Making the invisible ground visible through data science

Have you ever thought about the ground beneath the building you are in right now? When constructing buildings or structures, ground surveys are essential. However, it is practically impossible to excavate and examine every part of the ground in detail. In the field of civil engineering, accumulated knowledge and experience play a crucial role, but they also involve certain decisions that may not always be entirely rational. Professor Takayuki Shuku of the Faculty of Architecture and Urban Design is exploring rational decision-making in civil engineering by integrating geotechnical engineering with data science.

Professor Shuku’s research covers a wide range of topics. One example is reconstructing a three-dimensional model of the ground using sparse borehole data. Before constructing a building, a ground survey is conducted at the planned site. Borehole drilling is performed to collect soil samples, which are then tested in laboratories to determine their characteristics. Ideally, conducting surveys across all areas of a construction site would provide the most accurate data, but this approach is both time-consuming and costly. Therefore, engineers and geologists analyze a limited number of samples along with the site’s topographical data to make informed predictions about the ground conditions. Professor Shuku has successfully trained machine learning models to process sample and topographical data, generating predictions that closely align with human experts’ assessments. “Since we are dealing with underground conditions, we can never know the true answer. Human-generated assessments rely on trust, whereas computer-generated results can provide a quantifiable degree of certainty, allowing for more rational decision-making.”

Another area of Professor Shuku’s research involves analyzing ground movement using data obtained from various satellites. By collecting movement data from specific locations over time, he assesses whether the ground remains stable or is undergoing subsidence. While satellite-based observations allow for large-scale data collection, they also contain significant noise. To address this, he employs machine learning techniques to extract only the relevant data necessary for analysis.　”Satellites provide vast amounts of data, but there is simply too much for humans to analyze manually. That is why we use machine learning to filter and extract meaningful and appropriate information for observation.”

While analyzing data through computer programs may sound simple in theory, what makes it possible in the first place? Professor Shuku explains:　”Humans can only perceive the space directly in front of them. However, machines can freely bend and fold space, revealing aspects of the world that are invisible to us. I like to describe this as ‘projecting data into a space where it can be easily analyzed.'”

After completing graduate school, Professor Shuku joined a construction company. However, his passion for integrating geotechnical engineering with data science never faded. Even while working full-time, he devoted his weekends to research. “Since this is a field where research can be conducted with just a single computer, I spent my time programming and analyzing data. My research served as a way to recharge from work, but eventually, I felt it was too valuable to remain just a hobby. That realization led me to pursue a doctoral degree, bringing me to where I am today.”

From ground modeling and database construction to failure probability calculations and reliability analysis, Professor Shuku believes that geotechnical engineering presents many challenges that can be addressed through data analysis. Currently, due to the limited application of data science in the field, safety measures in ground engineering tend to be excessive. For instance, even if ten foundation piles are sufficient for structural stability, it is common practice to install forty for extra security. Moreover, research findings are not always immediately adopted in real-world design projects. “Alongside my research, I frequently conduct workshops on data science. My goal is to promote the use of data-driven decision-making in practical applications, making rational judgments more commonplace in the industry.”

Since joining Tokyo City University in 2024, Professor Shuku has reflected on conducting research in Tokyo: “Tokyo makes it easy to invite both domestic and international guests, facilitating connections and discussions. On the other hand, due to the high population density, the data we collect daily contains a lot of noise. Filtering out noise requires having an internal framework for evaluation, which in turn demands a solid foundation of knowledge. Additionally, I often feel that there are many inefficiencies in everyday life. If we could make decisions more rationally, I believe society as a whole could become much more efficient.”

By deciphering the invisible ground beneath us through data, Professor Shuku’s research is creating a new standard for ensuring structural safety. Rather than relying solely on intuition and experience, he advocates for rational decision-making, helping cities become stronger and more resilient. With a solid foundation, we can take bold steps toward the future.