What is a digital twin?

A digital twin is a digital representation of a specific real thing, such as a machine, a building or a field, that is kept up to date with data from that thing. What separates it from an ordinary model is the live link: when the real object changes, the twin changes too.

The three parts

Every digital twin has three parts.

The physical thing. One particular pump, one bridge, one field. A twin stands for an individual object, not for a type of object in general.
The digital representation. A structured record of that object: what it is, what it is made of, how it is arranged, its history and its current state.
The data connection. A flow of measurements and events from the real object to the digital one, often from sensors, and sometimes a flow of instructions back.

Take away the third part and you have a model or a drawing. These are useful, but they go out of date the moment the real object changes.

Model, shadow and twin

Writers on the subject often separate three levels, based on how data moves.

Term	Data from object to digital copy	Data from digital copy to object
Digital model	Manual, entered by a person	Manual
Digital shadow	Automatic	Manual or none
Digital twin	Automatic	Automatic

By this strict definition, a twin can send changes back to the real object, for example by adjusting a setting. In everyday use, many people call the middle level a digital twin as well. When someone offers you one, ask which level they mean. It is a fair question, and the answer tells you a lot.

What goes inside

A twin does not need to be a three-dimensional picture. Some are, because geometry matters for a building or an aircraft part. Many are closer to a well-organised database. What matters is the content.

Identity. A stable identifier for the object, so every record attaches to the right thing.
Fixed facts. Specifications, dimensions, location, materials and design documents.
Time series. Measurements over time, such as temperature, vibration, flow or moisture.
Events. Inspections, repairs, part replacements and faults.
Relationships. What the object is part of and what is part of it. A pump belongs to a line, which belongs to a plant.
Behaviour models. Physics-based equations, statistical models, or both, that estimate how the object responds to conditions.

Most of the effort goes into the unglamorous parts. Identity is a good example. Maintenance records may call a machine by one name, the sensor system by another and the purchase ledger by a third. Working out that these are the same machine is an entity resolution problem, and the twin is only as good as that matching.

What a twin is used for

The uses build on each other.

Seeing. Know the current state of the object without visiting it. This alone can justify the work for equipment that is remote, dangerous or spread over a large area.

Understanding. Compare current readings with the object's own history and with what the model expects. A bearing that runs a little warmer each week is easy to miss on a gauge and easy to spot in a record.

Predicting. Use the behaviour model to estimate what happens next. Maintenance teams use this to service equipment when its condition calls for it, not only by the calendar.

Testing changes safely. Ask "what if?" of the digital copy before trying it on the real one. What if we run this line faster? What if the building's heating starts an hour later? Simulation lets you try things that would be costly or unsafe to try for real.

Acting. At the strictest level, the twin sends changes back to the object. Many organisations keep a person in that loop for anything with consequences, which is a design choice and not a weakness.

Where twins are used

The idea grew out of engineering and manufacturing, where it is used for engines, turbines and production lines. It has since spread. Buildings and infrastructure use twins to manage energy and maintenance. Cities use them to study traffic and planning. Farming uses them for fields, where the "object" is a piece of land with its own soil, weather and crop history. What a digital twin of a farm looks like covers that case.

Common pitfalls

Bad data in, confident picture out. Sensors drift, fail and get moved. A twin that displays a wrong reading neatly is worse than no twin, because people believe it.
An unvalidated model. A simulation is only useful if its predictions have been checked against what really happened. Ask how the model was tested and how often it is checked again.
Twinning everything. A copy of a whole factory is a long project. A copy of the three machines that cause most of the stoppages is a useful one. Start where a decision depends on it.
No upkeep. The real object gets repaired, rebuilt and rewired. If nobody updates the digital copy to match, the two drift apart and the twin quietly becomes a model again.
Pictures over substance. A polished 3D view is easy to demonstrate. The harder and more valuable work is clean identity, trusted data and a record of where each value came from.

How to start small

Pick one kind of object and one decision you want to improve.
Give each object a stable identifier and gather what you already know about it.
Connect one reliable data feed and store its history.
Record where every value came from, so a surprising reading can be traced. See data provenance.
Add a simple model only when you have enough history to test it against.

Each step is useful by itself, which protects you if the project stops early. The plumbing underneath is an ordinary data pipeline, run on a schedule.

Where Prism fits

Prism Labs describes its work with four verbs. See: connect records, sensors and machines, which is the digital twin. Understand: ask a question and get an answer with its source. Plan: forecast and simulate before committing. Act: agents run systems and paperwork, and a person approves.

In agriculture, Prism Microclimate, built with FarmGrid, covers 961k fields across one whole state. It provides field-level weather, soil moisture, frost, spray and field-work windows, growing degree days and a morning brief per field. In other industries, this is what Prism can build, not what it has built.