AI-Based Generative Structural Design for Reuse: Integrating Material Banks and LCA

The transition of the construction industry towards a circular economy demands Structural Design for Reuse (StDfR) from end-of-life assets. However, designing with finite stock constraints imposes a complex combinatorial problem. Traditional approaches aiming for single optimal solutions limit diversity, presenting high computational costs and poor generalization. To overcome these limitations, this study proposes a generative StDfR framework integrating Life Cycle Assessment (LCA) and Artificial Intelligence (AI), employing Reinforcement Learning and Graph Neural Networks parameterized by material banks. The model generated hundreds of diverse and viable structural designs in approximately five minutes, covering three truss topologies and various scenarios ranging from entirely new steel to full reuse. Notably, LCA revealed that full reuse does not guarantee the lowest environmental impact due to oversizing with reused elements, the transport of this additional weight, distances from deconstruction sites, and the reconditioning process, highlighting the efficiency of hybrid designs. By automating the information workflow between structural and sustainability teams, this AI approach provides agile StDfR support. Ultimately, the framework generates the final Building Information Modeling (BIM) model in the open Industry Foundation Classes (IFC), consolidating geometric, structural, and logistical project information. This tool empowers decision-making in conceptual phases, promoting the circularity of end-of-life assets.