Digital Twin
Digital Twin
What is a digital twin?
- An exact digital copy of reality
- Simulate and predict in the virtual world
- Used in manufacturing, construction, and energy
What a digital twin really changes is the order of decision-making. Factories used to “just run it once” and look at the result. To decide whether to raise line speed by 5% or lower mold temperature by 2 degrees, you had to stop the real line, change the conditions, and run a trial batch, and each trial cost half a day and hundreds of defective parts. With a twin, most testing happens in the virtual world before the real one. When the instinct of a foreman with 30 years of experience clashes with a headquarters engineer's calculation, you can find out who is right without any physical loss.
What the field calls a twin actually falls into three tiers. A 3D copy of the drawings is a digital model. When sensor values flow one way into a virtual model, it is a digital shadow. When results calculated in the virtual world change the settings of real equipment, that is two-way connection, a digital twin in the true sense. Update frequency also depends on the use. Process control needs seconds, predictive maintenance needs minutes or hours, and for building or city planning a day or a week is enough. Setting the frequency shorter than the use requires only inflates communication and storage costs.
It also helps to draw a clear line between a twin and conventional simulation. A simulation runs on assumptions that people enter, while a twin keeps updating those assumptions with real data from the equipment, current as of yesterday. That makes adoption demanding. Sensors must already be in place, at least six months of data must have accumulated, and above all you need calibration, which means checking how well the model matches reality. Matching one machine with just two or three sensors is faster than modeling every line at once.
The most common failure is mistaking 3D visualization for a twin. Companies spend hundreds of millions of won on a screen that shows a factory rotating on a meeting-room monitor, and often that model has never once been checked or calibrated against the real equipment. After two or three missed predictions, the floor switches the screen off and the next investment request is rejected. It is also common for sensor wiring to change while the model runs unchanged for months. A twin is a twin only while it is maintained, not on the day it is built, and without someone who keeps tuning the model to reality, it stays an expensive 3D picture.
Ilyoung JungExpert in smart factories, equipment management, and Industry 4.0
John KimInstructor John Kim, author of “B2B Marketing Design” and CEO of Nanum ConsultingThe concept developed from uses such as NASA's spacecraft simulations, and spread across industries as IoT and sensor technology became widespread.
Consider a hypothetical case. Company A, an automotive-parts injection molder with 320 employees and annual revenue of around KRW 180 (≈ US$1) billion, runs 42 injection molding machines in three shifts. Last year, unplanned equipment stops averaged 62 hours a month, and half of them were defect-related stops caused by abnormal mold temperature. The production engineering team decided to try adding a twin of the machines as a 6-month project.
- Mr. Kim, a manager on the production engineering team, skipped all 42 machines and picked just Machine 3, the one with the most defects. He fitted it with seven sensors, covering mold temperature, injection pressure, cycle time, and more, and sent data every 2 seconds.
- The team fed 8 months of production history into a virtual injection molding machine, and its defect rate was off from the real one by 3.1 percentage points. Calibration to close that 3.1-point gap took 6 weeks on its own, and only then did they start to trust the predictions.
- Kang, the maintenance foreman, was skeptical at first: 'What does a screen know?' But three days after the twin flagged a crack in a hydraulic hose, the hose actually burst, and now he opens the prediction list first thing every morning.
- After running 40 simulations of lowering summer coolant temperature by 2 degrees, the team got the plant manager's approval and applied the change to Machine 3 only. Skipping physical trial runs saved about 900 parts that would have been lost in testing.
- Six months later, unplanned stops on Machine 3 fell from 14 hours to 5 hours a month, and the defect rate dropped from 4.2% to 2.6%. Company A approved a second project to extend the same approach to 11 identical machines.