Industrial digital twins

An industrial digital twin is a digital representation connected to information about a physical asset or process so that it can support decisions as conditions change. It may combine engineering models, operating data and simulation. Useful applications include process diagnosis, maintenance planning and testing proposed changes. Economic value depends on a specific decision improving enough to cover integration and continuing model maintenance, with evidence that predictions remain trustworthy within their intended operating range.

Core relationship

Physical operation linked to a digital representation

Decision uses

Diagnosis, prediction and process planning

Technical requirement

Validation and managed uncertainty

Commercial metric

Measured benefit after continuing operating cost

What the model must represent

A twin needs a defined scope: a machine, production cell, process or wider system. Data describes relevant operating conditions, while the model expresses relationships needed for the decision. The update frequency should match that decision. A maintenance-planning model and a control application can require very different response times.

NIST describes manufacturing twins as synchronised virtual models and highlights interoperability, validation and uncertainty. A visually detailed model is not enough. It must represent the variables that determine the outcome the user intends to change.

Evidence and limits of the model

A model can be technically correct in its implementation yet unsuitable for an operational decision. Validation asks whether it adequately represents the relevant real-world behaviour. That requires observations, a stated operating range and a way to quantify uncertainty. Equipment modifications or new product mixes may invalidate earlier assumptions.

NIST's manufacturing use-case work explains how the ISO 23247 framework can structure implementations. A standard framework supports common terminology and architecture; it does not certify that a particular prediction is accurate or that a deployment will produce savings.

Where the business case forms

Our assessment is that the strongest starting point is a costly, repeated decision with an observable result. Examples include choosing a maintenance intervention, adjusting a production sequence or evaluating a proposed process change before disrupting equipment. The benefit must be measured against an explicit baseline, with other operational changes accounted for.

NIST's economic work emphasises costs, benefits and decision making. An initial model should include sensors, connectivity, integration, licences, computing, engineering time and validation. Continuing costs include maintaining interfaces, refreshing models and investigating cases where predictions disagree with actual behaviour.

Who owns the continuing service

A twin may be delivered as software, an engineering project or an ongoing analytical service. Each route allocates responsibility differently. A subscription is more defensible when the supplier keeps the model useful as equipment and operations change, rather than simply maintaining access to a static model.

Contracts should establish access to operating data, rights to derived models, portability and responsibilities at the plant interface. Performance-linked pricing also needs an agreed measurement method and treatment of changes outside the supplier's control. For the buyer, a successful pilot becomes scalable only when the next installation needs less bespoke work while retaining validated decision quality.