Machine Learning-Based Design Approach for Concrete-Filled Stainless Steel Tubular Columns

• Published in: Arabian journal for science and engineering (2011) Vol. 48; no. 10; pp. 14105 - 14118
• Main Authors: de Carvalho, Adriano Silva, Rossi, Alexandre, Morkhade, Samadhan G., Martins, Carlos Humberto
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2023; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Austenitic stainless steels; Axial compression; Building codes; Compressive strength; Concrete; Correlation coefficients; Engineering; Humanities and Social Sciences; Machine learning; multidisciplinary; Neural networks; Research Article-Civil Engineering; Science; Stainless steel; Steel columns; Steel tubes; Supervised learning; Finite element modelling; CFST; Axial compression; Stainless steel; Artificial neural networks; Random Forest
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-023-08090-3

Concrete-filled steel tubes have become popular due to their desirable properties, including compressive strength, plasticity, and ease of construction. This study aimed to comprehensively analyze the axial response of short columns made of concrete-filled austenitic stainless steel tubes. The different parameters of these columns were carefully evaluated and compared to principal international design codes. This analysis gained a deeper understanding of the structural behavior of concrete-filled stainless-steel tubes under axial compression. Based on the data obtained from the parametric study, two supervised machine learning models were used to model the compression behavior of these elements, the artificial neural networks model and the random forest model. It was observed that the results obtained through machine learning algorithms provide a significantly more accurate response than the models available in design codes. Additionally, it was observed through the study that the best results were achieved with the artificial neural networks model, with a correlation coefficient of 0.99. The trained machine learning models were implemented into software, allowing the prediction of the behavior of these structures in the range of the data presented in the study.