Machine Learning Models Enhance Urban Resilience Against Liquefaction in Seismically Active Regions

This article discusses the development of machine learning models by Professor Shinya Inazumi and student Yuxin Cong, which enhance resilience to liquefaction in earthquake-prone areas. These models utilize geological data to predict soil behavior during earthquakes, aiming to improve infrastructure planning and urban safety. Their research has demonstrated increased accuracy in predicting soil layer depths, which is vital for assessing liquefaction risks and ensuring safer urban development.

As urban areas continue to expand, the risks posed by natural disasters necessitate urgent action from city planners and disaster management officials. In earthquake-prone countries, particularly Japan, liquefaction—where intense seismic shaking leads to the destabilization of loose, water-saturated soils—is a significant threat to infrastructure integrity. This phenomenon can result in catastrophic consequences, including the sinking of buildings, compromised foundations, and the failure of vital utilities such as water and sewage systems. Historically, significant earthquakes have demonstrated the destructive potential of liquefaction. Notably, the 2011 Tōhoku earthquake severely impacted 1,000 homes through this phenomenon. Similarly, the 6.2 magnitude earthquake in Christchurch resulted in widespread liquefaction damage, affecting 80% of the city’s water and sewage systems. In 2024, the Noto earthquake led to extensive liquefaction, directly impacting 6,700 residences. To address these pressing challenges and enhance urban resilience against liquefaction, Professor Shinya Inazumi and his student Yuxin Cong at the Shibaura Institute of Technology have developed innovative machine learning models that predict soil behavior during seismic events. Utilizing geological data, these models create detailed three-dimensional maps that delineate stable and vulnerable soil areas, offering a broader and more comprehensive assessment than traditional soil testing methods, which are often limited in scope. Their recent study, published in Smart Cities on October 8, 2024, employs artificial neural networks (ANNs) and ensemble learning techniques to predict the depth of bearing soil layers, crucial for assessing stability and liquefaction risk. “This study establishes a high-precision prediction method for unknown points and areas, demonstrating the significant potential of machine learning in geotechnical engineering,” states Professor Inazumi. The researchers utilized data from 433 locations in Setagaya-ku, Tokyo, obtained from standard penetration tests and mini-ram sounding tests, to train their ANN. They accurately projected bearing layer depths for ten additional locations, validating their predictions against actual site measurements. Moreover, applying a bagging technique—a method that enhances model accuracy through repetitive training on diverse subsets—resulted in a remarkable 20% increase in prediction precision. The resultant contour map illustrating the depths of bearing layers within a kilometer radius of selected sites serves as an invaluable tool for civil engineers. This map aids in identifying suitable construction zones with stable soils while concurrently assisting disaster management specialists in recognizing areas at heightened risk of liquefaction, thereby enabling enhanced risk assessment and mitigation strategies. Ultimately, the researchers aspire to support smart city development through data-driven methodologies that inform urban planning and infrastructure projects. “This study provides a foundation for safer, more efficient, and cost-effective urban development,” concludes Professor Inazumi. Moreover, they intend to further refine their model’s accuracy by incorporating diverse ground conditions and developing specific models tailored for coastal versus non-coastal environments, considering the critical role of groundwater in liquefaction dynamics.

The phenomenon of soil liquefaction poses a significant risk during earthquakes, particularly in urban areas with loose, water-saturated soils. As cities grow, the implications of such geological hazards become more pronounced, requiring forward-thinking strategies from urban planners and disaster response teams. Notable historical events have underscored the destructive impact of liquefaction, prompting the need for advanced predictive models that can enhance resilience and facilitate better planning for infrastructure in earthquake-prone regions.

In conclusion, the development of advanced machine learning models to predict soil behavior during earthquakes represents a critical advancement in civil engineering and disaster management. By providing accurate assessments of bearing soil depths and enhancing the understanding of liquefaction risks, these models contribute significantly to the resilience of urban infrastructure in earthquake-prone regions. As further refinements are made to incorporate diverse geological conditions, the potential for safer and more robust urban development remains promising.

Original Source: techxplore.com