Automated Compliance Checking for BIM Models via a Large Language Model–Driven BIM-to-Text Paradigm

Abstract

Abstract Automated compliance checking (ACC) for building information modeling (BIM) is a vital task in the architecture, engineering, and construction (AEC) industry. However, significant semantic gaps among Industry Foundation Classes (IFC) schemas, regulatory texts, and logical rules necessitate extensive manual intervention for domain-specific knowledge modeling and rule interpretation. Fully automated compliance checking for BIM models remains a challenging task. To address these issues, this study proposes an innovative BIM-to-text paradigm, which takes full advantage of the context understanding and generation capabilities of emerging large language models (LLMs). Unlike the existing mainstream methods, the original intention of our proposed method is to convert BIM semantic information into natural language descriptions, rather than converting the specification text into logical rules. First, regulatory entities and entity types are extracted from regulatory texts using LLMs with a self-correction mechanism. Second, we have designed a novel BIM parsing strategy and data storage schemes, and the LLM is used for semantic alignment between IFC entities and regulatory entities to identify BIM semantic fragments. Next, the BIM semantic fragments are converted into BIM description texts via LLM guided by the BIM-to-text motivation. Finally, the regulatory texts and the generated BIM description texts are fed into the LLM to infer the results of compliance checking. On the building BIM ACC data set and the bridge BIM ACC data set, our method achieves F1-scores of 95.70 and 96.13% in the end-to-end setting, respectively. The experimental results also indicate that the proposed framework can avoid complex ontology modeling and rule interpretation processes in the traditional semantic parsing–based BIM ACC methods, and achieves compliance checking with a higher level of automation. This work contributes a novel LLM-driven BIM-to-text paradigm for ACC.

The authors' abstract, as published at the source. Journal of Computing in Civil Engineering, 2026 · DOI ↗