LLM-Based IFC Mapping and Automatic IDS Generation for Supporting Automated Code Compliance

Abstract

Building regulations are typically expressed in natural language, which makes their interpretation heavily dependent on the manual work of domain experts. Practitioners must read regulatory clauses, identify the relevant requirements, and convert them into rule sets for BIM-based compliance checking. This process is time-consuming, error-prone, and difficult to scale across different regulatory domains. Although recent research has begun to automate parts of this workflow, the subsequent steps of integrating regulatory information into BIM processes and generating machine-readable rules still rely largely on manual effort. This paper introduces a two-stage approach that automates these remaining tasks using large language models (LLMs) through the Gemini API. The method operates on regulatory clauses that have already been semantically tagged with objects, properties, qualities, and values, providing a structured representation of the text from which the two stages proceed. In the first stage, the LLM performs schema-aware reasoning over a knowledge base of Industry Foundation Classes (IFC) 4.3 definitions to map each tagged element to its corresponding IFC class, property set, or attribute. In the second stage, these mappings are used to automatically generate Information Delivery Specifications (IDS) that encode the relevant attribute-based regulatory requirements. The workflow is validated using clauses from the International Building Code (IBC), with the generated IDS executed in Solibri to assess correctness and practical applicability. The results demonstrate that LLM-driven semantic mapping, combined with automated IDS generation, significantly enhances the consistency and scalability of digital compliance processes, providing a practical approach to integrating regulatory knowledge into openBIM environments.