2026 Volume 7 Issue 3 Pages 727-735
The aging of approximately 730,000 road bridges in Japan and a chronic shortage of engineers pose significant challenges for establishing a sustainable maintenance management system. While the application of Building Information Modeling/Construction Information Modeling (BIM/CIM) is expanding in new construction projects, its adoption in the maintenance sector remains limited. This limitation stems from existing Level of Development (LOD) definitions, which demand high precision suited for the design and construction phases, making them impractical for existing bridges. Furthermore, constructing 3D models from paper or PDF drawings requires excessive labor.
To address these issues, this study proposes a methodology to drastically reduce modeling effort using generative AI. First, by organizing the use cases of BIM/CIM for newly constructed and existing bridges, this study extracts the challenges and proposes a new detail level, "e-LOD". It identifies the limitations of current LODs, which require uniform detail across the entire model, and proposes a new hybrid concept called "e-LOD 250". In this approach, the overall bridge is represented by a lightweight, simplified model (e-LOD 200), while specific critical areas, such as bearing supports equipped with IoT sensors, are modeled with high detail (equivalent to LOD 300). This ensures both practical operability and adequate structural comprehension.
Furthermore, a case study using generative AI is conducted to output Python code for Blender based on natural language prompts containing basic design conditions extracted from bridge drawings. The results demonstrate the successful automated generation of a simplified e-LOD 200 model in approximately 15 minutes. It was also proven that shape modifications could be flexibly executed using natural language prompts.