In the continuous pursuit of improved performance and efficiency of structural engineering components, fiber-reinforced polymer (FRP) composites have emerged as a promising alternative to traditional construction materials, but challenges such as the confinement effect of microcirculation on noncirculating sequences and inadequate study of elliptical double skin rows (DSTCS). Meanwhile, the application of finite element methods (FEM) and machine learning (ML) technologies in structural engineering demonstrates the great potential for accurate prediction of structural behavior and provides new approaches to address these challenges.
Therefore, researchers at Qujing Normal University in China, BHU Institute of Technology, National Institute of Technology, India, and Gable University in Sweden collaborated on a study entitled “Proposed numerical and machine learning models for hollow and solid elliptical columns of fiber-reinforced polymer concrete steel.”
This study employs a hybrid approach that integrates FEM simulations using ABAQUS software with ML technology to investigate oval cross sections of glass fiber reinforced polymer (GFRP) reinforced DSTC and double skin filled tubular columns (DSFTC). A comprehensive investigation of key parameters including aspect ratio, concrete strength, number of GFRP confinement layers, and steel pipe dimensions is carried out through comparative analysis and parametric studies on 54 modelled specimens. To ensure reliability of the findings, FEM simulation results are rigorously verified against experimental data, showing a high agreement with a maximum deviation of 11.71% at the ultimate load and 20.69% at the final strain, confirming the accuracy and reliability of the analysis. Seven ML algorithms are evaluated, namely Adaboost, LightGBM, CatBoost, Random Forest (RF), Extra Tree Regressor (ETR), XGB, and Deep Neural Network (DNN) to predict load-bearing capacity and ultimate confined strain. The results show that all algorithms achieve good reliability, and that ETR stands out as the best performance model, showing the highest coefficient of determination (maximum 0.9967) and lowest root square error. Additionally, Python web applications that incorporate RF models have been developed to facilitate practical prediction of limited ultimate loads. This study provides valuable insight into the structural behavior of elliptical FRP concrete steel composite rows, contributing to advances in FEM and ML applications in the design and evaluation of structural engineering.
Full text of the paper, “Proposed numerical and machine learning models for hollow and solid elliptical columns of fiber-reinforced polymer concrete steel,” created by Tang Qiong, Ishan Jha, Alireza Bahrami, Haytham F. Isleem, Rakesh Kumar, and Pijush Samui.
