The renovation of Europe's ageing building stock is essential for sustainable development, and digitalization plays a key role in streamlining renovation efforts. Traditional methods, such as manual drafting and material classification, are often slow and inefficient. This thesis proposes an innovative approach using artificial intelligence (AI) to automate material detection in architectural drawings, improving the circularity assessment process. AI, particularly computer vision (CV), has seen rapid advancements and increasing use in the architecture, engineering, and construction (AEC) industry. The methodology starts with the annotation and preprocessing of diverse datasets, followed by comparing various convolutional neural network (CNN) architectures. The best-performing model is fine-tuned to enhance material detection accuracy. This refined CNN model is applied to building information modelling (BIM) data, demonstrating its effectiveness in both simple and complex geometries. The semantic information generated is critical for creating material passports, enabling systematic analysis of building materials. Results show that the fine-tuned MobileNetV2 CNN model achieves a Mean Intersection over Union (IoU) of 38.91%, though overfitting on certain materials is noted. Testing reveals minimal deviations in reinforced concrete wall detection in simpler cases, while complex geometries present larger discrepancies.     «

The renovation of Europe's ageing building stock is essential for sustainable development, and digitalization plays a key role in streamlining renovation efforts. Traditional methods, such as manual drafting and material classification, are often slow and inefficient. This thesis proposes an innovative approach using artificial intelligence (AI) to automate material detection in architectural drawings, improving the circularity assessment process. AI, particularly computer vision (CV), has seen...     »