driven Japanese scientists to innovate and develop technology to map the potential impacts of these disasters.
One key focus of their efforts is predicting liquefaction, a condition where soil loses its strength, often causing severe damage during earthquakes.
During liquefaction, the soil behaves like a liquid, causing structures above it to shift or sink. For this reason, predicting liquefaction is a critical priority for scientists.
Technological Innovations to Predict Liquefaction
Japan has faced numerous major earthquakes, such as the 2011 Tohoku earthquake, which caused liquefaction that damaged 1,000 homes. Similarly, the Christchurch earthquake severely impacted 80% of its water and sewage systems, and the 2024 Noto earthquake damaged over 6,700 homes.
To mitigate the effects of liquefaction, Professor Shinya Inazuma and student Yuxin Cong from the Shibaura Institute of Technology, Japan, developed an advanced machine learning model to predict how soil will react during earthquakes.
The model uses geological data to create a more accurate 3D map of soil layers and identify areas more susceptible to liquefaction. This method surpasses traditional manual soil tests, which are typically limited to specific locations.
The new approach offers a broader and more detailed understanding of soil behavior across various areas.
In a study published in Smart Cities on October 8, 2024, researchers utilized artificial neural networks (ANN) and ensemble learning techniques to predict the depth of bearing layers, a key indicator of soil stability.
“This study establishes a high-precision prediction method for unknown points and areas, showcasing the significant potential of machine learning in geotechnical engineering. This enhanced model facilitates safer and more efficient infrastructure planning, particularly crucial in earthquake-prone regions,” said Prof. Inazuma, as quoted from shibaura-it.ac.jp.
Predicting Bearing Layer Depths to Identify Stable Locations
Researchers collected data from 433 points in Setagaya-ku, Tokyo, using standard penetration tests and mini-ram sound tests to determine the depth of the bearing layers. This data was then used to train the artificial neural network to predict bearing layer depths in untested locations.
To improve the model’s accuracy, the researchers also applied a bagging technique (bootstrap aggregation), which involves training the model on several subsets of the training data. This method increased prediction accuracy by 20%.
Using the data obtained, they created contour maps depicting the depth of bearing layers within a 1 km radius around four selected locations in Setagaya. These maps serve as valuable visual aids for civil engineers, helping them select construction sites with stable soil conditions.
“This predictive model facilitates safer and more efficient infrastructure planning, which is essential for earthquake-prone regions,” Prof. Inazuma explained.
Long-Term Goals of AI-Based Technology
For researchers, this approach holds significant potential for reducing liquefaction risks and enhancing urban resilience to natural disasters. By leveraging artificial intelligence (AI) in geotechnical analysis, smart cities in Japan can be designed more efficiently and cost-effectively.
The maps produced allow disaster management experts to identify areas more vulnerable to liquefaction, enabling better risk assessment and mitigation strategies.
“This study lays the foundation for safer, more efficient, and cost-effective urban development,” said Prof. Inazuma.
In the future, researchers plan to develop more accurate models by incorporating additional soil conditions and creating specialized models for coastal and non-coastal regions.
