Automating Structural Engineering Workflows with Large Language Model Agents

TL;DR

MASSE presents a first-of-its-kind Multi-Agent System for Structural Engineering that couples LLM-based agents with real-world workflows to automate end-to-end design, analysis, and verification. By organizing agents into Analyst, Engineer, and Management teams and embedding memory and structured tool use (e.g., FEM solvers, code documents), MASSE achieves large reductions in expert workload while maintaining reliability. The authors validate MASSE with a domain-specific racking system dataset and show performance competitive with larger models, highlighting a strong trade-off between speed, cost, and accuracy and demonstrating real-world deployment potential. The work advances the practical application of retrieval-augmented, tool-assisted reasoning in safety-critical engineering and suggests broader applicability to other verbalizable, tool-mediated domains.

Abstract

We introduce $\textbf{MASSE}$, the first Multi-Agent System for Structural Engineering, effectively integrating large language model (LLM)-based agents with real-world engineering workflows. Structural engineering is a fundamental yet traditionally stagnant domain, with core workflows remaining largely unchanged for decades despite its substantial economic impact and global market size. Recent advancements in LLMs have significantly enhanced their ability to perform complex reasoning, long-horizon planning, and precise tool utilization -- capabilities well aligned with structural engineering tasks such as interpreting design codes, executing load calculations, and verifying structural capacities. We present a proof-of-concept showing that most real-world structural engineering workflows can be fully automated through a training-free LLM-based multi-agent system. MASSE enables immediate deployment in professional environments, and our comprehensive validation on real-world case studies demonstrates that it can reduce expert workload from approximately two hours to mere minutes, while enhancing both reliability and accuracy in practical engineering scenarios.

Figures (13)

• Figure 1: Analogy of future human–AI collaborations. Traditional practice relies on long apprenticeships, with senior engineers transferring expertise to junior engineers through mentorship and problem solving. In contrast, LLM-based multi-agent systems instantiate scalable junior engineer agents that inherit workflows, perform specialized tasks, and coordinate under senior engineers' oversight. As these systems evolve toward self-planning and adaptive learning, fully agentic hierarchies—with senior engineer agent directing junior engineer agents—could transform engineering into a continuously improving, highly efficient practice.
• Figure 2: Failure Modes of Single-Agent. We evaluate the same structural engineering problem over 10 trials, comparing single-agent and multi-agent setups. The single-agent failed in all 10 trials, with error distributed as shown in the figure. In contrast, our multi-agent framework succeeded in every trial. See details in Appendix \ref{['appendix:singlevsmulti']}.
• Figure 3: MASSE Overall Framework. The Analyst Team combines four layers of tools (LLMs, FEM solvers, engineering documents, and fundamentals) to enable multi-agent collaboration. The Engineer Team performs limit state verification (Response< Capacity), while the Project Manager coordinates workflows and the Structural Memory stores analysis data. The Safety Manager conducts the final adequacy check. Three specialized teams are shown: Analyst (data extraction and analysis), Engineer (design and verification), and Management (coordination and decision-making).
• Figure 4: MASSE Analyst Team with Tool Integration
• Figure 5: MASSE Engineer Team: Analyze, Design and Verify