Large language model driven BIM collaborative automatic trestle-bridge modeling technology

Abstract Building Information Modeling (BIM) has been increasingly adopted in bridge engineering, yet the modeling of trestle bridges still involves repetitive manual parameter entry, component-family selection, coordinate calculation, API scripting, clash checking, and iterative correction in conventional BIM software. To address these interaction-intensive and weakly automated workflows, this paper proposes a large language model (LLM)-driven multi-agent framework for automatic trestle-bridge modeling. The framework is organized into a platform-independent reasoning layer and a Revit-specific execution layer. In the reasoning layer, the Requirement Analyst agent parses natural-language commands, validates modular design constraints, and completes missing parameters, while the Architect agent interprets external engineering parameter files, constructs geometric baselines, and generates a structured component layout scheme. In the execution layer, the Programmer agent maps the layout scheme into Revit API function calls, and the Revit plug-in executes the generated function list within the BIM environment. The proposed system can transform heterogeneous design inputs into traceable, constraint-aware, and executable modeling instructions, and the execution layer supports closed-loop diagnosis and local correction of modeling failures, completing an end-to-end process from design intent to 3D BIM model. Experiments on trestle-bridge modeling scenarios show that the proposed framework improves modeling robustness, reduces manual intervention, and provides a feasible path for automated bridge-engineering design.