Summary
Turning scanned blueprints of reinforced-concrete buildings into detailed computer models ready for safety checks and earthquake upgrades is usually done by hand and can be error-prone. The authors introduce BlueprintAgent, a smart tool that reads these blueprints and checks its own work against engineering rules while focusing on small problem areas for fixes. This approach helps it produce accurate models much more reliably than other methods that guess or apply fixed steps. Their tests on 300 real blueprints show BlueprintAgent almost perfectly identifies structural beams and columns needed for further analysis. This method shows that using rules to guide focused re-examination works better than just filtering or guessing at the end.
reinforced concreteblueprintssimulation-ready modelsfinite element methodmultimodal large language modeloptical character recognitioncomputer visionengineering constraintsbeam and column detectionstructural safety assessment
Authors
Zhouyuan Xu, Chen Yang, Linhao Wang, Jiansheng Fan, Chen Wang
Abstract
Converting in-service reinforced-concrete (RC) building blueprints into simulation-ready models---structured frame representations that support deterministic FEM export and qualified-engineer review---underpins safety assessment and seismic retrofit, but the process remains manual. Direct prompting of a multimodal large language model (MLLM) over a scanned sheet is unreliable: outputs often violate engineering constraints on beam--column support, span count, or 3D continuity. We present BlueprintAgent (BPA), a constraint-triggered multimodal agent for simulation-ready frame extraction from scanned blueprints. BPA treats the MLLM as the primary reader and decision maker, with OCR and computer vision supplying localized evidence. Its central mechanism realizes engineering constraints as callable validators whose entity-level conflict reports trigger targeted MLLM revisits over the local region---an inference-time control distinct from fixed pipelines and free-form self-reflection. We evaluate BPA on 300 real scanned blueprint sheets from 20 anonymized RC frame projects, against five baselines and six ablations. BPA reaches a macro-averaged Beam F1 of 0.994, against 0.301 for single-MLLM zero-shot and 0.820 for a fixed pipeline; removing MLLM-led axis adjudication collapses Beam and Column F1 on complex multi-sheet projects. For dense technical drawings, engineering constraints are best deployed as triggers for entity-level targeted revisits rather than as post-hoc output filters.