Scholarlink

<p>In the design of prefabricated bridges, translating unstructured natural language requirements into standardized Building Information Modeling (BIM) models remains a key efficiency bottleneck. Existing BIM tools lack knowledge-driven alignment between natural language and engineering standards, as well as end-to-end automation from textual instructions to geometric modeling. To tackle these challenges, we propose AutoBIM, a framework that integrates Large Language Models (LLM), Retrieval-Augmented Generation (RAG), and structured prompts to automate the conversion of open-domain instructions into fully compliant BIM models. The core innovation lies in using RAG to ground LLM outputs within a verified knowledge base of standard components, eliminating hallucinations with 100% parameter accuracy and ensuring strict adherence to engineering specifications. Experimental results from a real prefabricated box-girder bridge case show that AutoBIM achieves 100% task success across diverse instructions, completing modeling in approximately 3.5&#xa0;min—significantly outperforming pure LLM and keyword-based methods. This work demonstrates the potential of LLM–RAG integration for semantics-driven, standardized engineering design and provides a feasible knowledge-driven technical pathway for intelligent BIM modeling in the architecture, engineering, and construction (AEC) industry.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Knowledge-driven automated prefabricated bridge modeling from natural language using LLM and RAG

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月25日

查看全文
  • Fei Huang,
  • Dapeng Mei,
  • Canwen Yang,
  • Chuanhai Su,
  • Runping Ma

摘要

In the design of prefabricated bridges, translating unstructured natural language requirements into standardized Building Information Modeling (BIM) models remains a key efficiency bottleneck. Existing BIM tools lack knowledge-driven alignment between natural language and engineering standards, as well as end-to-end automation from textual instructions to geometric modeling. To tackle these challenges, we propose AutoBIM, a framework that integrates Large Language Models (LLM), Retrieval-Augmented Generation (RAG), and structured prompts to automate the conversion of open-domain instructions into fully compliant BIM models. The core innovation lies in using RAG to ground LLM outputs within a verified knowledge base of standard components, eliminating hallucinations with 100% parameter accuracy and ensuring strict adherence to engineering specifications. Experimental results from a real prefabricated box-girder bridge case show that AutoBIM achieves 100% task success across diverse instructions, completing modeling in approximately 3.5 min—significantly outperforming pure LLM and keyword-based methods. This work demonstrates the potential of LLM–RAG integration for semantics-driven, standardized engineering design and provides a feasible knowledge-driven technical pathway for intelligent BIM modeling in the architecture, engineering, and construction (AEC) industry.