SeoulTech research team uses machine learning to optimize structural design safety

Researchers at SeoulTech have developed a new machine learning model that can accurately predict the strength performance of carbon fiber reinforced steel columns (CFRP reinforced CFST columns). This breakthrough is expected to achieve higher structural safety and durability in the construction and infrastructure fields. In order to cope with the increasingly complex needs of modern buildings, engineers have gradually adopted advanced materials and technologies, such as composite structures that combine carbon fiber reinforced polymer (CFRP) with concrete-filled steel tube columns (CFST). This combination of materials not only improves load-bearing capacity, but also has lightweight and corrosion-resistant properties, which can significantly improve the performance and life of buildings while reducing maintenance requirements.

First, data scarcity and performance prediction challenges. Although CFRP reinforced CFST columns have broad prospects in modern buildings, the lack of relevant experimental data has always been a major problem in performance prediction. Although existing machine learning models can provide prediction results to a certain extent, their reliability is often limited by insufficient data sample size. In response to this bottleneck, a research team led by Associate Professor Jin Zhenguo of SeoulTech developed a new hybrid machine learning model that successfully broke through this limitation.

Second, the design and verification of innovative models. The study was published online in the journal Applied Expert Systems on November 13, 2024, and will be officially published in March 2025. The model designed by the team can effectively predict the ultimate axial strength of CFRP-reinforced CFST columns, which is an important parameter that determines the safety of structural design. To make up for the lack of data, the research team used generative AI and used conditional table generative adversarial networks (CTGAN) to generate synthetic data with real data characteristics. Dr. Jin said: "These synthetic data greatly expanded the database size and provided reliable support for model training and verification." Subsequently, the researchers developed a hybrid model based on the database that combines Extra Trees (ET) technology and moth flame optimization (MFO) algorithm.

In addition, performance testing and result analysis. After multiple rounds of rigorous testing, the prediction performance of the hybrid model was significantly better than that of the traditional empirical model. Dr. Jin pointed out: "Our MFO-ET model performed well in prediction accuracy and reliability, especially in a number of key indicators. It showed a low error rate." In addition, the study also conducted a reliability analysis to confirm that the prediction results of the model were consistent under various conditions.

At the same time, the practical application and prospects of the model. This innovative model provides engineers with a new tool to design safer and more efficient CFRP-reinforced CFST columns. Its application scenarios include the design of skyscrapers, high-rise buildings and offshore platforms. In addition, the model can help evaluate the reinforcement of old buildings and bridges. By strengthening structures with CFRP materials, buildings will be more resistant to natural threats such as corrosion and extreme weather, especially in the context of increasing environmental pressure caused by climate change.

Finally, developing convenient tools to promote widespread application. To promote the popularization of the model, the research team also launched a free web browser-based tool that users can access from any device without installing any software. This tool enables engineers to quickly predict the ultimate axial strength of CFRP-reinforced CFST columns, which facilitates practical engineering applications.

In summary, this research by SeoulTech provides important support for improving the design and evaluation of CFRP-reinforced CFST columns. Through reliable strength prediction, engineers can not only optimize the construction process, but also improve the safety and efficiency of new and old structures at a lower cost. This achievement is expected to be widely used in the future, bringing new momentum to the development of the construction industry.