The terms this platform relies on, defined precisely

The set of engineering decisions, assumptions and limits an asset was designed against — nameplate ratings, performance curves, materials, allowable conditions and the operating case each was chosen for. It exists in datasheets, curves and commissioning records, and it is the reference every later judgement is measured against.

The region of operating conditions within which an asset may run without exceeding a design limit. It is multi-dimensional — pressure, temperature, flow, speed and load are bounded together, not independently — so a machine can sit inside every individual limit and still be outside its envelope.

The machine-evaluable expression of an asset's limits: the envelope, its conservation laws and its statutory restrictions written so that a proposal can be tested against them. Assembled per asset in four passes — class defaults, design basis, site amendments, regulatory overlay — and pinned to the revision of every document it drew from.

A deliberate reduction of an asset's permitted operating limits below its design values, applied after damage, repair, ageing or a change in service. A derate tightens the envelope and must carry an expiry or a review date — an unlifted derate that outlives its cause is one of the most common sources of a wrong limit.

The manufacturer's declared performance at a stated reference condition. It is a single point on a curve, not a capability across conditions — reading a nameplate as an all-conditions limit is a frequent error, because ambient temperature, fouling and inlet conditions move the real limit continuously.

The difference between the net positive suction head available at a pump inlet and the head the pump requires to avoid cavitation. Expressed as a ratio or a difference in metres of liquid. Margin erosion is gradual and is usually a system change rather than a pump fault.

The position in a plant that a function is performed at, independent of the physical unit currently installed there. Distinguishing location from equipment is what allows failure history to survive a replacement: the location keeps its history, the removed unit keeps its own.

The manner in which an item fails to perform its required function — stated as the loss of function, not as its cause and not as its symptom. "Bearing seizure" is a mode; "inadequate lubrication" is a cause; "high vibration" is a symptom. Conflating the three is the most common defect in reliability records.

Failure mode and effects analysis: a systematic enumeration of the ways an item can fail, the effect of each on the system, and the detection and mitigation in place. FMECA extends it with criticality. Its value depends entirely on being maintained after commissioning — a static FMEA describes an asset that no longer exists.

The consequence of an asset's failure to safety, production, environment and cost, used to rank engineering attention. Criticality belongs to the function in its plant context, not to the equipment type — the same pump model can be critical in one service and spare in another.

Measurement and trending of parameters that indicate a change in an asset's condition — vibration, temperature, oil chemistry, acoustic emission, electrical signature. It detects change; it does not by itself explain mechanism, which is why an unexplained trend is a question rather than a finding.

A physics-based model of a specific asset, kept aligned to that asset by measurement. A twin is defined by the equations it solves and the asset it is bound to — a dashboard fed by live tags is not a twin, because it predicts nothing and therefore cannot disagree with anything.

The difference between a modelled value and a measured one. The residual — not the measurement — is where developing degradation first becomes visible, because a fault appears as a growing disagreement between what physics expects and what the plant reports.

Closing mass and energy balance across an asset boundary so that measured and modelled states can be compared honestly, then attributing any remaining difference to instrument error, boundary-condition error or genuine model error. Attribution is the step that decides whether an engineer inspects a machine or a transmitter.

The residual magnitude beyond which disagreement is treated as significant. It is a judgement made by a responsible engineer with a written basis — not a value inferred from historical variance, which only encodes whatever the asset happened to be doing while the data was collected.

Carrying the measurement uncertainty of every input through a calculation to the uncertainty of its result — analytically for linear expressions, by Monte Carlo sampling where the expression is strongly non-linear or the inputs are correlated. A calculated value without a band is an incomplete statement.

Treating a quantity's physical dimension as part of its type, so that adding or comparing incompatible quantities is rejected before evaluation rather than producing a dimensionally coherent but numerically wrong result. Dimensions are tracked as exponents over the seven SI base quantities.

The declared epistemic status of a value: VALIDATED, CALCULATED, ESTIMATED, SIMULATED, VIOLATED or UNKNOWN. Rendered wherever the value is rendered, never in a tooltip — a number whose status is hidden invites being read as more certain than it is.

Evaluating a proposed action against an asset's full constraint set before it is shown to an engineer. Deterministic, total — every constraint is checked, not just until the first failure — and returning three verdicts rather than two: pass, block, or indeterminate where an uncertainty band straddles a limit.

The retained account of an engineering decision: what was proposed, on what evidence, which constraints applied, what the physics verdict was, who accepted it and when. It is written whether or not anyone reads it, because its value appears in a review months later.

A recorded departure from expected engineering state, addressed to a named responsible engineer: a residual past tolerance, a blocked action, a constraint in dispute, an expired calibration. An exception is closed by a decision, never by a timeout.

What a person actually did with a recommendation or an exception — accepted, rejected, deferred, escalated — recorded against their name. Disposition is what turns a set of proposals into an auditable history rather than a queue.

A recorded, authorised change to a twin's model — treated as an engineering change with a written basis and a named signature, not as maintenance. Prior revisions are retained and replayable, because a decision that cites a model which no longer exists cannot be audited.

A governed vocabulary of entity types and the permitted relationships between them. It is not a database schema: a schema says where data goes, an ontology says what a thing is and what may be true of it. Without one, the same physical situation is described three different ways in three plants.

A store in which relationships are first-class and traversable, so a question can follow a path — this asset, to its design basis, to the clause that set this limit — rather than requiring a join to be anticipated in advance.

A time-series archive of process and equipment measurements. It records what happened, at high fidelity, with no account of why — which is why a decade of historian data can coexist with no improvement in engineering understanding.

Piping and instrumentation diagram: the drawing that states what is connected to what, with which instruments and which control elements. It carries the plant's topology, which is why it is a primary source for an asset record rather than a reference document.

The layer between an instrumented physical asset and a defensible engineering decision: a canonical engineering record, a physics gate over it, and a retained account of every decision made through it. Distinct from data, which describes what happened, and from prediction, which proposes what might.

A persistent, addressable entity bound to one physical asset, holding that asset's design basis, history, FMEA and open exceptions. It proposes; it never decides and never acts. One agent per asset tag — never a fleet-wide assistant.

A deterministic result is calculated from first principles and is reported as calculated or validated. A probabilistic result is model output and is reported as recommended, predicted or estimated, always with a confidence value. The two are never presented in the same register.