Graph neural networks for node classification and attribute allocation in architectural BIM

Jabi, Wassim ORCID: https://orcid.org/0000-0002-2594-9568 and Li, Yang 2024. Graph neural networks for node classification and attribute allocation in architectural BIM. Presented at: eCAADe 2024 Conference, Nicosia, Cyprus, 09 - 13 September 2024. Published in: Kontovourkis, O., Phocas, M. C. and Wurzer, G. eds. Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe. , vol.1 Cyprus: eCAADe, pp. 675-684. 10.52842/conf.ecaade.2024.1.675

Preview

PDF - Accepted Post-Print Version Download (1MB) | Preview

Abstract

Building Information Modelling (BIM) marks a notable shift in architectural design, extending beyond simple digital reproductions by capturing the spatial, physical, and operational characteristics of structures. Unfortunately, these representations are often complex in nature and difficult to inspect, analyze, and understand which can lead to errors and omissions during model construction. This research aims to leverage graph machine learning systems, utilizing learned datasets, to detect and rectify these issues, improving model quality and minimizing costly mistakes. To illustrate the application of graph neural networks in this domain, this paper applied a graph-based geometric and topological editor coupled with a graph neural network to a real-world dataset of residential building complexes. The developed workflow operates by converting traditional architectural floor plans into graph-structured data, enabling precise node classification predictions. The paper details the overall workflow, data preparation and conversion, hyperparameter optimization and experimental results. Comparing the performance of various graph neural network models has validated the efficiency of the chosen prediction model in processing and analyzing architectural floor plans, achieving an overall accuracy rate of approximately 95%. The paper concludes with a discussion of the potential and limitations of graph-based machine learning methodologies within the architectural domain and an outline of future work plans.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Publication
Status:	Published
Schools:	Schools > Mathematics Schools > Architecture
Publisher:	eCAADe
Related URLs:	Organisation
Date of First Compliant Deposit:	9 July 2024
Date of Acceptance:	12 May 2024
Last Modified:	18 Feb 2025 15:09
URI:	https://orca.cardiff.ac.uk/id/eprint/170462

Actions (repository staff only)

Edit Item