CORE ARCHITECTURE
All seven layers →Engineering Ontology — EIO
One canonical vocabulary for assets, functions, failure modes and constraints.
Intelligence Graph — EIG
Every asset, document and decision connected as a queryable engineering structure.
Digital Twin Fabric — EDTF
Design intent and live operating state reconciled continuously, with the delta named.
ENGIN Solver
Deterministic engineering calculation — thermodynamics, hydraulics, structures, rotating equipment.
PhysicsNET
The validation gate — no recommendation reaches an engineer without passing the constraint set.
System of Truth — ESOT
One reconciled record per asset, every field carrying its source, revision and verification date.
Industrial assets produce data. They do not produce understanding.
Two decades of instrumentation gave every plant more signals than any engineer can read. What it did not give them was a machine-readable account of how the asset actually behaves — the design intent, the operating envelope, the failure physics, the decisions already taken and why.
Engineering knowledge is trapped in documents
P&IDs, datasheets, FMEAs, commissioning reports and vendor manuals hold the asset's real behaviour. None of it is queryable, and none of it reaches the control room at the moment of decision.
Analytics predict without knowing the physics
A model trained on historian data can flag a correlation. It cannot tell you whether the recommended setpoint violates a NPSH margin, a thermal limit, or a code-stamped design condition.
Nobody can audit an AI recommendation
When a suggestion cannot be traced to a governing equation, a measurement, or a named engineer, it cannot be signed off. Unauditable intelligence does not survive contact with a regulated plant.
One asset, four questions, and the people who have to answer them.
ENGINPILOT is built around the people who carry engineering, operational, and reliability accountability for critical physical assets.
Different roles. Different decision horizons. One requirement: make the engineering decision defensible.
Reliability engineers
Will this machine survive to the next outage — and on what evidence?
A canonical engineering record for each asset, connecting its design basis, FMEA, operating condition and open exceptions. Recommendations carry their governing evidence and physics validation before an engineer approves the decision.
Facility managers
Which competing intervention actually has to happen this shift?
Intervention proposals ranked by consequence against the operating envelope — not by alert age, dashboard severity or alarm volume. The decision is prioritised in engineering context.
Chief engineers
Is the engineering basis across this fleet defensible in a review?
One canonical engineering record per asset, with versioned constraints traced to their source documents, engineering assumptions preserved, and blocked decisions retained as part of the audit record.
Operations directors
Where is unplanned downtime coming from, which decisions are driving it, and is it getting better?
Decisions, evidence, interventions and outcomes recorded as one auditable series across sites — turning operational history into an engineering reliability case, not another management slide.
Engineering Intelligence is a new layer, not a better dashboard.
- —A CMMS or work-order system
- —An asset-performance dashboard
- —A general-purpose AI assistant pointed at plant data
- —A data lake with charts on top
- +A canonical ontology of engineering concepts, assets and relationships
- +A physics layer that evaluates every proposed action against real constraints
- +An evidence trail from raw measurement to signed engineering decision
- +Infrastructure other engineering systems build on
Three rules govern everything the platform is allowed to do.
Engineering defines
Intent, requirements, constraints and system behaviour are set by engineering — never inferred from data alone. Every model in the platform is bounded by a design basis a competent engineer could defend.
AI assists
Models retrieve, correlate, summarise and propose. They are rendered in a distinct visual language, always carry a confidence value, and never present a probabilistic output as an engineering fact.
Physics validates
Before any recommendation is shown, it is evaluated against conservation laws, material limits and the asset's certified operating envelope. A recommendation that violates physics is blocked, not ranked lower.
Seven layers between a sensor reading and a signed decision.
Every value declares how it was obtained.
A measured value, a solver result, an inference within tolerance and a twin simulation are four different kinds of claim. ENGINPILOT never renders them the same way.
Measured and confirmed against the model
Deterministic first-principles result
Inferred and inside stated tolerance
Twin output, not measured
Envelope or constraint breach
Insufficient data to assert
Awaiting acceptance by a named engineer. Nothing is actioned until it is signed.
Built for facilities where failure is expensive and physics is unforgiving.
Three sectors, at three different stages. Each is published with the stage it has actually reached, so an evaluator can tell a design-partner deployment from an active research programme.
Data centres
Chiller plants, CRAH loops, UPS trains. Thermal headroom governs uptime, and PUE is a physics problem before it is a reporting one.
The constraint library, twin models and validation cases are built for this sector, and running with design partners.
Hospitals
Medical gas, isolation-room pressure cascades and critical power, all under accreditation regimes that demand evidence.
Constraint encoding is under way with clinical engineering partners.
Energy
Generation, transmission and storage assets operating close to design limits with regulatory reporting attached to every excursion.
Open research on grid-scale constraint modelling, published as it is reviewed.
Read the argument before you evaluate the product.
The Engineering Intelligence Manifesto
The founding document of the category: why intelligence over physical assets must be constitutionally bounded.
Read essay →EP-WP-001 — Engineering Intelligence
The manifesto as a formal paper, with definitions, references and a citable revision number. No form, no gate.
Download PDF →Architecture papers and standards work
Ontology specifications, validation methodology, and how the platform maps to ISO 55000, IEC 62443 and ISO 14224.
Browse research →