Machine learning-based design for high-strength steel tubular section columns

Pengliang Yang, Cheng Chen, Yao Sun,
Machine learning-based design for high-strength steel tubular section columns,
Structures,
Volume 80,
2025,
109872,
ISSN 2352-0124,
https://doi.org/10.1016/j.istruc.2025.109872.
(https://www.sciencedirect.com/science/article/pii/S235201242501687X)
Abstract: Machine learning is increasingly transforming structural design by addressing the limitations of traditional design methodologies, especially as innovative high-strength materials challenge conventional approaches. By providing a data-driven framework, machine learning enables accurate predictions of material behavior and structural performance under complex conditions, improving design safety and efficiency in civil engineering. This study develops a machine learning-based design approach for high-strength steel tubular section columns, leveraging advanced algorithms to enhance predictive accuracy. A database consisting of 300 test data points on high-strength steel tubular section columns (with a wide range of cross-sectional profiles, geometric dimensions, manufacturing techniques and material grades) from existing literature was pre-processed for analysis, and nine machine learning models based on nine different algorithms were trained and evaluated. It was found that the best-performing model was developed based on the Light Gradient Boosting Machine algorithm. The column capacity predictions from the best-performing model were then compared against the design capacities predicted from the latest European and American codified design rules. The comparison results reveal that the machine learning model significantly outperforms traditional design codes, offering significantly improved design accuracy and consistency in compressive capacity predictions for high-strength steel tubular section columns.
Keywords: Design analysis; Tubular section columns; Machine learning; High-strength steel; Model training