Graph neural networks for structural engineering : improving BIM workflows and IFC model exchange

Tools

Nabrotzky, Toni; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Graph neural networks for structural engineering : improving BIM workflows and IFC model exchange. In: EG-ICE 2025. University of Strathclyde Publishing, GBR, pp. 417-423. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093299)

[thumbnail of Nabrotzky-EG-ICE-2025-Graph-neural-networks-for-structural-engineering]
Preview
 Text. Filename: Nabrotzky-EG-ICE-2025-Graph-neural-networks-for-structural-engineering.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Model-based planning is a fundamental component of the BIM-supported design process. In structural engineering, architectural models must be transformed into analytical models, a task typically performed by rule-based converters. However, these converters often reach their limits, especially when handling geometries with complex details, necessitating manual adjustments. This paper explores the application of Graph Neural Networks (GNNs) to support automated model conversion. Structural models are represented as analytical graphs, where nodes correspond to building components with their attributes, and edges describe the load transfer between them. An algorithm first extracts the load-bearing elements from an IFC model and generates an unconnected graph. Subsequently, a GNN classifies the edges to predict the connections and reconstruct the load flow. Training and validation are conducted on a parametrically generated IFC dataset. The analysis demonstrates that GNNs achieve high accuracy in edge classification, although further optimization is necessary. The findings confirm the suitability of GNNs for model conversion and highlight future research directions to enhance interoperability between different geometric representations.

CORE (COnnecting REpositories)