What is an industrial knowledge graph?
An industrial knowledge graph is a graph of a plant's data in which equipment, tags, documents, inspections and events are nodes and their relationships are labeled links. Often built on W3C standards such as RDF and fed from vocabularies such as ISO 15926 and CFIHOS, it lets one query follow a tag to its datasheet, inspections and damage mechanisms.
By Jun Zhong, Editor·Researched and drafted by Beacon's AI newsroom·Checked by Jun Zhong, Editor·Last checked ·Next review
Why does it matter?
A 2004 study for the US National Institute of Standards and Technology (NIST) estimated that poor interoperability between the software systems used to design, build and run US capital facilities, industrial plants included, cost $15.8 billion in 2002 1. About $9.1 billion of that fell in the operations and maintenance phase, and owners and operators carried about $10.6 billion, roughly two-thirds 1. The study linked the operating-phase cost to information that was hard to manage and reach, and owners traced it back to as-built data poorly kept and passed on during design and construction 1. A knowledge graph is one way of storing plant data so that each record keeps its connections to the others.
What is in an industrial knowledge graph?
A 2021 survey in ACM Computing Surveys describes a knowledge graph as data arranged as a graph to collect and share knowledge about the world: nodes for the things of interest, edges for the ways they relate 2. Three further elements usually come with it: a schema giving the graph its overall structure, identity saying which nodes refer to the same real thing, and context saying when or where a fact holds 2.
In the World Wide Web Consortium (W3C) Resource Description Framework (RDF) 1.1, a Recommendation of February 2014, every fact is a triple of subject, predicate and object, and a graph is a set of such triples 3. Subjects and predicates are named by internationalized resource identifiers (IRIs), so two systems using one identifier refer to one thing 3.
What makes a graph industrial is its subject matter. ISO 15926-2:2003 sets out a data model for technical information about process plants, oil and gas production facilities included, across planning, specifying, installing, operating and maintaining equipment 5.
What does a small one look like?
Illustrative numbers, not from any real asset or standard table
Illustrative data, not from any real asset or standard reference library. Dates sit in the object column for reading; a real graph would hang them on a separate installation event node.
| Subject | Predicate | Object |
|---|---|---|
| Tag P-101A | is a | Pump position in the produced water system (a functional location) |
| Pump SN-4471 | was installed at | Tag P-101A, May 2016 to September 2024 |
| Pump SN-5820 | is installed at | Tag P-101A, from September 2024 |
| Pump SN-5820 | is described by | Datasheet DS-P-101A, revision 3 |
| Datasheet DS-P-101A, revision 3 | gives casing material | Duplex stainless steel |
| Tag P-101A | handles | Produced water with sand |
| Inspection record IR-2025-031 | inspected | Pump SN-5820 |
| Inspection record IR-2025-031 | found | Casing minimum wall 11.2 mm, by ultrasonic testing, March 2025 |
| Erosion-corrosion | is a credible damage mechanism at | Tag P-101A |
Query: "Every inspection of the pump now at P-101A, with its material and damage mechanisms." It walks from the tag to the item installed since September 2024, then to its inspection record, its datasheet and the mechanisms linked to the tag. The records of SN-4471 stay with SN-4471.
The example keeps the tag, a place in the process, apart from the pump, a physical item installed there for a time. In its Key Programme 4 (KP4) on ageing offshore installations, the UK Health and Safety Executive (HSE) found that maintenance systems usually identified equipment by functional location, so a replacement pump took the old tag, traceability was lost and performance histories could be confused 10. HSE suggested a unique tag for each item of kit 10.
Which standards feed an industrial knowledge graph?
| Standard, publisher, date | What it covers | Role in a graph |
|---|---|---|
| RDF 1.1, W3C, 2014 | Triples of subject, predicate and object, named by IRIs 3 | The data model |
| OWL 2 (Second Edition), W3C, 2012 | An ontology language giving classes, properties and individuals a formal meaning 4 | The vocabulary and rules |
| ISO 15926-2:2003, ISO | A conceptual data model for process plant life-cycle data, used with reference data from other parts of the series 5 | Upper model for plant data; the POSC Caesar Association (PCA) published an OWL version in 2008 and reworked it into the Part 14 ontology 6 |
| Capital Facilities Information Handover Specification (CFIHOS), begun 2012; with the International Association of Oil & Gas Producers (IOGP) as JIP36 since 2020 | Information handover between operators, contractors and suppliers 7,8 | Shared classes and properties for tags and equipment at handover |
| OSDU Data Platform, The Open Group | A data model and application programming interface (API) standard for technical and operational energy data, with an open-source reference implementation; Standard Version 1.0, June 2026 9 | A shared data store the graph can link to |
How is it different from a relational database?
| Relational database | Knowledge graph | |
|---|---|---|
| Unit of data | A row in a table | An edge between two nodes |
| Schema | Designed first, changed as new kinds of data arrive | Can be deferred and grown with the data |
| Following connections | Joins between tables | Joins plus paths of any length |
| Meaning of terms | In column names and application code | Can be stated in an ontology and used to infer facts |
Source: Hogan et al., 2021 2.
