| Aydan Aghabayli, Ricardo de Matos Camarinha, Manuel Esteves Luís, José Luís Granja, Bruno Figueiredo Machine Learning Applied to BIM: BIM2GNN Framework |
|---|
| Abstract. Building Information Modelling (BIM) has increasingly gained momentum throughout the AEC industry, allowing nowadays access to large repositories of data. However, converting data into knowledge requires an iterative process of contextualization and interpretation. Thus, this paper addresses the challenge of efficiently leveraging BIM data through Machine Learning (ML) techniques. The paper is structured in two main parts: firstly, a comprehensive literature review on the use of ML techniques in the AEC industry is presented; secondly, a BIM2GNN framework was designed and tested to make fully transparent iterative use of the Building Information Model data in Machine Learning algorithms, by introducing Convolutional Neural Network and Graph Neural Network algorithms approaches. The authors concluded that BIM data is relevant to implementing learning techniques in construction, however, there are yet several hurdles to overcome at the industry level. These include access to data, data format and file types, data structure and interoperability. By shedding light on these obstacles, this paper not only advances theoretical understanding but also provides practical insights essential for overcoming barriers to effective implementation of ML techniques in the AEC domain. |
| Keywords: Machine Learning, GNNs, BIM, Architectural Design, Space Syntax |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2024.06 |