Scholarlink

<p>The bond strength between concrete and reinforcing bars in lap splices is crucial for the structural integrity of RC members. Inadequate bond strength can cause brittle failures, compromising safety and durability. However, predicting bond strength is challenging due to complex nonlinear relationships and limitations in current design codes. Developing a more accurate method to predict spliced bar tensile strength is essential for ensuring safe and cost-effective RC designs. This study proposes a novel approach to predict the tensile strength of unconfined lap-spliced steel bars using deep residual neural networks (DRNNs) and variance-based sensitivity analysis (VBSA). The combination of DRNNs with VBSA for predicting the tensile strength of unconfined spliced bars offers interpretable outputs that highlight critical input factors. DRNNs are particularly well-suited to model the nonlinear, complex interactions among the various parameters affecting bond strength. Moreover, the VBSA is employed in the post-processing phase to rank input variables based on their impact on output predictions and assess interactions between parameters. To train and validate the model, an extensive dataset of 149 for the unconfined lap splice test was compiled, incorporating data from the existing literature and experimental work conducted by the authors. The performance of the DRNNs model was then evaluated by comparison with established design codes and empirical models commonly used in practice. The DRNNs model outperforms traditional methods in predicting the tensile strength of lap-spliced bars, revealing trends that diverge from existing models, particularly those assuming uniform bond stress along the spliced bars. VBSA identified concrete compressive strength, bar diameter, and splice length as the most influential factors, while concrete cover had minimal impact. Finally, a parametric study is conducted based on the DRNNs model and the best traditional models to examine the influence of input parameters on the tensile strength of unconfined spliced bars. The approach proposed in this study improves prediction accuracy, reduces computational costs, and minimizes effort while also providing valuable insights for optimizing designs and updating current design codes.</p> Graphical abstract <p></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

Modeling tensile strength of unconfined lap-spliced steel bars using deep residual neural networks and variance-based sensitivity analysis

  • 期刊论文

  • 发表日期：

    2025年7月5日

查看全文
  • Ebrahim S. AL-Shami,
  • Mahmoud Owais,
  • Shehata E. Abdel Raheem,
  • Waleed Abo El-Wafa Mohamed,
  • Mohamed F. M. Fahmy

摘要

The bond strength between concrete and reinforcing bars in lap splices is crucial for the structural integrity of RC members. Inadequate bond strength can cause brittle failures, compromising safety and durability. However, predicting bond strength is challenging due to complex nonlinear relationships and limitations in current design codes. Developing a more accurate method to predict spliced bar tensile strength is essential for ensuring safe and cost-effective RC designs. This study proposes a novel approach to predict the tensile strength of unconfined lap-spliced steel bars using deep residual neural networks (DRNNs) and variance-based sensitivity analysis (VBSA). The combination of DRNNs with VBSA for predicting the tensile strength of unconfined spliced bars offers interpretable outputs that highlight critical input factors. DRNNs are particularly well-suited to model the nonlinear, complex interactions among the various parameters affecting bond strength. Moreover, the VBSA is employed in the post-processing phase to rank input variables based on their impact on output predictions and assess interactions between parameters. To train and validate the model, an extensive dataset of 149 for the unconfined lap splice test was compiled, incorporating data from the existing literature and experimental work conducted by the authors. The performance of the DRNNs model was then evaluated by comparison with established design codes and empirical models commonly used in practice. The DRNNs model outperforms traditional methods in predicting the tensile strength of lap-spliced bars, revealing trends that diverge from existing models, particularly those assuming uniform bond stress along the spliced bars. VBSA identified concrete compressive strength, bar diameter, and splice length as the most influential factors, while concrete cover had minimal impact. Finally, a parametric study is conducted based on the DRNNs model and the best traditional models to examine the influence of input parameters on the tensile strength of unconfined spliced bars. The approach proposed in this study improves prediction accuracy, reduces computational costs, and minimizes effort while also providing valuable insights for optimizing designs and updating current design codes.

Graphical abstract