Graph neural networks for classification and error detection in 2D architectural detail drawings

The assessment and classification of architectural sectional drawings is critical in the architecture, engineering, and construction (AEC) field, where the accurate representation of complex structures and the extraction of meaningful patterns are key challenges. This paper established a framework for standardizing different forms of architectural drawings into a consistent graph format, and evaluated different Graph Neural Networks (GNNs) architectures, pooling methods, node features, and masking techniques. This paper demonstrates that GNNs can be practically applied in the design and review process, particularly for categorizing details and detecting errors in architectural drawings. The potential for visual explanations of model decisions using Explainable AI (XAI) is also explored to enhance the reliability and user understanding of AI models in architecture. This paper highlights the potential of GNNs in architectural data analysis and outlines the challenges and future directions for broader application in the AEC field. •Demonstrate the application of Graph Neural Networks (GNNs) for classifying architectural detail drawings.•Establish a framework for standardizing various architectural drawings into a graph format.•Evaluate different GNN architectures, pooling methods, node features, and masking techniques.•Highlights the potential of explainable AI (XAI) in improving the reliability and understanding of AI models.•Suggest future directions for the broader application of GNNs in the architecture, engineering, and construction (AEC) field.