Ontology Selection for Manufacturing: 12 Standards Decision Matrix & Implementation Roadmap

Introduction

Ontology is not merely an academic concept; it is the engineering infrastructure that solves the "different terminology, different standards" problem in enterprises. At least 12 vertical domains already have mature standard ontologies, so companies do not need to build from scratch. This article delivers a selection matrix, a toolchain guide, a three-phase rollout plan, six pitfalls to avoid, and domestic case studies.

Why Manufacturing Needs Ontology

Manufacturing systems suffer from four core pain points:

Data silos : ERP uses material codes, MES uses part numbers, WMS uses SKUs, requiring manual mapping.

Integration cost explosion : N systems require N×(N-1)/2 point-to-point interfaces, with maintenance cost growing exponentially.

Master data chaos : Excel dictionaries are maintained manually and must be rebuilt whenever business changes.

No foundation for AI : Large language models do not understand business concepts, making intelligent Q&A unreliable.

Ontology solves these by providing "define once, reuse globally."

Standard Selection: 12 Manufacturing Ontology Standards

The article evaluates five key standards (representing the 12) across five dimensions: applicable scenario, maturity, open-source availability, and integration difficulty.

BFO 2.0 (Basic Formal Ontology) : Cross-industry top-level ontology, mandatory for all enterprises. ISO standard, open-source on GitHub, low integration difficulty (schema layer only).

ISO 10303 STEP : Product data for manufacturing, automotive, aerospace. ISO standard, OntoSTEP open-source, medium integration difficulty (requires PLM connection).

AAS (Asset Administration Shell) : Industry 4.0 / digital twin scenarios. IDTA standard, open-source on GitHub, medium integration difficulty (new asset models needed).

GS1 Web Vocabulary : Supply chain and retail. GS1 standard, open, low integration difficulty (easy GTIN alignment).

ISO 15926 : Oil & gas / process industries. ISO standard, POSC open-source, high integration difficulty (complex model).

Recommended starter combination for manufacturing: BFO (top-level) + STEP/AAS (product & equipment) + GS1 (supply chain). Start with one high-value scenario such as product master data unification rather than attempting a comprehensive ontology upfront.

Toolchain Selection

Four stages with recommended tools:

POC modeling : Protégé (free, Stanford open-source, best OWL editing experience).

Automated modeling : Owlready2 (Python, integrates seamlessly with data pipelines).

Team collaboration : PoolParty / TopBraid EDG (enterprise-grade semantic management platforms).

Runtime storage : Stardog / Neo4j (graph databases with reasoning to support queries).

Start path: Learn modeling with Protégé → write scripts with Owlready2 → migrate to enterprise platform when scaling.

Implementation Roadmap (Three Phases)

Phase 1: POC Validation (0-3 months)

Select one high-value scenario (product master data unification).

Build minimal ontology based on BFO + STEP.

Connect SAP and PLM systems.

Team: 1-2 semantic engineers.

Milestone: Schema review completed, cross-system query operational.

Phase 2: Business Pilot (3-9 months)

Extend to 2-3 business departments.

Introduce AAS asset model.

Integrate MES, WMS, SCM.

Establish ontology governance process.

Team: 3-5 people (including domain experts).

Milestone: Cross-department unified query live, mapping automation rate > 60%.

Phase 3: Scale-Up (9-18 months)

Roll out to all business units.

Build enterprise knowledge graph platform.

Connect customer systems (e.g., Weichai, Cummins).

Integrate AI large models for intelligent Q&A.

Team: 8-12 people.

Six Pitfalls to Avoid

Building a comprehensive ontology upfront → Start from a single scenario, iterate and expand.

Treating ontology as a database → Ontology defines only the schema; instances remain in business systems.

Excluding domain experts → Domain experts are co-creators, not just validators.

Ignoring legacy systems → Ontology is a mapping layer, not a replacement layer.

Pursuing full standard compliance too early → Reference standards but tailor and adapt, align gradually.

Lacking version management → Establish ontology change notification mechanism; otherwise downstream systems break.

Domestic Case Studies

Haier COSMOPlat : Platform enterprise using ontology to connect cross-industry ecosystems.

CATL (宁德时代) : Unified semantics around product data, achieving lighthouse factory status.

Midea MeiQing (美的美擎) : Multi-division group unifying product and equipment semantics.

CRRC (中车) : Long-cycle equipment manufacturing, starting from master data governance.

Conclusion

Ontology is not a silver bullet, but semantic integration without ontology will inevitably repeat wheel-reinventing. The principle: selection first, reuse open standards, scenario-driven. Begin with one scenario, show results in three months — don't wait.

Reference diagrams (decision matrix, toolchain, roadmap, pitfalls, case studies):