3D reconstruction of building information models with weakly-supervised learning for carbon emission modelling in the built environment

Vincent J.L. Gan, Kexin Li, Mingkai Li and Leanna B.E. Halfian

Applied Energy, 2025, vol. 377, issue PD, No S0306261924020786

Abstract: Buildings and construction activities contribute around 38.0 % of global CO2 emissions, emphasising the importance of carbon emission modelling in assessing the built environment's carbon footprint. For accurate carbon emission modelling, detailed as-built information on building structures is crucial, which is often challenged by uncertainties due to design modifications and construction variations. Scan-to-BIM technology mitigates this problem by capturing precise as-built geometries through point clouds and transforming them into detailed 3D digital models. However, traditional 3D modelling methods often rely on manual intervention in creating as-built BIM models, which leads to low accuracy and efficiency. This paper presents an AI-enhanced approach that employs weakly-supervised learning for automated BIM reconstruction, aiming at accurate carbon performance evaluation in the built environment. By employing weakly-supervised semantic segmentation, this approach segments structural components from 3D point clouds and formulates the topological relationships of building objects, which enhances the automation of BIM reconstruction. This ensures a detailed representation of dimensions and materials, facilitating effective carbon emission modelling. The results reveal marked improvements in both semantic segmentation and BIM model accuracy. Specifically, the WSSIS network demonstrates an OA of 99.42 % and a mIoU of 98.27 % for semantic segmentation. Moreover, over 80.0 % of the point clouds meet the 5.0 mm tolerance requirements for BIM model accuracy. These BIM models are then used to assess the upfront carbon footprint of construction materials, as well as to model carbon emissions from usage to demolition. This method significantly enhances accuracy in evaluating and understanding the carbon impact within the built environment, and represents a noteworthy leap towards eco-friendly architectural and construction practices that aim to reduce carbon emissions.

Keywords: Carbon emissions modelling; Weakly-supervised learning; Semantic instance segmentation; 3D reconstruction; Building information model (search for similar items in EconPapers)
Date: 2025
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0306261924020786
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:appene:v:377:y:2025:i:pd:s0306261924020786

Ordering information: This journal article can be ordered from
http://www.elsevier.com/wps/find/journaldescription.cws_home/405891/bibliographic
http://www.elsevier. ... 405891/bibliographic

DOI: 10.1016/j.apenergy.2024.124695

Access Statistics for this article

Applied Energy is currently edited by J. Yan

More articles in Applied Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().