New AI model enhances durability of construction materials

Researchers at Seoul National University of Science and Technology have developed a hybrid machine learning model aimed at improving the safety and durability of structures through accurate strength predictions of construction materials.

Concrete-filled steel tube (CFST) columns, reinforced with carbon fibre-reinforced polymers (CFRP), are becoming popular for their ability to enhance structural durability and reduce maintenance. These composite materials promise to strengthen infrastructure while maintaining lightweight and corrosion-resistant features. However, the limited data available on these materials has posed significant challenges to accurately predicting their properties.

In response, a research team led by Associate Professor Jin-Kook Kim has introduced a hybrid machine learning model detailed in a paper published in Expert Systems with Applications. This model aims to improve the prediction of the ultimate axial strength of CFRP-strengthened CFST columns, a crucial parameter in construction.

"We employed a conditional tabular generative adversarial network, or 'CTGAN,' to generate new data with similar characteristics to real data," explains Dr. Kim. This synthetic database compensates for the insufficiency of existing real-world data.

The research team combined the Extra Trees (ET) technique with the Moth-Flame Optimization (MFO) algorithm to train and validate their model. Their findings suggest that this hybrid approach results in superior prediction accuracy, surpassing the best current empirical models. "Compared to existing empirical models in the literature, the predictive and reliable performances of the MFO-ET model are outstanding," highlights Dr. Kim.

The creation of this model is expected to provide engineers with enhanced tools for designing safer skyscrapers, high-rise buildings, and offshore structures. Additionally, it offers a means to retrofit older buildings and bridges with CFRP materials, providing resilience against corrosion and other natural degradations, which is essential in addressing climate change concerns.

To facilitate accessibility and application of the model, the researchers have also developed a web-based tool capable of making ultimate axial strength predictions of CFRP-strengthened CFST columns. This tool can be accessed freely from any device, eliminating the need for local software installation.

The hybrid machine learning model presents a valuable asset for improving the construction and assessment of CFRP-strengthened structures. Its performance in delivering reliable predictions under various conditions has been affirmed through rigorous testing and reliability analyses.

Seoul National University of Science and Technology, or SeoulTech, is a university focused on contributing to the nation and fostering creativity for humankind. It offers various programs in engineering, data science, business administration, and international studies, and has a strong research and industry-university cooperation foundation.