The production of annotated technical drawing in the Architecture, Engineering, and Construction (AEC) industry requires much work, necessitating the adoption of tools that can limit the workload especially with the advent of artificial intelligence (AI). This thesis sought to revolutionize architectural documentation by using Graph Neural Net-works (GNNs) as a step toward automating this process. The overarching goal is to automatically generate annotations for Building Information Modeling (BIM) designs, sparing architects from labor-intensive tasks. This thesis proposes a method to represent BIM models as graphs and classify nodes using GNNs thus predicting interconnectivity of elements. For this purpose, a method has been developed in the BIM authoring software ArchiCAD by introducing a plugin, named ServCAD utilizing C++ API. The information gathered from BIM models is streamlined and is essential for the next steps. The extracted data are transformed into detailed graphs that capture architectural elements' semantic and spatial aspects. These graphs serve as the GNN model's foundational data set. The objective is to evaluate the efficacy and accuracy of GNN in automating the annotation process, spe-cifically in label type prediction based on node classification, thorough testing and eval-uation. In summary, the prediction results for different types of elements (walls, zones, doors) were provided via Graph Convolutional Network (GCN) and Graph Attention Network (GAT). Both models demonstrated exceptional performance in classifying certain label types. Notably, GAT model achieved accuracy rates of over 90% for labels '0', '2', and '3' (GCN was also above 80 %), with label '2' achieving a perfect accuracy of 100%. However, there were discrepancies in the accuracy rates for labels '1' and '4'. The GCN outperformed the GAT model in accurately predicting these labels. Later, the prediction results were examined and mapped on the example projects having different complex-ities to complete the automatic annotation process by using ServCAD.     «

The production of annotated technical drawing in the Architecture, Engineering, and Construction (AEC) industry requires much work, necessitating the adoption of tools that can limit the workload especially with the advent of artificial intelligence (AI). This thesis sought to revolutionize architectural documentation by using Graph Neural Net-works (GNNs) as a step toward automating this process. The overarching goal is to automatically generate annotations for Building Information Modeling (BI...     »