Nerf2BIM

AI-Driven Detailing-on-Demand through Sustainable Point Cloud Surveys and Semantic 3D Understanding for Advanced Modeling of Existing Buildings

Recently, the optimization of existing buildings, especially in Germany, where they account for up to 80% of all projects, has increased sharply, with a strong focus on the energy-efficient refurbishment of post-war residential buildings. In the coming years, residential buildings constructed up to 1918 (15%) and between 1920 and 1940 (13%) offer particularly great potential for energy savings.

Significant challenges arise from unavailable or outdated plans and data, which make complex building surveys necessary. Unlike new construction, existing buildings evolve throughout their life cycle, turning their investigation into a kind of "reverse design and planning process." This requires measuring "visible" surfaces and points, drawing on supplementary documentation and partial findings, and recovering crucial domain knowledge that is often forgotten today.

Because existing buildings can never be replicated with perfect precision, the resulting models are inevitably imperfect. NERF2BIM addresses these challenges by leveraging recent AI advancements to introduce a detailing-on-demand strategy, enhancing established BIM methodology with variable information integration along three lines.

The first line enhances point cloud–based surveys by addressing data acquisition, georeferencing, and uncertainty quantification, with a focus on sustainable solutions; scanning performance and reliability are improved through strategies for evaluating and documenting uncertainty. The second develops semantic interpretation and reconstruction of indoor and outdoor acquisitions, integrating 2D and 3D data to enable a 3D-consistent semantic understanding of building components within raw scans and supporting the reconstruction of simple BIM components from geometric representations. The third applies detailing to these "simple" BIM components by drawing on a knowledge base of architectural and construction insights, using novel AI methods and a user feedback system that continuously refines the results.

Through the integration of uncertainty-aware data gathering, context-aware reconstruction, and domain expertise, NERF2BIM bridges the data gap for existing buildings and enables more efficient, knowledge-driven renovation processes.