Applying Graph Neural Networks on BIM data

Automatic Classification of building elements using Machine Learning (ML) is a powerful technique for constructing, enriching, and validating Building Information Models (BIM). We propose improving existing classification methods by introducing contextual information into the model via Graph Neural Networks (GNNs).  In this approach, we build a graph data structure of a building where each element node has a basic geometric description and is connected to its immediate neighbors by edges. The graph structure allows us to efficiently enrich queries about specific building elements with contextual information. Good classification results will pave the way for introducing error correction and auto-complete functionality to BIM modelling software.