Scholarlink

<p>Global urbanization and transportation acceleration have heightened the collisions risk of vehicle on concrete filled steel tube (CFST) columns. The catastrophic fracture failure in structural column might ultimately cause the entire structure to collapse. Machine learning technologies were created in this work to forecast CFST lateral deflection and identify failure modes (bending deformation, crack and fracture failure). A database comprising 875 samples was constructed to create Gaussian process regression, Least-Squares Boosting, Random Forest, Support Vector Machine (SVM), Multilayer Perceptron (MLP), Long Short-term Memory, Naive Bayesian Classifier and k-Nearest Neighbors algorithms. MLP delivered the most accurate prediction of deflection with a correlation coefficient of 0.95, while SVM performed the best in classifying failure modes with an accuracy of 0.81. Shapley additive interpretation was used to analyze feature importance, it was discovered that geometric dimensions exerted greater effect on model output than material attributes. Finally, this research developed an intelligent application and provided a visual presentation of predicted results, which provides a promising, user-friendly and time-saving tool for investigating intricate impact response.</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

An Intelligent Application to Predict Deflection and Failure Modes of Concrete Filled Steel Tube Members Under Lateral Impact

  • 期刊论文

  • 发表日期：

    2026年7月5日

查看全文
  • Nan Xu,
  • Yanhui Liu

摘要

Global urbanization and transportation acceleration have heightened the collisions risk of vehicle on concrete filled steel tube (CFST) columns. The catastrophic fracture failure in structural column might ultimately cause the entire structure to collapse. Machine learning technologies were created in this work to forecast CFST lateral deflection and identify failure modes (bending deformation, crack and fracture failure). A database comprising 875 samples was constructed to create Gaussian process regression, Least-Squares Boosting, Random Forest, Support Vector Machine (SVM), Multilayer Perceptron (MLP), Long Short-term Memory, Naive Bayesian Classifier and k-Nearest Neighbors algorithms. MLP delivered the most accurate prediction of deflection with a correlation coefficient of 0.95, while SVM performed the best in classifying failure modes with an accuracy of 0.81. Shapley additive interpretation was used to analyze feature importance, it was discovered that geometric dimensions exerted greater effect on model output than material attributes. Finally, this research developed an intelligent application and provided a visual presentation of predicted results, which provides a promising, user-friendly and time-saving tool for investigating intricate impact response.