A graph seldom replaces the systems it draws on. A 2021 review of knowledge graphs in manufacturing and production found that combining data from several sources was their main use, and that empirical research and industrial application were still largely missing 12.
Is a knowledge graph the same as an ontology?
No. The ontology is the vocabulary: OWL 2 gives classes, properties and individuals a formal meaning, and the ISO 15926 Part 14 ontology is one built for plant data 4,6. The knowledge graph is the body of facts written with that vocabulary, such as which pump sits at which tag; the 2021 survey treats ontologies as one way to state what a graph's terms mean 2.
Where integrity problems show up
| Integrity task | Link the graph holds | What goes wrong without it |
|---|---|---|
| Trending an item's condition | Tag to the item installed, with dates | A replacement pump under the old tag loses traceability 10 |
| Keeping inspection records | Each inspection to the equipment it covered | US rules require each record to name the equipment by serial number or other identifier 11 |
| Knowing what an item is made of | Item to materials and design codes | Process safety information must include both 11 |
| Management of change (MOC) | A change to the documents it affects | Changed process safety information must be updated 11 |
| Handover to operations | As-built data to tags and items | Owners traced operating costs to as-built data poorly kept upstream 1 |
Sources: HSE KP4 10; 29 CFR 1910.119(d)(3), (j)(4)(iv) and (l)(4) 11; NIST GCR 04-867, 2004 1.
Common questions
- Is a knowledge graph the same as an ontology?
- No. An ontology, for example one written in W3C OWL 2 such as the ISO 15926 Part 14 ontology, defines the classes, relations and rules. The knowledge graph holds the plant's actual facts described with that vocabulary.
- Does an industrial knowledge graph have to use RDF?
- No. The 2021 ACM Computing Surveys review of knowledge graphs names several graph data models, including directed edge-labeled graphs such as RDF and property graphs. RDF is the W3C standard model, which helps when data must be shared between organizations.
- Does a knowledge graph replace the maintenance or inspection system?
- Usually not. In a 2021 review of knowledge graphs in manufacturing and production, the main use found was knowledge fusion, combining data that stays in several source systems, and industrial evidence was still largely missing.
Beacon explains these concepts in its own words for education. It does not reproduce or replace the standards cited, and Winda is not affiliated with or endorsed by the organizations that publish them. Engineering decisions need the current edition of the standard and a qualified engineer.
Sources
- 1.Cost Analysis of Inadequate Interoperability in the U.S. Capital Facilities Industry, NIST GCR 04-867 (M.P. Gallaher, A.C. O'Connor, J.L. Dettbarn Jr., L.T. Gilday; RTI International and Logistics Management Institute), August 2004 · US National Institute of Standards and Technology (NIST)
- 2.Knowledge Graphs (A. Hogan, E. Blomqvist, M. Cochez and others), ACM Computing Surveys 54(4), article 71, 2021 · ACM Computing Surveys (preprint on arXiv) ·
- 3.RDF 1.1 Concepts and Abstract Syntax, W3C Recommendation · World Wide Web Consortium (W3C) ·
- 4.OWL 2 Web Ontology Language Document Overview (Second Edition), W3C Recommendation · World Wide Web Consortium (W3C) ·
- 5.ISO 15926-2:2003, Industrial automation systems and integration: Integration of life-cycle data for process plants including oil and gas production facilities, Part 2: Data model · Standards Council of Canada (SCC), standards database ·
- 6.PCA Reference Data and Services: ISO 15926-2 in OWL (background) · POSC Caesar Association (PCA)
- 7.New initiative will ease information handover · International Association of Oil & Gas Producers (IOGP) ·
- 8.Year of renewal for CFIHOS (P. Townson) · International Association of Oil & Gas Producers (IOGP) ·
- 9.The Open Group OSDU Forum · The Open Group
- 10.Key Programme 4 (KP4): Ageing and life extension programme · UK Health and Safety Executive (HSE), Energy Division
- 11.29 CFR 1910.119, Process safety management of highly hazardous chemicals, paragraphs (d)(3), (j)(4)(iv) and (l)(4) · US Occupational Safety and Health Administration (OSHA)
- 12.Knowledge Graphs in Manufacturing and Production: A Systematic Literature Review (G. Buchgeher, D. Gabauer, J. Martinez-Gil, L. Ehrlinger), IEEE Access 9, 55537 to 55554, 2021 · IEEE Access (preprint on arXiv)