An Agentic AI Framework for Sectional Analysis of Reinforced Concrete Columns when Subjected to Fire

Abstract

This paper presents an agentic AI framework that automates the multi-stage analysis process for evaluating the load-bearing capacity of reinforced concrete columns under fire conditions through natural language interaction. First, a large language model orchestrator, equipped with a formal ontology for reinforced concrete structures based on international design codes, interacts with the user. The ReAct technique and LangChain framework are then utilized to decompose user requests into sub-tasks for a series of specialized agents: a data ingestion agent; a thermal analysis agent integrated with ANSYS for heat transfer simulations under ISO 834 standard fire exposure; and a structural capacity and visualization agent, named CFire, which computes residual axial capacities and plots axial force- biaxial bending moment interaction diagrams based on temperature-dependent stress-strain relationship theory. Furthermore, the retrieval-augmented generation paradigm equips the agentic system with expert domain knowledge. The applicability of the proposed agentic system is demonstrated via different reinforced concrete columns, showing that it automates repetitive and tedious tasks in structural fire engineering, such as the manual translation of geometric, material, and temperature data into heat-transfer models and code-compliant load-bearing calculations, thereby significantly improving productivity across the entire analysis process.