The Digital Twin Factory

District heating networks, electricity grids and public transport systems have something important in common. Each is a web of connected parts, and a decision taken in one place travels through the rest of the system in ways that are difficult to predict. At Arcada, we build digital twins for systems like these, shaping every model around a single operator's network, data and questions. The work draws on our research into AI-assisted modelling and on our ongoing collaboration with industrial partners in Finland.

What is a digital twin?

A digital twin is a working computer model of a real system. It is built from the system's own history: operational records, sensor readings, and the knowledge of the engineers who run it day to day. Because it behaves the way the real system behaves, it can be used to try out decisions before anyone commits to them.

A digital twin is a working computer model of a real system.

Operators tend to put two kinds of question to a twin. The first starts from a change and asks where it leads. If a production plant is taken offline for maintenance, what happens to pressures and temperatures across the network? If demand climbs sharply on a cold morning, where does capacity begin to run short? The second kind works in the opposite direction, starting from present conditions and asking which course of action is best. Given this week's energy prices, which combination of production units keeps costs lowest while respecting the network's limits?

Testing either question on the real system would be slow, costly or simply unsafe. On a twin, it becomes a routine exercise.

The trouble with building twins by hand

Most digital twins in industry today are crafted by hand. A team of specialists studies the system, writes out the governing equations, assembles the model and then tunes it until its behaviour matches reality. The process can take months, and the finished model is fitted so closely to one site that little of the effort carries over to the next.

Nor does the work end at delivery. Networks grow, equipment is replaced and operating policies change, and a twin that is not maintained drifts steadily away from the system it was meant to describe. Commercial platforms help with the underlying infrastructure, but they generally supply the building blocks and leave the customer to construct the simulation itself. For many operators, and smaller ones in particular, the overall cost is hard to justify.

Our approach

Our research asks how much of that manual work can be automated. We are developing a low-code/no-code method in which machine learning takes on the repetitive parts of model building, such as uncovering relationships in the data and selecting a suitable model structure.

Domain knowledge still matters a great deal. Your engineers tell us which variables count, which limits must never be crossed, and whether the model's behaviour looks credible. We then build the physical rules of the system, such as how heat travels through pipes or how load spreads across a grid, directly into the model.

Once the twin has been trained and checked against real measurements, a simulation layer lets you explore scenarios, and an optimisation layer searches for the best available decision within your constraints. When the physical system changes, the twin can be retrained on fresh data instead of being rebuilt from the beginning.

This is applied research carried out with industrial partners, and it remains under active development. That is also why every engagement is a close collaboration. The model we deliver reflects your network, your data and your priorities, because your people help shape it from the outset.

Our vision

To democratise digital twin technology, making advanced industrial simulation accessible and affordable for businesses of all sizes, driving efficiency and innovation across sectors.