Digital Twin

A digital twin is a representation of a real thing inside a computer. This can be a machine, a wind turbine, an aircraft engine, or even an entire city. The special thing about it: the twin is not just a rigid 3D model, but is connected to the real original via measured values. Sensors on the original constantly measure temperature, pressure, rotational speed, or position and send these numbers to the model. If the real wind turbine turns more slowly, the value in the computer image changes as well. You can think of it like a mirror that not only shows appearance, but also calculates behavior along with it.

Why companies build machines twice

The greatest benefit lies in being able to experiment on the twin without risk. What happens if a turbine runs three months longer than planned? How much does a motor heat up if it spins twice as fast? Testing these questions on the real device would be expensive and sometimes dangerous. On the model, a failed attempt only costs computing time.

A second reason is predictive maintenance. If the model detects from the measurement data that a bearing is vibrating unusually strongly, it can be replaced before it breaks. A planned workshop appointment is always cheaper than a sudden production stoppage. Airlines and power plant operators save considerable sums this way.

It’s important to distinguish this from a simple simulation. A simulation is a one-time calculation with made-up assumptions. A digital twin runs continuously and uses real, current data. Without this ongoing connection to reality, it is not a twin, just a model.

Sensors, model, and feedback channel

A digital twin roughly consists of three parts. First, the sensors on the real object that measure constantly. Second, the model in the computer that calculates a state from these numbers. Third, a feedback channel through which insights from the model flow back into reality, for example as a warning message or as a changed setting.

The model itself can arise in two ways. It can be based on physical formulas, such as the laws of thermodynamics. Or it can arise through machine learning, meaning a program derives patterns from many hours of operation. In practice, both are combined: the formulas provide the basic framework, while the learned patterns capture the deviations that no formula describes.

A typical misconception is that such a twin is automatically accurate. It is only ever as good as its data and its assumptions. If a sensor fails or is incorrectly calibrated, the model confidently calculates nonsense. That’s why every serious twin needs a check on whether its predictions actually came true later on.

From the factory floor to the heart valve

Digital twins are most widespread in industry. Car manufacturers plan entire factories virtually first, before the first machine is even delivered. Software companies like Siemens, Nvidia, and Dassault earn money with the platforms on which such twins run. That’s why the term often appears in business news in connection with industrial software.

There are also examples outside the factory. Cities build twins of their road networks to calculate traffic flows or floods. In medicine, work is underway on models of individual organs so that doctors can plan a procedure in advance on the computer. Even sports teams use models of vehicles or bodies to optimize settings.

For you personally, the term is usually invisible, but not far away. Your phone’s battery is monitored by software that maintains a simplified model of its aging state. This is a digital twin in miniature. When your device tells you the battery capacity is at 87 percent, that number comes from such a calculation, not from a direct measurement.