|
Title |
A Study on the Application of Large Language Models (LLMs) for Automated Structural Drawing Generation Based on Structured Prompts
|
|
Authors |
박경규(Kyung-Kyu Park) ; 최원준(Won-Jun Choi) ; 이상현(Sang-Hyun Lee) ; 허석재(Seok-Jae Heo) |
|
DOI |
https://doi.org/10.11112/jksmi.2025.29.4.60 |
|
Keywords |
구조도면 자동화; 대규모 언어 모델; 구조화 프롬프트; 프롬프트 엔지니어링; 건축구조설계 Structural drawing automation; Large language model (LLM); Structured prompt; Prompt engineering; Architectural structural design |
|
Abstract |
This study proposes a novel methodology for automatically generating structural drawings by leveraging large language models (LLMs) in architectural structural design. To apply LLMs to the highly repetitive process of structural drafting, a Structured Prompt (S-Prompt) technique was developed. The S-Prompt explicitly incorporates regulations from the Korean Building Code (KBC) into the prompt, hierarchically organizes parameters, and utilizes numerical constraints to minimize hallucinations by the LLM. To validate the effectiveness of the proposed method, experiments were conducted using two LLMs?OpenAI’s GPT-4o and Anthropic's Claude 3.7 Sonnet?for generating reinforced concrete structural member drawings. The results showed that the Claude 3.7 Sonnet model achieved an average drawing accuracy of 92.8%, significantly outperforming GPT-4o’s 49.0%. Furthermore, the structured prompt technique notably reduced interference errors between drawing elements and the omission of critical information. This study demonstrates the feasibility of LLM-based automation in structural drawing generation and highlights the high practicality and scalability of the proposed S-Prompt technique for future applications in automated architectural design.
|