Scholarlink

<p>The structural performance of circular concrete-filled steel tubes (CFSTs) is strongly influenced by complex interactions between steel and concrete, particularly under eccentric loading. Traditional analytical approaches often struggle to predict load capacity when parameters such as reinforcement configuration, steel section shape, and eccentricity vary simultaneously. This study addresses this challenge by proposing a novel integration of finite element (FE) simulations and machine-learning (ML) models for accurate capacity prediction and parametric analysis. A database of 66 experimentally tested CFST specimens under concentric loading was compiled from the literature and used for FE model validation. Building on this, 200 additional FE models with an eccentricity of <i>e</i> = 40&#xa0;mm were developed for a comprehensive parametric study. The FE analysis confirmed that eccentricity significantly reduces column strength: the average load capacity of eccentrically loaded specimens was only 51.8% of their concentric counterparts, with a coefficient of variation of 10.9%. Further parametric investigations highlighted the effects of yield stress, H-, X-, box-, O-, and E-shaped steel sections, spiral reinforcement, and longitudinal reinforcement ratio on load behavior. The validated data set was subsequently employed to train seven ML algorithms, including support vector regression (SVR), Gaussian process regressor (GPR), random forest regressor (RF), gradient boosting regressor (GBR), eXtreme gradient boosting (XGBoost), multi-layer perceptron (MLP), and K-neighbors regressor (KNN). The best-performing model (RF) achieved <i>R</i><sup>2</sup> = 0.9855 with correspondingly low RMSE, MAE, and MAPE values, confirming the robustness of the proposed framework. This study's novel approach, which integrates finite element (FE) analysis and machine-learning (ML) techniques, offers a dual benefit: quantifying the negative effects of eccentricity and providing a dependable predictive tool for the design and optimization of CFST structures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Finite-Element–Machine-Learning Framework for Predicting Load Capacity and Eccentricity Effects in Circular Concrete-Filled Steel Tubes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月27日

查看全文
  • Yizhuo Li,
  • Haytham F. Isleem,
  • Anlin Shao,
  • Mohammad Khishe

摘要

The structural performance of circular concrete-filled steel tubes (CFSTs) is strongly influenced by complex interactions between steel and concrete, particularly under eccentric loading. Traditional analytical approaches often struggle to predict load capacity when parameters such as reinforcement configuration, steel section shape, and eccentricity vary simultaneously. This study addresses this challenge by proposing a novel integration of finite element (FE) simulations and machine-learning (ML) models for accurate capacity prediction and parametric analysis. A database of 66 experimentally tested CFST specimens under concentric loading was compiled from the literature and used for FE model validation. Building on this, 200 additional FE models with an eccentricity of e = 40 mm were developed for a comprehensive parametric study. The FE analysis confirmed that eccentricity significantly reduces column strength: the average load capacity of eccentrically loaded specimens was only 51.8% of their concentric counterparts, with a coefficient of variation of 10.9%. Further parametric investigations highlighted the effects of yield stress, H-, X-, box-, O-, and E-shaped steel sections, spiral reinforcement, and longitudinal reinforcement ratio on load behavior. The validated data set was subsequently employed to train seven ML algorithms, including support vector regression (SVR), Gaussian process regressor (GPR), random forest regressor (RF), gradient boosting regressor (GBR), eXtreme gradient boosting (XGBoost), multi-layer perceptron (MLP), and K-neighbors regressor (KNN). The best-performing model (RF) achieved R2 = 0.9855 with correspondingly low RMSE, MAE, and MAPE values, confirming the robustness of the proposed framework. This study's novel approach, which integrates finite element (FE) analysis and machine-learning (ML) techniques, offers a dual benefit: quantifying the negative effects of eccentricity and providing a dependable predictive tool for the design and optimization of CFST structures.