Getting smarter with digital twins together

At Eindhoven University of Technology, digital twinning is gaining traction. Whether scientists want to optimize machines, study the logistics in a hospital or improve the performance of a truck, a good virtual model can accelerate their research immensely. However, the High Tech Systems Center – now part of the AI institute EAISI – noticed that researchers were sometimes lagging compared to the industry. “They used to look for open-source and freely accessible tools on the internet, but those are often rather primitive,” says EAISI fellow Marc Hamilton. “Businesses are moving at a much faster pace than researchers, so before they got to the end of their PhD track, they could already be well outdated.”

HTSC initiated the Digital Twin Lab to bring researchers together, share experiences and reduce the cost of setting up experimental environments. “The lab is necessary to facilitate research with real added value,” according to Hamilton. “As a researcher, you need to have access to the latest tools and the best technology. You can’t rely on the basic software you can find online, when there are comprehensive tools on the market, like for instance Prespective, that give you a strong base to build on and that will prevent you from reinventing the wheel.”

The Digital Twin Lab started about two years ago and has grown to a network of over sixty scientists, professors and researchers. Since last December, it’s located in the House of Robotics on TUE campus. “The lab is like a kitchen where people can find all the necessary ingredients, bring them together, spice it up with the knowledge from other researchers and optimize their digital twin,” explains Shane Ó Seasnáin, director of the EAISI Program Board. “Actually, the coffee machine might be the most important part of the lab, as one of the biggest goals is to communicate with each other and share knowledge.”

Eye on the future

At the moment, the Digital Twin Lab is used mostly by researchers from TUE, but it’s emphatically reaching out to industry. “We’re still starting up and accelerating, but already about a dozen companies have shown concrete interest in collaborating. And we’re happy to help them,” tells Ó Seasnáin. “Some of them are just starting, others are already some time underway on their digital twin journey. What they have in common, though, is that they’re all innovative companies.”

That vision is almost a prerequisite for the proper use of digital twinning. Ó Seasnáin: “Building a virtual model is an investment, so you have to be smart.” Many users start their digital twin with a focus on one specific use case. Although that might be wise to get a fast return and get the ball rolling, Ó Seasnáin recommends looking further. “It would be foolish to waste all the hard work for only that one goal. You need to have your eye on the future as well. What do you need now for the first use case, and what other plans and ideas do you have in mind? Try to find the balance between short term and long term, and adapt your digital twin accordingly. In that sense, a digital twin isn’t that much different from buying a new machine or hiring a new employee; you always need to look where you want to go.”

With that in mind, when you start setting up your digital twin, you have to decide how accurate you need it to be. “Ideally, the digital twin is an exact carbon copy of the real world, reacting precisely as the physical system would,” Hamilton explains. “The problem is that we’re not there yet, and we might never reach that level with all the intricacies of the real world. A twin needs to fit the job. How you can determine the level of detail and add the most value to your project, without overcomplication, is something we investigate at the Digital Twin Lab.”

Self-constructing

As the Digital Twin Lab is part of a university, helping fellow scientists and companies construct a twin is only part of the job. Research on improving digital twin technology is just as important. Self-constructing digital twins are one topic TUE has set its eye on. “In its core, a digital twin is a collection of all the aspects you know about the system,” says Hamilton. “When there’s a physical counterpart, operational data streams can be a source. But the input can come from many sides and multiple disciplines – it can be about software, cost, behavior or electronics. All those data have to be collected and added to your model.”

“It’s rather complicated to bring everything together,” Ó Seasnáin adds. “A tool like Prespective can be helpful. But you also need physics engines and other tools. Plus, there are people in the loop that interact with the twin to improve it. And you want to embed the knowledge from within the organization into the twin as well. How should we combine all these sources in a way that we can understand them and use them to our advantage?”

Hamilton explains that the coupling is possible with the right software code. “When we understand what the data are and where they’re coming from, you can create programs and data pipelines to connect them properly with the right component in, for instance, Unity. But actually, you want to get rid of that programming step. What you need is enriched data that contains the necessary information to automatically find its way through the model and make the right connections without any programming. That shift is what’s needed and what we’re working on at TUE.”

Thinking assistant

Another research topic is the combination of digital twinning and artificial intelligence. “A twin is a fantastic representation of your system and AI can help to make it even more valuable,” states Ó Seasnáin. “Based on the model, AI can give recommendations, to improve your design, for better control or to optimize the maintenance planning. Imagine the combination of AI and your digital twin as a thinking assistant that, for instance, a maintenance engineer can use when he’s on-site, trying to fix a malfunctioning machine. Such a sparring partner could help him make better decisions and find the solution much faster.”

But AI can do a lot more when combined with a digital twin. “You can optimize for yield or throughput, of for energy consumption. You can better plan all your valuable assets and make better use of materials and resources,” sums up Ó Seasnáin. “A digital twin is a brilliant tool to facilitate and accelerate innovation.”

Hamilton adds: “A twin can also work in service of AI. In the virtual world, you can create and try out many variations. In case the learning engine is insensitive to the difference between the twin and the real system, the algorithm can learn in the twin environment. Think of an autonomous car. When you want to teach the AI algorithms how to brake for crossing children, they can fail a thousand times in the virtual world, until it’s robust enough for the real world.”

This article was written in close collaboration with EAISI/High Tech Systems Center. Main picture credit: EAISI/High Tech Systems Center