Why AI Needs a Semantic Digital Thread
AI on engineering data is only as good as its foundation. Semantics, interconnection, configuration context, and provenance are what make system-level AI trustworthy.
AI is moving into digital engineering fast. The goal is no longer a chatbot over one tool’s documents. It is AI that works at the system level, across requirements, MBSE, design, simulation, analysis, ALM, PLM, and manufacturing, end to end.
That is where most AI efforts stall. The data AI needs is spread across tools that were never designed to share it:
- Cross-domain relationships don’t exist, and they can’t be inferred after the fact.
- Every tool has its own data schema, and those schemas keep changing.
- Data is duplicated through re-entry and import/export, so nobody is sure which copy is right.
- No two tools manage configuration the same way.
Point AI at that and you get answers that sound right but can’t be trusted. High-confidence AI needs a foundation built for it: a semantic digital thread.
Defense programs already have a mandate pointing the same way. DoD Instruction 5000.97, issued in December 2023, directs new programs to use digital engineering and to maintain an authoritative source of truth for their models and data. A semantic digital thread is how that source of truth becomes something AI can reason over.
What makes a digital thread semantic
A digital thread connects every stage of a product’s lifecycle into one cohesive set of data across engineering teams, while building on the PLM and engineering tools you already have. A semantic digital thread goes a step further: it captures what the data means, not just where it is.
That meaning comes from an ontology: a formal model of the concepts in a domain and how they relate. In engineering, that means things like requirements, system elements, parts, analyses, and verifications, and relationships like satisfies, allocated to, and verified by. When every tool’s data is mapped to the ontology, a requirement is a requirement wherever it came from, and its links mean the same thing everywhere.
SBE Vision’s Model Based Engineering (MBE) ontology is built on the Basic Formal Ontology (BFO), an ISO-standard top-level ontology (ISO/IEC 21838-2). MBE inherits BFO’s foundation and extends it with engineering concepts refined over years of real customer programs. The platform can also work with multiple ontologies, so the model fits the program. In January 2024, the DoD and Intelligence Community data councils directed BFO, along with the Common Core Ontologies built on it, as the baseline standard for formal ontology work across DoD and the IC.
The four foundations of system-level AI
Ontology-based digital threads provide the foundation for system-level AI by enabling four things. Take one example throughout: a motor mass limit drops from 13 kg to 12 kg.
Semantics
A structured representation of otherwise unintelligible raw data, enabling accurate and relevant AI results.
To a model reading raw exports, “13” in one tool and “mass_max” in another are just strings. With semantics, the AI knows it is looking at a mass requirement on the motor, in kilograms, and what depends on it.
Interconnection
Cross-domain linkages necessary for AI to make system-wide, powerful, and well-grounded inferences.
The mass limit links to the rotor drive allocation in the system model, the rotor group in the BOM, and the tests that verify it. The AI follows the change to everything it touches instead of guessing.
Configuration context
Data organized by configuration, ensuring AI operates within the correct system context.
Programs run many baselines and variants at once. Configuration context tells the AI which revision applies to which build, so its answer reflects the system you are actually changing.
Provenance
Where the data came from, who it is shared with, and where it is being used, enabling trusted AI outputs.
Every answer traces to its sources: which tool, which revision, which owner. Engineers can check the AI’s work, and reviewers can defend the decision later.
Published research backs up the first foundation:
In one benchmark of enterprise questions, GPT-4 answered correctly 16% of the time when querying a database directly, and 54% of the time when the same data was represented as a knowledge graph defined by an ontology. Using the ontology to check and repair the AI’s queries raised accuracy above 70%.
The benchmark used insurance data, but the principle carries over: meaning has to be in the data before AI can use it.
What happens without them
| Without | What AI does |
|---|---|
| Semantics | Reads raw values and field names with no meaning, so answers are generic or wrong |
| Interconnection | Sees one tool at a time and misses the downstream impact of a change |
| Configuration context | Mixes revisions and variants, answering for the wrong version of the system |
| Provenance | Gives answers no one can trace or verify, so they can’t be used for real decisions |
There is one more catch. The relationships, context, and history a semantic thread captures happen during engineering: when a requirement is allocated, a part is released, a test is linked. They cannot be reconstructed later from exports.
If you want system-level AI you can trust, you need a semantic digital thread. And if you don’t build it during engineering, you will never have it.
Frequently asked questions
What is a semantic digital thread?
A digital thread whose data is mapped to an ontology, so it captures what the data means and how it relates across tools, not just where it is stored.
Why can’t AI just read our engineering tools directly?
It can read them, but it can’t reliably understand them. Each tool has its own schema and configuration approach, and the relationships between tools usually aren’t recorded anywhere. AI working from that data guesses at connections instead of following them.
What is BFO, and why build on it?
The Basic Formal Ontology is an ISO-standard top-level ontology (ISO/IEC 21838-2). In 2024, DoD and the Intelligence Community adopted it as their baseline standard for formal ontology work. Building on it gives SBE Vision’s MBE ontology a well-defined, standards-based foundation that other ontologies can align to.
Do we need to replace our PLM or other tools?
No. A semantic digital thread builds on the tools and PLM investments you already have.
Can we add the semantic layer later, once we’re ready for AI?
Only partly. Relationships and history created during engineering can’t be fully inferred after the fact, so the sooner the thread is in place, the more complete the foundation your AI will have.
How does SBE Vision use this for AI?
ThreadIQ™ works over the semantic digital thread, with AI autonomy matched to engineering consequence and every answer tied to its sources.
Build the foundation for AI you can trust.
See how SBE Vision builds a semantic digital thread across the tools your program already uses.
Sources
- U.S. Department of Defense, DoD Instruction 5000.97, “Digital Engineering”, Office of the Under Secretary of Defense for Research and Engineering, December 21, 2023
- Mark Jensen, DoD/IC Ontology Working Group, “Common Core Ontologies: Path to Standardization”, Ground Systems Architecture Workshop, February 28, 2024 (includes the CDAO and IC CDO directive)
- Juan Sequeda, Dean Allemang, and Bryon Jacob, “A Benchmark to Understand the Role of Knowledge Graphs on Large Language Model’s Accuracy for Question Answering on Enterprise SQL Databases”, arXiv, November 2023
- Dean Allemang and Juan Sequeda, “Increasing the LLM Accuracy for Question Answering: Ontologies to the Rescue!”, arXiv, May 2024