UK How a digital twin works
A digital twin is a virtual model of a physical asset, system or place that is kept up to date with data from the real thing. It has three parts: the physical object, the virtual model, and the data connection between them. Sensors, cameras and scans feed the model, so it reflects the object’s current state rather than its design.
That live link is what separates a twin from an ordinary 3D model or a one-off simulation. Engineers can look at the twin to see how a machine, a building or a stretch of railway is doing now, compare it with last month, run what-if tests on the model, and send decisions back to the real asset. The idea was set out by Michael Grieves in the early 2000s, and NASA popularised the term for spacecraft that engineers needed to model from the ground.
Digital twin vs simulation, 3D model and digital shadow
The terms overlap, and the easiest way to separate them is by how data flows between the real object and the virtual one.
| Term | Data from the real object | Data back to the real object | Typical use |
|---|---|---|---|
| 3D model or CAD | None, built from the design | None | Design and drawings |
| Simulation | Assumed inputs, run on demand | None | Testing a design or scenario |
| Digital shadow | Automatic, continuous | Manual, if any | Monitoring and reporting |
| Digital twin | Automatic, continuous | Decisions or control sent back | Monitoring, prediction and optimisation |
In practice, many projects called digital twins start as digital shadows and add the return path later. A twin also usually contains simulations: it runs them on live data instead of assumed inputs.
Types of digital twin
Twins are often grouped by how much of the real world they model.
-
//
Component twin. A single part, such as a bearing, motor or battery, modelled to track wear.
-
//
Asset twin. A whole machine, such as a pump, robot or vehicle, built from its components’ data.
-
//
System twin. Several assets working together, such as a production line, a building’s plant room or a rail corridor.
-
//
Process twin. A complete operation, such as a factory or a network, used to test changes to scheduling and flow before making them.
How the data reaches the twin
A twin is only as current as the data feeding it. Most pipelines follow the same four steps, and the first two usually run on edge computers beside the asset or on the vehicle collecting the data.
-
//
Capture. Cameras, LiDAR scanners, vibration and temperature sensors record the asset.
-
//
Locate. For mobile capture, GNSS and an inertial measurement unit (IMU) tie every frame to a precise position. Our GNSS vs GPS comparison explains the positioning side.
-
//
Analyse. AI models detect, classify and grade assets and defects, often at the edge to cut the data sent upstream.
-
//
Update. Repeat runs or continuous streams keep the model current and show change over time.
Where digital twins are used
A rail corridor twin is built by driving cameras, LiDAR and positioning sensors along the track. Each run is matched to a precise location, so software can compare today’s images with last month’s and flag what has changed, from vegetation and fencing to defects in the track itself. The railway track inspection case study shows rugged railway computers capturing that data from the train.
Factories use twins of lines and machines to plan changeovers and feed predictive maintenance, as covered in our guide to AI in manufacturing. Buildings, utilities, ports and wind farms use them to monitor condition and plan work without sending crews to look.
Benefits and limits
The main benefit is seeing change early. A twin shows a drifting temperature, a widening crack or a slowing line before it becomes a breakdown, and it lets engineers test a fix on the model before touching the asset. It also cuts site visits, because condition can be checked remotely.
The limits are mostly practical. Twins are costly to build at full scale, they drift from reality if sensors fail or data stops, and a model is only trusted once its predictions have been checked against what actually happened. Security matters too, since a twin with a return path can change how a real asset behaves.
What it takes to build one
Most of the effort goes into data, not graphics. A twin needs reliable sensors, accurate positioning, a shared asset register so every reading is tied to the right object, and a way to keep the model current without drowning in raw data. Processing at the edge, beside the sensors, keeps bandwidth and storage manageable: the edge computer sends results and changes upstream rather than every frame. Start with one asset and one question, such as which bearings are wearing fastest, and extend the twin once that answer is trusted.
Quick answers on digital twins
Digital twins: common questions
What is digital twin technology?
The sensors, software and models that keep a virtual copy in step with a real asset. It combines IoT data collection, 3D modelling, simulation and, increasingly, AI analytics.
What is a digital twin in IoT?
The virtual side of an IoT system. IoT devices supply the live data, and the twin turns it into a model engineers can query, compare and simulate.
Is a digital twin the same as AI?
No, but they work together. The twin holds the current state of the asset. AI analyses that state to detect faults, predict failures or suggest changes.
How do you create a digital twin?
Start with the question, then the data. Choose the asset and decision it should support, fit the sensors to measure it, connect them to a model, and check its predictions against the real asset before relying on it.
Does a digital twin need the cloud?
Not always. Many twins are hosted in the cloud, but capture and first-pass analysis usually run at the edge, and secure sites often keep the whole twin on premises.
We supply rugged edge and railway computers for digital twin capture, with the cameras, positioning and storage integrated and tested as standard.