Semantic property enrichment for circularity assessments using BIM and labeled property graphs

Integrating circularity assessments into early-stage building design supports informed decisions that advance a circular economy by evaluating, among other aspects, how easily building components can be separated and reused without damage. This study investigates how building information models (BIM) can be semantically enriched to automatically calculate disassembly potential and the building circularity indicator (BCI), two types of circularity assessment. The study presents a method combining semantic textual similarity (STS), matching of IFC elements to archetypal elements using fine-tuned large language models (LLM), with labeled property graphs (LPG). The IFC models are enriched with disassembly knowledge of the matched archetypes, encoding connection types, accessibility, and assembly logic. Across three architectural-model case studies and two LLMs, the fully automated pipeline yielded BCI values within ¡1.5% of manual benchmarks, but the low archetype-matching accuracy necessitated manual verification. Four competency questions validate the circularity graph, and manual recomputation of BCI confirms the calculation consistency. This approach contributes to reducing manual modeling effort and supports architects and engineers in evaluating design for disassembly principles during the early design phase. The work lays a foundation for future design decision-making, enabling circular building designs by combining AI- and graph-based reasoning systems. • Integration of BIM and archetype data into a fused circularity graph. • Archetype data enriches vague BIM with detailed circularity semantics. • Semantic matching and enrichment of BIM using textual similarity and LLM. • Alignment of different connection type classifications enables several data sources & metrics. • Labeled property graphs enable the dynamic assessment and visualization of BIM and circularity metrics.