Scholarlink

<p>As critical components in bearing and transferring superstructure loads, bridge pile foundations play a decisive role in ensuring structural safety. However, current single-pile design practices heavily rely on iterative normative calculations and engineering expertise, resulting in inefficiencies and labor-intensive work. To overcome these limitations, this study proposes an intelligent single-pile design method based on Graph Neural Networks (GNN), termed PileGNN, which enables end-to-end automated generation of pile design drawings from geological maps. This approach introduces two topological modeling strategies: edge-driven graph representation and node-driven graph representation. Through deep learning on graph data, PileGNN predicts the penetration ratio of each stratum as an intermediate variable, thereby transforming the pile-sizing problem into a linear combination. To validate this method, real-world engineering case studies were conducted. Results show that 95% of the designed piles meet the safety margin requirements specified by design codes, with an average deviation of less than 4% compared to engineer-designed schemes. Moreover, the entire process can be completed in under 2&#xa0;s—just 0.1% of the time required by manual design. The model outputs can be directly converted into CAD drawings without manual drafting. Overall, PileGNN demonstrates high accuracy, efficiency, and reliability, offering a theoretical basis for AI-driven automation across the whole bridge system in the future.</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

Automated design of bridge single-pile based on graph neural networks

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Buyu Jia,
  • Zhaofan Liang,
  • Wentao Ou,
  • Yujun Lin,
  • Xiaolin Yu

摘要

As critical components in bearing and transferring superstructure loads, bridge pile foundations play a decisive role in ensuring structural safety. However, current single-pile design practices heavily rely on iterative normative calculations and engineering expertise, resulting in inefficiencies and labor-intensive work. To overcome these limitations, this study proposes an intelligent single-pile design method based on Graph Neural Networks (GNN), termed PileGNN, which enables end-to-end automated generation of pile design drawings from geological maps. This approach introduces two topological modeling strategies: edge-driven graph representation and node-driven graph representation. Through deep learning on graph data, PileGNN predicts the penetration ratio of each stratum as an intermediate variable, thereby transforming the pile-sizing problem into a linear combination. To validate this method, real-world engineering case studies were conducted. Results show that 95% of the designed piles meet the safety margin requirements specified by design codes, with an average deviation of less than 4% compared to engineer-designed schemes. Moreover, the entire process can be completed in under 2 s—just 0.1% of the time required by manual design. The model outputs can be directly converted into CAD drawings without manual drafting. Overall, PileGNN demonstrates high accuracy, efficiency, and reliability, offering a theoretical basis for AI-driven automation across the whole bridge system in the future.