دانشگاه شهید بهشتی صفه 1683-870X 36 2 2026 05 22 Integrated Framework for IFC Model Generation and Structural Layout from Architectural Sketch Plans Integrated Framework for IFC Model Generation and Structural Layout from Architectural Sketch Plans 83 108 106785 10.48308/soffeh.2026.242295.1468 FA Shima Afshar M.Sc. in Architecture, Department of Architecture and Urban Planning, Shahid Beheshti University, Tehran, Iran Mohammad Tahsildoost Associate Professor, Department of Architecture and Urban Planning, Shahid Beheshti University, Tehran, Iran 0000-0002-9016-9429 Journal Article 2025 11 02 Background and Objectives: Recent advances in Artificial Intelligence (AI) and Building Information Modelling (BIM) have enabled closer integration between data-driven automation and architectural design processes. Nevertheless, current AI-based BIM modelling approaches—particularly in the plan recognition stage—remain heavily reliant on large labelled datasets, making data preparation time-consuming and labour-intensive. This dependency limits rapid iteration in early-stage architectural design, where speed and responsiveness are essential. To address these challenges, this study proposes an integrated automation framework that leverages hybrid computational methods to accelerate the translation of sketch-based floor plans into structured BIM models. Materials and Methods: The proposed workflow comprises three main components. First, 2D raster floor plans are converted into vector data using a hybrid approach that combines deep learning-based opening detection (YOLOv11) with rule-based wall extraction. Second, the framework automatically recommends initial column layouts optimised according to spatial and functional design criteria. Third, it programmatically generates Industry Foundation Classes (IFC)-compliant 3D models. Geometric information extracted from sketch plans produces an IFC file encompassing walls, floors, doors, windows, and preliminary structural column layouts, closely reflecting the original design intent. Results and Conclusion: Experimental evaluation demonstrates robust performance across all stages. The object detection model achieved a mAP50:95 of 81%, with an inference time of 0.0041 seconds per image. Wall recognition reached 100% accuracy, while the column layout algorithm attained 95.83% alignment with executed plans. IFC generation was completed in under one minute, ensuring efficiency and practicality. These results underscore the framework’s capacity to substantially reduce manual effort, minimise reliance on large labelled datasets, and expedite the transition from conceptual sketches to BIM. By integrating hybrid computational techniques, the proposed approach offers a reliable, automated, and scalable solution for early-stage architectural design, bridging the gap between manual plan interpretation and BIM workflows, and providing a foundation for future research in intelligent architectural modelling. Background and Objectives: Recent advances in Artificial Intelligence (AI) and Building Information Modelling (BIM) have enabled closer integration between data-driven automation and architectural design processes. Nevertheless, current AI-based BIM modelling approaches—particularly in the plan recognition stage—remain heavily reliant on large labelled datasets, making data preparation time-consuming and labour-intensive. This dependency limits rapid iteration in early-stage architectural design, where speed and responsiveness are essential. To address these challenges, this study proposes an integrated automation framework that leverages hybrid computational methods to accelerate the translation of sketch-based floor plans into structured BIM models. Materials and Methods: The proposed workflow comprises three main components. First, 2D raster floor plans are converted into vector data using a hybrid approach that combines deep learning-based opening detection (YOLOv11) with rule-based wall extraction. Second, the framework automatically recommends initial column layouts optimised according to spatial and functional design criteria. Third, it programmatically generates Industry Foundation Classes (IFC)-compliant 3D models. Geometric information extracted from sketch plans produces an IFC file encompassing walls, floors, doors, windows, and preliminary structural column layouts, closely reflecting the original design intent. Results and Conclusion: Experimental evaluation demonstrates robust performance across all stages. The object detection model achieved a mAP50:95 of 81%, with an inference time of 0.0041 seconds per image. Wall recognition reached 100% accuracy, while the column layout algorithm attained 95.83% alignment with executed plans. IFC generation was completed in under one minute, ensuring efficiency and practicality. These results underscore the framework’s capacity to substantially reduce manual effort, minimise reliance on large labelled datasets, and expedite the transition from conceptual sketches to BIM. By integrating hybrid computational techniques, the proposed approach offers a reliable, automated, and scalable solution for early-stage architectural design, bridging the gap between manual plan interpretation and BIM workflows, and providing a foundation for future research in intelligent architectural modelling.

https://soffeh.sbu.ac.ir/article_106785_655a24492da711cd2e13bcb95d76f855.pdf