Large Language Models for BIM-Based Generative Design of Structural Models

Building information modeling (BIM) has revolutionized the architectural, engineering, and construction (AEC) industry by reshaping how projects are designed and managed. BIM-based structural design workflows often involve repetitive tasks to modify designs, which are time-consuming and labor-intensive for designers. Large language models (LLMs), such as generative pretrained transformers (GPTs), offer a promising solution by automating these workflows. However, LLMs heavily depend on the clarity and structure of prompts for effective generation. Poorly designed prompts can lead to errors or incomplete results, potentially compromising the design process. This study addresses these challenges by proposing a new generative design approach that leverages prompt engineering for LLMs to enhance Dynamo capabilities in programmatically creating structural Revit models. Testing results demonstrated that the proposed approach enabled GPT to produce effective Python scripts, which were successfully implemented in Dynamo workflows. These workflows generated industrial shed structural models in Revit with varied specifications and high precision.