A Large Language Model–based Approach for automatically Optimizing BIM

BIM (Building Information Modeling) is a way to represent, store and share real building digital information, and its construction method mainly relies on the design software tool, i.e. REVIT, while the generation of BIM is time-consuming and has strong professional barriers. In the case of low accuracy requirements, sometimes we want to quickly generate architectural prototypes also need to rely on the manual operation of professionals. How to quickly generate or optimize BIM based on partial building information is an important and challenging problem. Since BIM is usually described using IFC (Industrial Foundation Class) text files based on the EXPRESS language, the emergence of Large Language Models (LLMs), which specialize in text generation, provides a possibility to solve this problem. In this paper, we propose an automatic optimization method for BIM based on large language model, which can automatically supplement or optimize BIM from some building model information. Firstly, based on the analysis of IFC standard, we design an extraction algorithm for IFC instances; Secondly, we interpret and annotate the extracted instances, and guide the large model to generate the corresponding IFC text by means of dialogue; Finally, we validate the proposed method through the practice of automatic optimization of BIM for residential building.