When it comes to practical applications for AI and digital twins, Emulate3D’s John Pritchard has plenty of examples to share.
Artificial intelligence (AI) for manufacturing sounds like a straightforward proposition — or at least a catchy one. Yet despite the hype surrounding AI, it can be difficult to identify instances of practical use cases for it in manufacturing. The same might be said for the digital twin, though that argument is looking less sound all the time.
Beyond mirroring its development, AI may also have something to contribute to the digital twin, as John Pritchard, business manager for Emulate3D, explains: “Regarding AI, there are two areas immediately applicable to customers: generative AI to assist with building digital twins and AI that helps find relevant documentation.”
Before either of these use cases can be applied, however, manufacturers need to overcome a fundamental limitation on both digital twins and AI: as the models become more complex, they require more computing power. Emulate3D’s parent company, Rockwell Automation, is aiming to overcome this limitation by adopting NVIDIA’s Omniverse platform.
Collaboration in the Omniverse
“We have lots of 3D data sources in manufacturing, which is good for manufacturers, because it helps provide insights, but it can be tough running simulation models because it drives greater demand for compute,” explains Pritchard. “Plus, the models have to run in real time because we’re connecting them to real controls. Omniverse helps aggregate all those various 3D data sources and improve collaboration and performance.”
So, what does the integration of all this data look like? One of Pritchard’s examples involves the design of an automated storage and retrieval system (ASRS) and a separate but adjacent robotic racking system inside a new factory. By combining digital models for the ASRS and the racking system with data from architectural software, the three separate engineering teams can collaborate on the layout of the factory without having to wait days or weeks to see each other’s adjustments.
“It’s been possible to do this in a static way, but the Omniverse is dynamic,” Pritchard explains. “So when you’ve got loads going down a conveyor or being delivered by AMRs [autonomous mobile robots], that’s also happening in the model in real time. That’s really helpful when you’re trying to workout layout collisions and where to put people.”
Generative AI applications for engineers
Generative AI and digital twins both depend on data, and with the Omniverse serving as a platform, these tools can access data across business units with potentially game-changing results.
“Building digital twins is something that generative AI could and should help with in the future,” says Pritchard. “What I think will happen is that we’ll be able to produce multiple first versions of a model with natural language that domain experts can then choose from. With an iterative approach and some fine tuning and customization, you could quite quickly get from there to your end product.”
The other area where Pritchard hinted AI could make a difference in manufacturing is in documentation. Engineering documents contain not just text but drawings, schematics and other visual information that’s traditionally been difficult to search through. A machine learning model (or more likely a collection of them) trained on this dataset could be an invaluable assistant to working engineers.
“Also with the ability of AI to interpret technical documentation, you can ask it questions like, ‘Which screw terminal is input 15 on a 1756-IB16?’ and it will come back and say, ‘Oh, it’s screw terminal #9 on the left-hand side.’ Now, that’s a personal view of where I think AI is heading, but it seems to me like a completely obvious use case for engineers.”
Getting started with AI and digital twins
Given these kinds of applications, one might wonder whether all their data needs to be in the cloud before they can start to see the benefits of AI or digital twins, but in Pritchard’s experience, that’s not the case:
“I would say, nine times out of ten, people don’t start on the cloud. However, nine times out of ten, when they start scaling, they end up there. We did a project recently for a solar manufacturer and we ended up with 12 digital twin models and over 100 controller emulators. The only way you’d really want to stand up that many instances is on the cloud.”
Another obvious question is whether these tools only apply to greenfield projects, since those tend to be the leading examples in keynotes and presentations. Pritchard suggests this is more the result of marketing—showing off the newest, most exciting case studies—rather than a reflection of reality.
“There are at least as many brownfield projects as greenfield ones,” he says. “There’s a surprisingly large amount of retooling in manufacturing, so often the primary use for the digital twin is figuring out how to make those changes in an existing facility.”
One final issue to consider is whether AI and digital twins have a place in smaller organizations as well as larger ones. While Pritchard acknowledges the inherent advantage large players have in the form of bigger budgets, he also argues that there’s nothing in the technology of AI or digital twins that would prevent it being scaled down for small- and medium-sized (SMEs).
“We’re already seeing medium-sized equipment manufacturers adopt this technology,” he says, “Although, interestingly, one of the reasons why is because their customers are starting to expect that they’ll share a digital model of their equipment. In the automotive industry, for instance, some companies are not even allowing new equipment on site until they’ve validated it through simulation.”
As is often the case in manufacturing, innovation comes partly from a desire for process improvement, but largely from customer demand. Whether they’re prepared or not, SMEs should expect increasing pressure to adopt AI and digital twins in the coming years to ensure competitiveness and customer satisfaction.
“It’s also becoming part of the curriculum in colleges,” says Pritchard, “which means the engineers graduating now are going to be thinking differently about problems than the ways we did in the past.”