What Engineering Knowledge Graphs Must Express

Context: MIN v1.1.0 — foundational ontology for engineers.

Engineering knowledge graphs face a specific challenge: they must represent not only what exists and what happens, but also what determines, what is believed, and what is uncertain. MIN addresses this through seven axes.


Axis 1: Materiality — what exists physically?

Question MIN construct
What is physically there? Object
What happens / was done? Process
What data is available? Data
What arises only between partners? Boundary

A tensile test has a specimen (Object), the testing procedure (Process), measurement data (Data), and contact friction between specimen and grip (Boundary). All four are causally efficacious — they cause something in the world.

Key relations: - hasInput / hasOutput — what goes in, what comes out - hasComponent — mereological composition - generates / generatedBy — process produces data - describes / describedBy — data describes nexus - bounds / hasBoundary — boundary between partners


Axis 2: Formality — what determines without causing?

Question MIN construct
Which natural law holds? Lex
Which formal model describes it? Structura
What could happen? Possibile
What requirement must be met? Norma
What is collectively recognized (grade, standard, type)? Institutio
What is held to be true? Epistemicum

A materials engineer needs all six: Hooke's law (Lex), Johnson-Cook model (Structura), fatigue crack scenario (Possibile), Rm >= 270 MPa (Norma), DC04 steel grade (Institutio), and "Hooke holds here up to 200 MPa" as an evaluated hypothesis (Epistemicum).

Key principle: Forma causes nothing — but without Forma, every effect would be different. Forma is the grammar, Nexus is the sentences.

Key relations: - governs — Lex constrains Process - formalizes — Structura constrains Nexus - evaluates — Norma constrains Nexus - concerns / alternativeTo — Possibile relates to Nexus - typifies — Institutio determines what kind a Nexus counts as - comprises — Institutio bundles atomic Forma instances


Axis 3: Agency — who acts?

Question MIN construct
Who executes? Agent ∩ Object (human, machine)
Which software decides? Agent ∩ Data (ML model)
Which organization is responsible? Agent ∩ Institutio

Not just humans. A CNC machine, an ML model, a standards committee — all act selectively. Agent is a cross-category under Entity, not a subcategory of Nexus. Co-typing is mandatory.

Key relations: - performs / controls — agent carries out process - actsOn — agent acts on object (derived via property chain) - owns — ownership / responsibility - constitutes / recognizes — agent creates / acknowledges institution


Axis 4: Causality — how does it connect?

Within the world (Nexus ↔ Nexus)

  • What goes in, what comes out? → hasInput / hasOutput
  • Who carries out the process? → performs / controls
  • What is it made of? → hasComponent

Between world and determination (Nexus ↔ Forma)

Three generic bridge relations form the Forma lifecycle:

Relation Direction Semantics
originates Nexus → Forma Nexus brings forth NEW formal determinants
realizes Nexus → Forma Nexus makes EXISTING formal determinants actual
constrains Forma → Nexus Forma restricts what Nexus can be or do

Plus the encoding bridge:

Relation Direction Semantics
encodes Data → Forma Data carries formal content (DIN document encodes Norma)

Key distinction: - realizes changes the world (ontological). - confirms changes what we know (epistemic).


Axis 5: Epistemics — what do we know, and how certain?

Question How MIN expresses it
Who believes it? heldBy Agent (or agent-free)
About what? about Entity
How certain? hasConfidence [0..1] + hasConfidenceType
Which status? hasEpistemicStatus (5 values)
Based on what? supportedBy / underminedBy Nexus
What was believed before? supersedes (Forma → Forma)

This is what engineering knowledge graphs typically could not do before MIN v1.1.0: formalize the difference between "42% recyclate content" and "42% recyclate content, statistically confirmed with confidence 0.92 supported by XRF measurement data."

Two complementary patterns

Pattern Carrier Semantics Since
Popperian Process Process confirms/refutes Forma v1.0.0
Evidence-centric Epistemicum Stance supported/undermined by Nexus v1.1.0

Both are needed. The Popperian pattern is compact for simple assertions ("the tensile test confirms Hooke's law"). The evidence-centric pattern is needed when epistemic state evolves, confidence must be typed, or multiple agents disagree.


Axis 6: Uncertainty — what do we not know?

Question MIN construct
Quantifiable risk? Possibile + isQuantifiable = true
Deep / Knightian uncertainty? Possibile + isQuantifiable = false
Missing evidence? Epistemicum without supportedBy
Dissent? Two Epistemicum, same about, different agents, different status

Known unknowns: "Component failure with 2% probability." Normal risk, a distribution is assignable.

Unknown unknowns: "Climate change impact on supply chain." No distribution assignable. This is not a bug but a deliberate statement.


Axis 7: Provenance — where does the information come from?

Question MIN construct
Who produced the data? Data + generatedBy Process
Who encoded the formal content? Data + encodes Forma
Who brought forth the formal determinant? Forma + originatedBy Nexus
What evidence supports the belief? Epistemicum + supportedBy Nexus

Provenance in MIN runs through three complementary mechanisms: - Data provenance: generatedBy traces which process produced the data. - Forma provenance: originatedBy traces which nexus brought forth the formal determinant. - Epistemic provenance: supportedBy / underminedBy traces which evidence supports or weakens a belief.


Axis 8: Temporality — when does it happen?

Question MIN construct
When was it recorded? hasTimestamp (Entity → xsd:dateTime)
When did the process start? startedAt (Process → xsd:dateTime)
When did the process end? endedAt (Process → xsd:dateTime)
What temporal sequence? Process chains via hasInput / hasOutput
What replaced what? supersedes (Forma → Forma)

MIN embeds temporality implicitly rather than as a separate temporal model. Timestamps attach directly to entities and processes. Temporal ordering emerges from process chains: if Process B uses the output of Process A, then A precedes B. Epistemic succession uses supersedes — when a new Epistemicum replaces an old one, the temporal dimension is captured structurally, not through explicit time intervals.

What alignment provides: - OWL-Time (alignment/min-time.ttl): min:Process rdfs:subClassOf time:ProperInterval — processes can be treated as time intervals by OWL-Time reasoners. - PROV-O (alignment/min-prov.ttl): min:startedAt rdfs:subPropertyOf prov:startedAtTime — timestamps become visible to PROV-O tools.

What belongs in the domain layer: - Time series data (sensor readings at millisecond resolution) - Lifecycle phase models (design → manufacture → use → end-of-life) - Calendar and scheduling constraints - Allen interval relations (before, during, overlaps)


Summary: eight axes

1. Materiality     What exists physically?                → Nexus
2. Formality       What determines without causing?       → Forma
3. Agency          Who acts?                              → Agent
4. Causality       How does it connect?                   → Relations
5. Epistemics      What do we know, how certain?          → Epistemicum
6. Uncertainty     What do we not know?                   → Possibile + isQuantifiable
7. Provenance      Where does the information come from?  → Data + originatedBy + supportedBy
8. Temporality     When does it happen?                   → hasTimestamp + startedAt/endedAt

What MIN deliberately does NOT cover

Concern Why not Where instead
Truth "X is true" is not a MIN concept. Epistemicum says what is believed, not what is.
Bayesian propagation Confidence propagation is a Process, not an axiom. Domain layer
Time series Temporal resolution belongs in the domain layer. sdata-measurements
Physical units Unit systems are orthogonal to ontological structure. QUDT alignment
Emotions "I feel the component will fail" is not propositional.
Self-reference "I believe that I believe" is not cleanly modelable in OWL.