Multi-agent large language model framework for code-compliant automated design of reinforced concrete structures
Jinxin Chen, Yi Bao
Abstract
The current manual approach for designing reinforced concrete, guided by structural design codes, is inefficient and susceptible to human error. This paper presents a Large Language Model (LLM) framework to automate code-compliant design and achieve interpretability and verifiability. The framework decomposes complex tasks into subtasks handled by coordinated LLM agents with specialized expertise, enabling automatic structural design and human-robot interaction for exploring alternative solutions and explanations. This framework was tested using case studies on the design and evaluation of 30 beams and compared against commercial engineering software SAP2000, demonstrating how the agents collaborate and cross-check results while maintaining high accuracy (97 %), high efficiency (90 % time-saving), and transparency in structural analysis and design. An intuitive Graphical User Interface (GUI) that supports natural language queries was developed to facilitate practical use. By bridging the gap between intuitive communication and rigorous structural analysis, this framework provides a paradigm shift for automatic structural design.