Text-to-Bridge: An Intelligent Agent for Finite Element Analysis of Bridge Structures Based on Natural Language

This article introduces the innovative Text-to-Bridge project, which combines large language models (LLMs) with the Abaqus finite element analysis software to realize an intelligent workflow for automatically generating bridge structure models and performing structural mechanics analysis via natural language descriptions. The project aims to address pain points in traditional bridge analysis workflows and promote engineering automation and intelligence.

• Original Author/Maintainer: Fei-hu111
• Source Platform: GitHub
• Original Title: Text-to-bridge
• Original Link: https://github.com/Fei-hu111/Text-to-bridge
• Source Publication/Update Time: 2026-05-28T12:44:54Z

Bridge engineering is a complex and critical subfield of civil engineering. Traditional bridge structure design and analysis workflows rely on professional engineers' experience and manual operations. The finite element analysis phase requires proficiency in professional software, precise definition of model parameters, and writing tedious scripts, which is time-consuming, labor-intensive, and has a high threshold. Human errors can easily lead to later mistakes. In recent years, large language models have shown outstanding capabilities in code generation and other areas, inspiring the idea of using natural language descriptions to let AI complete modeling and analysis.

The project builds an intelligent agent workflow, with its core architecture based on Python, integrating LLMs and the Abaqus API. The technical architecture is divided into four layers:

1. Natural Language Understanding Layer: Responsible for intent recognition, parameter extraction, ambiguity resolution, and specification mapping;
2. Knowledge Representation and Modeling Layer: Designs domain-specific intermediate representations and includes built-in parameterized templates for common bridge types;
3. Abaqus Integration Execution Layer: Converts intermediate representations into executable scripts, covering the complete analysis workflow and supporting graphical/headless modes;
4. Result Interpretation and Reporting Layer: Automatically extracts analysis results, which are interpreted by LLMs to generate multi-format reports.

Taking a newly built simply supported T-beam bridge as an example, the user inputs a natural language description (e.g., span of 20 meters, deck width of 12 meters, 5 T-beams, C40 concrete, highway class II load, etc.). The system sequentially completes understanding (extracting parameters and objectives), modeling (calculating dimensions, defining materials, setting boundaries and loads), analysis (generating scripts and submitting jobs), and reporting (extracting stress results and evaluating safety), with no manual operation of the Abaqus interface required throughout the process.

The project faces four major challenges: geometric complexity, load specification complexity, diversity of analysis types, and result verification and quality control. The corresponding solutions are:

• Geometric complexity: Parameterized modeling + template library + semi-automatic mode;
• Load specifications: Built-in mainstream specification database + keyword matching + custom working conditions;
• Analysis types: Explicit specification/intelligent inference + dedicated templates + multi-stage support;
• Result verification: Rationality check + model preview + historical case comparison + script retention.

The application value includes engineering education (learning tool), design optimization (rapid scheme comparison), and specification research (large-scale parameter analysis). Future directions: Expansion to other finite element software, introduction of multi-physics analysis, integration with computer vision, and integration of optimization algorithms. The project promotes the integration of AI with traditional engineering software, transforms work modes, and contributes to the intelligence of the industry.