What is a digital twin?
A live virtual model of a physical thing, kept synchronised with it by a continuous stream of sensor data — which is what distinguishes it from an ordinary simulation or a 3D model.
The three requirements, all of which must be present:
A physical object or system — a machine, a building, a factory, a wind farm, a rail network, a patient.
A virtual representation, detailed enough to be useful.
A live data connection between them. Without the connection it is a model, not a twin, and a great deal of what is marketed as a digital twin is simply a model with a better name.
What it is for:
Monitoring the current state of something inaccessible, distributed or dangerous to inspect.
Predictive maintenance, detecting drift from expected behaviour before failure — the most commercially proven application.
Simulating changes safely. Testing a different operating strategy, a modification or a fault response on the twin rather than the real thing.
Optimisation, continuously tuning operation against actual conditions.
Training, on a model that behaves like the real system.
Design feedback, where real-world performance informs the next generation.
Where it is genuinely used: jet engines and turbines, where manufacturers monitor fleets in service; manufacturing lines; buildings and their energy systems; utility networks; and increasingly whole cities for planning and infrastructure.
Why it is harder than it sounds:
Data quality and sensor coverage, since a twin is only as good as what it can observe.
Model fidelity. A model accurate enough to trust for decisions is expensive to build and validate, and it drifts as the physical asset ages and is modified.
Integration across systems that were never designed to share data.
Cost, which frequently exceeds the value for anything other than high-value assets.
Security, since a live connection to operational technology is an attack surface.
The honest test for any claimed digital twin: what decision does it change, and how would you know if it were wrong?