“For a digital twin, there has to be a digital master and a digital shadow.”
The digital twin was first described by American Michael Grieves at the University of Michigan in 2002 (see page 14). Yet only modern technologies have allowed the concept to expand its potential: big data applications, IoT, the cloud and sensors. NASA was one of its first adopters: While developing a robot to be used on Mars, NASA combined a model with real data from the Red Planet.
However, the digital twin has been out of the exotic spheres of space technology for some time and has come to industry, at least as a vision. The possible applications are extensive (see box).
The technology can realize its full potential in combination with applications when, for example, the data from production monitoring is used as input for virtual start-up. Or when quality management finds the causes of known defects in the digital twin and the parameters for future production are then modified accordingly.
Ultimately, the continued networking between these applications could lead to a bidirectional system where the digital twin provides feedback to its physical brother - resulting in a self-controlling system. “Technically, that’s definitely feasible,” confirmed Lindow. “However, for that it would be necessary to enrich the model with artificial intelligence and machine learning.”
Where are digital twins already being used in industry? “It’s not yet happening nationwide,” said Lindow. “I have the feeling it’s still being researched in most companies.” For a field test, his institute worked with a scooter sharing service provider. The scooters were outfitted with sensors, so the use of every vehicle could be documented. Where was it checked out? Where was it checked back in? How far did it travel? And what rate was paid? This made it possible to compile a detailed analysis of individual users and the entire fleet, which may even make it possible to adjust the price structure and battery charging times.
According to Lindow, the application is used at best in subfields. One example is logistics: At Airbus, digital twins help coordinate the 12,000 partners supplying the three million parts that make up an A319. Another is product development: In automobile development, engineers test load scenarios for individual components or entire vehicles, down to a virtual crash test. Fiat Holding’s sportscar manufacturer, Maserati, has used this to cut the development time of its vehicles nearly in half.
“Digital twins have an effect on the business model,” explained Lindow. That hinders its adoption. If a machine manufacturer, for example, realizes its customers want to buy the machine capacity but not the machine itself, does it then become simply a service provider?
Lindow emphasized that the adoption of a digital twin also needs to be preceded by an extensive analysis. “I need to be clear about what I want to achieve with the twin and where the business value is.” Does my system already have sensor technology? How much data do I need? Is real-time monitoring necessary or is the data collected at specific points in time adequate? In the end, sensors, data transmission and data analysis are not free.
Lastly, a company needs to look at what data outside of production could be relevant. For a company that maintains aircraft turbines, it could be interesting not only to know how many flight hours a turbine has on it, but also what routes they flew. This is not obtained from the airline’s data, but from a third party, such as a flight tracker. “At the end of the day, aircraft flying primarily over the Sahara are exposed to entirely different loads than those flying over the Atlantic,” added Lindow. For the same reason, automobile manufacturers prefer to test their prototypes in wastelands - whether in the Arctic Circle or Death Valley. Considering the demands of those locations by dust, sand and co., the Formula 1 cars have it much easier - they always drive on hard asphalt.
The data continuously collected from production creates a comprehensive image of the current state of the production system in operation.
Analysis of production data can be used to increase efficiency. Comparison with a similar production system at another location can also be used.
By reviewing historical data or comparing with a similar production system, it is possible to calculate upcoming failure in components and wear parts.
Continuous monitoring of product quality offers clear advantages over random inspection.
Conversions of production to other products or smaller series (keyword: customizing) can be run through first in the digital twin.
Analysis of historical comparative data makes it possible to predict the performance of a system that has not yet been constructed.
Optimization of the supply chain can result in significant increases in efficiency, especially for just-in-time or just-in-sequence production.
Product lifecycle management is relevant for capital goods especially. Even for the end of life of a product, it can be interesting to know what materials are in the product to facilitate recycling.
Virtual simulations help with development. Data collected from the use of a product can also help develop and improved version (design feedback).