A minimal standards-only carrier profile for semantic PMI in OpenUSD: design, open implementation, and validation on the NIST MBE test models
Industrial digital-twin platforms increasingly ingest engineering CAD into OpenUSD. Semantic product and manufacturing information (PMI) — tolerances, datums, and their links to geometry — does not survive this ingestion: in the measured outputs of three production CAD-to-USD pipelines, no machine-readable carrier remains. The loss is fixed at the specification level: OpenUSD Core Specification 1.0 defines no carrier for semantic PMI, boundary-representation face identity, or annotation-to-face association. We propose a minimal carrier profile that gives these semantics a standards-only representation in USD, built from per-face subsets, typed attributes, and native relationships. We implement it in an open STEP-to-USD writer that serves as an executable reference definition of the profile, together with a fingerprint algorithm aligning STEP face identifiers with mesh faces. On the NIST MBE PMI test models (16 single-part B-rep models, 838 top-level annotations), the empirical results are: typed parsing covers 731/736 annotations (99.3%); face alignment resolves 3824/3843 faces (99.5%); associations resolve for 688/690 face-anchored annotations (99.7%), with target-level completeness of 4836/4864 (99.4%); and 11 annotations lack machine-readable geometry links in the NIST sources themselves. Survival (838/838) and exact typed round-trip hold by construction and are audit-confirmed. Carrying the semantics changes no geometric data; area-uniform mean deviations of all four outputs lie at or below the 0.06 mm level. Downstream, geometry-only baselines on production outputs recover hole geometry only partially or with an unflaggable error tail, and cannot link any feature to its tolerance; profile outputs are measurable by construction, with residuals at the numerical-precision floor (median 1.8e-06 mm).
Authors
- Jiawei Jing (ORCID: https://orcid.org/0009-0000-8365-367X)
Institutions
- Ai Corporation (United Kingdom) (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-08-28
- DOI
- https://doi.org/10.5281/zenodo.21299879
- Primary Topic
- Digital Transformation in Industry
- Type
- preprint