Intelligent prediction methods for N-M interaction of CFST under eccentric compression

Hou, Chao1; Zhou, Xiao-Guang1; Shen, Luming2

2023-07-12

10.1007/s43452-023-00734-3

ARCHIVES OF CIVIL AND MECHANICAL ENGINEERING   影响因子和分区

1644-9665

2083-3318

Machine learning (ML), as a promising artificial intelligence method, gradually begins to be applied in predicting the behavior of structural members and demonstrates its superior capability in capturing the nonlinear data laws between various structural parameters and the targeted capacity. For concrete-filled steel tube (CFST) under eccentric compression, the nonlinear material confinement and the unsymmetrical loading make the application of ML methods both challenging and appealing. In this study, a comprehensive literature review is conducted to gather data from 92 reported test programs, and an experimental database consisting of 899 eccentrically loaded circular CFST samples is established. Through structural behaviors and correlation analysis of the data, input parameters for the ML models are rationally identified. Prediction models are developed using seven single ML algorithms, two ensemble ML algorithms and two reference regression methods, with their predictions compared and explained by the shapely additive explanation (SHAP) approach. Based on Monte Carlo simulation, the uncertainty quantification of predicted capacities is conducted considering four degrees of randomness. The results reveal that extreme gradient boosting (XGBoost) yields the best prediction performance, providing less than 5% prediction errors for more than 58% samples. Compared with existing capacity calculation methods for CFST under eccentric compression in design standards, the established ML models exhibit higher accuracies and wider applicable ranges.

Machine learning Concrete-filled steel tube (CFST) Eccentric compression Capacity prediction N-M interaction curve prediction

SCI

英语

第一 ; 通讯

Shenzhen Science and Technology Program[RCYX20210706092044076]

Engineering ; Materials Science

Engineering, Civil ; Engineering, Mechanical ; Materials Science, Multidisciplinary

WOS:001027103800001

SPRINGERNATURE

Web of Science

1.Southern Univ Sci & Technol, Dept Ocean Sci & Engn, Shenzhen 518055, Peoples R China
2.Univ Sydney, Sch Civil Engn, Sydney, NSW 2006, Australia

Hou, Chao,Zhou, Xiao-Guang,Shen, Luming. Intelligent prediction methods for N-M interaction of CFST under eccentric compression[J]. ARCHIVES OF CIVIL AND MECHANICAL ENGINEERING,2023,23(3).

Hou, Chao,Zhou, Xiao-Guang,&Shen, Luming.(2023).Intelligent prediction methods for N-M interaction of CFST under eccentric compression.ARCHIVES OF CIVIL AND MECHANICAL ENGINEERING,23(3).

Hou, Chao,et al."Intelligent prediction methods for N-M interaction of CFST under eccentric compression".ARCHIVES OF CIVIL AND MECHANICAL ENGINEERING 23.3(2023).