Scholarlink

<p>Concrete-filled double steel tubular (CFDST) and concrete-filled double-skin steel tube (CFDSST) columns have emerged as an efficient steel–concrete composite system, offering improved axial capacity than conventional columns. However, traditional methods, such as experimental and numerical studies, for assessing their capacity can be both complex and time-consuming, highlighting the need for accurate and efficient data-driven predictive tools. Hence, this paper aims at developing an ANN-based model for predicting the axial capacity of short circular CFDSST columns, incorporating both normal and high-strength materials. A comprehensive database comprising 296 CFDSST specimens, consisting of 138 validated FEA results and 158 experimental data collected from the literature, was used for model development. The ANN model was trained using eight input parameters representing the geometric and material properties of the column. The developed ANN model achieved an overall correlation coefficient of 0.9978, demonstrating excellent predictive capability. Furthermore, the proposed model exhibited improved prediction accuracy when compared with existing design provisions and previously published theoretical models, using various performance indices, achieving RMSE of 207.74, MAPE of 4.62%, <i>R</i><sup><i>2</i></sup> of 0.99, VAF of 99.38% and a20 index of 0.97, demonstrating improved prediction accuracy and reliability over existing design approaches. In addition, the Shapley additive interpretation (SHAP) technique is also adopted to examine the contribution of input design parameters for predicting axial capacity. To further strengthen the explainable AI framework, Partial Dependence Plots (PDPs) were also incorporated alongside SHAP to illustrate the nonlinear effect of the governing input variables on the predicted axial capacity. Lastly, an explicit ANN-based equation is proposed and based on these, a Graphical User Interface (GUI) is also developed to facilitate user-friendly approach for the prediction of the capacity of stub CFDSST columns. Overall, the proposed ANN framework, coupled with explainable AI and a user-friendly Excel-based design tool, proves an accurate, efficient and practical approach for predicting the axial capacity of the CFDSST columns, thereby supporting reliable design and optimization in engineering practice.</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

ANN-powered design equations and graphical user interface for capacity prediction of short concrete-filled double skin steel tubular columns

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月30日

查看全文
  • Vanira Pradhan,
  • Sajal Sarkar,
  • Partha Sarathi Nayek,
  • Shantharam Patil

摘要

Concrete-filled double steel tubular (CFDST) and concrete-filled double-skin steel tube (CFDSST) columns have emerged as an efficient steel–concrete composite system, offering improved axial capacity than conventional columns. However, traditional methods, such as experimental and numerical studies, for assessing their capacity can be both complex and time-consuming, highlighting the need for accurate and efficient data-driven predictive tools. Hence, this paper aims at developing an ANN-based model for predicting the axial capacity of short circular CFDSST columns, incorporating both normal and high-strength materials. A comprehensive database comprising 296 CFDSST specimens, consisting of 138 validated FEA results and 158 experimental data collected from the literature, was used for model development. The ANN model was trained using eight input parameters representing the geometric and material properties of the column. The developed ANN model achieved an overall correlation coefficient of 0.9978, demonstrating excellent predictive capability. Furthermore, the proposed model exhibited improved prediction accuracy when compared with existing design provisions and previously published theoretical models, using various performance indices, achieving RMSE of 207.74, MAPE of 4.62%, R2 of 0.99, VAF of 99.38% and a20 index of 0.97, demonstrating improved prediction accuracy and reliability over existing design approaches. In addition, the Shapley additive interpretation (SHAP) technique is also adopted to examine the contribution of input design parameters for predicting axial capacity. To further strengthen the explainable AI framework, Partial Dependence Plots (PDPs) were also incorporated alongside SHAP to illustrate the nonlinear effect of the governing input variables on the predicted axial capacity. Lastly, an explicit ANN-based equation is proposed and based on these, a Graphical User Interface (GUI) is also developed to facilitate user-friendly approach for the prediction of the capacity of stub CFDSST columns. Overall, the proposed ANN framework, coupled with explainable AI and a user-friendly Excel-based design tool, proves an accurate, efficient and practical approach for predicting the axial capacity of the CFDSST columns, thereby supporting reliable design and optimization in engineering practice.