Agentic Large Language Models for Automated Structural Analysis of 3D Frame Systems

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Structural Engineering · Depth: Expert, extended

Summary

An agentic Large Language Model (LLM) framework has been developed for automated structural analysis of 3D frame systems from natural language inputs. This framework addresses challenges in irregular geometric representation, topological consistency, and long-horizon reasoning. It represents 3D frames by projecting them onto a 2D plan with orthogonal gridlines and a matrix of number of stories (MNS) for vertical extrusion. A multi-agent pipeline then processes this input: a problem analysis agent parses input into JSON, a floor decomposition agent derives floor layouts, and node, girder, slab, and column agents assemble the 3D geometry. Support and load agents assign conditions, and code translation agents generate executable SAP2000 scripts. Evaluated on ten representative 3D frames, the framework achieved an average accuracy of 90% across repeated trials, with an average runtime of 175 seconds and a cost of USD 0.193 per run, significantly outperforming general-purpose LLMs like GPT-5.4 and Gemini-3.1 Pro.

Key takeaway

For AI Architects or Research Scientists developing automated engineering tools, this agentic LLM framework offers a robust approach for 3D structural analysis. You should adopt multi-agent decomposition and structured geometric representations to overcome long-horizon reasoning and topological consistency challenges. This method significantly outperforms general-purpose LLMs, providing high accuracy, efficiency, and cost-effectiveness for complex domain-specific tasks. Consider integrating vision language models for nonorthogonal geometries in future work.

Key insights

Agentic LLMs with structured decomposition and specialized agents can reliably automate complex 3D structural analysis tasks.

Principles

• Decompose complex tasks into modular subtasks.
• Use structured intermediate representations (JSON).
• Employ specialized agents for narrow reasoning scopes.

Method

The framework uses a multi-agent pipeline: problem analysis, floor decomposition, parallel geometry generation (nodes, girders, slabs, columns), support/load assignment, and two-stage SAP2000 script translation.

In practice

• Represent 3D frames via 2D plan + MNS.
• Use GPT-OSS 120B for spatial inference.
• Use Llama-3.3 70B Instruct Turbo for code translation.

Topics

• Agentic LLMs
• Structural Analysis
• 3D Frame Systems
• SAP2000
• Multi-Agent Architecture
• Geometric Representation

Best for: AI Scientist, Research Scientist, AI Architect

Related on AIssential

• A Lightweight Multi-Agent Framework for Automated Concrete Barrier Design — A multi-agent "generation-evaluation-optimization" framework automates concrete barrier design, achieving high accuracy with lightweight LLMs.
• Integrating Graphs, Large Language Models, and Agents: Reasoning and Retrieval — Integrating graphs with LLMs enhances reasoning, retrieval, and decision-making by combining structured knowledge with semantic understanding.
• Agentic AI Vision System: Object Segmentation with SAM 3 and Qwen — Agentic AI systems combine VLMs and segmentation models for iterative, self-correcting visual understanding and object segmentation.
• LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks — Agentic frameworks like LEAP enable LLMs to achieve state-of-the-art formal theorem proving by bridging informal reasoning with verifiable proofs.
• SlideAgent: Hierarchical Agentic Framework for Multi-Page Visual Document Understanding — Hierarchical agentic reasoning across global, page, and element levels significantly enhances multi-page visual document understanding.
• Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems — Latent communication in LLM multi-agent systems reduces cost and information loss by exchanging continuous representations directly.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.