Fine-tuning a domain-specific language model for truss structural analysis

This research investigates the feasibility of fine-tuning a domain-specific vison-language and large-language for truss structural analysis. General-purpose AI models often struggle with engineering-specific problems due to insufficient domain knowledge. To address this, we propose a hybrid workflow for truss analysis via the stiffness method as a case study. The project leverages a curated dataset of 27 truss templates and expanded through geometric augmentation, load randomization, and support variations. Llama 3.2 Vision Instruct was fine-tuned using the parameter-efficient fine-tuning to produce truss description from images, and T5-large was fine-tuned to convert these text description into JSON format for analysis using the stiffness method. Model performance was evaluated against the ground truth for node coordinates, elements, loads, and support conditions. This research demonstrates the potential of fine-tuned domain-specific language models to automate engineering analysis and design workflows, offering engineers and students a practical tool for rapid and accurate structural analysis.

Grantor dc:publisher

University of Tennessee at Chattanooga

Year dc:date.available

2027

Author dc:creator

• Dey, Rajon

Contributors dc:contributor

• Wu, Weidong
• Owino, Joseph; Liang, Yu; Fomunung, Ignatius; Onyango, Mbakisya A.
• College of Engineering and Computer Science

• Machine learning
• Natural language processing (Computer science)
• Structural analysis (Engineering)

Licence

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language

English, eng

Repository record dc:identifier

https://scholar.utc.edu/theses/1070

OAI identifier oai:identifier

oai:scholar.utc.edu:theses-2266