What a digital twin actually is (and isn't)

"Digital twin" has become one of those phrases that gets stretched to cover almost anything on a screen. A spinning 3D model of a machine? Someone will call it a digital twin. A dashboard with a few live gauges? Digital twin. A VR walkthrough of a factory? Also, apparently, a digital twin.

Most of these aren't. And the gap between what the term promises and what's often delivered matters — especially in an engineering lab, where the whole point is to teach students how real systems behave. So it's worth being precise about what the words actually mean.

Start with the analogy that gives it the name

A twin is not a picture of a person. It's another person — one who changes as the first one does. That's the idea the term borrows. A digital twin isn't a representation of a physical thing; it's a living counterpart that stays in step with it.

The key word is "stays." A photograph of your heat exchanger is a representation. A CAD model is a representation. A beautifully rendered animation of coolant flowing through it is a representation. None of them change when the real exchanger fouls up, runs hotter, or springs a leak. They were true at the moment they were made and frozen ever since.

A digital twin is different because it's connected. When the physical system changes, the twin changes with it. When you'd want to know what the physical system will do next, you ask the twin.

The four things it actually takes

Strip away the marketing and a real digital twin needs four ingredients working together. Miss any one and you have something else — a simulation, a dashboard, an animation — but not a twin.

1. A physical asset that's actually instrumented

There has to be a real thing, and it has to be wired up to report on itself — temperatures, flow rates, pressures, vibration, whatever matters for that system. Sensors are what make the twin's connection to reality possible. Without live measurement, there's nothing for the model to stay in step with.

2. A model that predicts, not just displays

The model is the brain. Given the current conditions, it works out what the system should be doing — and what it will do if something changes. This can be built from physics (the governing equations of heat transfer, fluid flow, and so on) or learned from data, or both. What matters is that it computes an answer rather than replaying a pre-recorded one.

3. A live link between the two

This is the ingredient most fakes are missing. Real sensor data flows into the model, and the model's predictions flow back out to whoever's watching. The twin updates as the physical system runs. Break this link and you're left with a simulation running in isolation — useful, but not a twin, because it no longer knows what the real system is doing right now.

4. Validation against reality

A prediction nobody has checked is just a guess with good production values. A digital twin has to be validated — run the real experiment, compare it against the model, and quantify how closely they agree. The output is an error band you can state and defend: "the model tracks the measured outlet temperature to within two percent." Until that exists, you don't know whether the twin is telling you anything true.

So what are the things people mistake for twins?

None of these are bad. Several are genuinely useful, and we build some of them ourselves. They're just not digital twins, and calling them that muddies what students are actually looking at.

The pattern is clear once you see it. A visualisation has the looks but no brain. A dashboard has live data but no prediction. A simulation has a brain but no connection to anything real. Each has three of the four ingredients at most — and three out of four isn't a twin, it's a component of one.

Why the distinction matters in a teaching lab

You might reasonably ask: if the animation looks convincing and the students are engaged, does the label matter? In a teaching context, it matters more than anywhere.

The reason engineering students do lab work at all is to close the gap between theory and reality — to see that the neat equation on the board only approximates a messy physical world, and to develop judgement about when it holds and when it doesn't. A pre-recorded animation shows them the theory dressed up as reality. It can't ever surprise them, because it was authored to show exactly one outcome.

A real twin can surprise them, and that's the entire value. When a student raises the flow rate and the measured outlet temperature doesn't match what the idealised model predicted, they've learned something a manual could never teach: that real systems have fouling, losses, and imperfections, and that engineering is the discipline of accounting for them. The discrepancy between prediction and measurement is where understanding actually forms.

Sell a school a glossy animation labelled "digital twin," and you've given them a more expensive textbook. Give them a validated, sensor-linked model, and you've given them a place where the physics can push back. Only one of those is worth the name.