Civil Engineering Infrastructures Journal

Civil Engineering Infrastructures Journal

A Real-World Dataset for Beam and Column Dimension Prediction in Reinforced Concrete Structures Using Graph Neural Networks

Authors
1 Department of Civil Engineering, University of Qom, Qom, Iran.
2 Department of Civil Engineering, Isfahan University of Technology, Isfahan, Iran.
3 Department of Computer and Information Technology, University of Qom, Qom, Iran.
Abstract
Conventional structural design methods are often time-consuming, repetitive, and heavily dependent on designer expertise, requiring significant manual effort. Deep learning models offer a means to accelerate this process and reduce reliance on individual skills; however, their development is limited by the scarcity of large, well-structured datasets. To address this challenge, this paper introduces a standardized real-world dataset, AV_SStruct313, collected from 313 real reinforced concrete projects in Iran. Unlike previous studies that rely on synthetic datasets or limited real-world samples, AV_SStruct313 is derived from fully executed, code-compliant structural projects and explicitly captures practical design and construction constraints. The proposed dataset provides a realistic and reliable benchmark for developing accurate predictive models of beam and column dimensions. To demonstrate its effectiveness, five state-of-the-art graph neural networks—GCN, GIN, GraphSAGE, GAT, and Graph Transformer—are evaluated on AV_SStruct313. Among, the Graph Transformer achieved the highest accuracy, with an average prediction error of approximately 5 cm, which is consistent with practical design discretization levels in reinforced concrete design, and inference time under one second. These results indicate that graph-based learning can significantly reduce manual design effort and accelerate reinforced concrete design workflows, demonstrating a clear improvement over conventional methods and existing learning-based approaches.
Keywords
Subjects


Articles in Press, Accepted Manuscript
Available Online from 18 November 2026

  • Receive Date 05 December 2025
  • Revise Date 02 November 2026
  • Accept Date 18 November 2026