Nvidia expert explains: What AI brings to mechanical engineering

The current breakthroughs in digital AI enable its use in the physical world, especially in industry. What does the term physical AI specifically mean in the industrial context for you?

Timo Kistner: What we mean by this is the combination of any AI models with a physics-based simulation. For example, if I use a language model on my smartphone today - no matter what tool it is - the tool itself is very powerful and I get generated answers that I couldn't get a few years ago. The limitation these tools have today is that they don't have spatial context. That means the tool doesn't know the context in which I'm moving and doesn't give me corresponding answers to that context. And that's what we mean by physical AI: We establish the context between the AI model and the space in which we move.

What does this mean for the industrial environment? For example, where I previously trained my model with the data of a machine for quality control, in the future I will be able to create a digital twin of the machine and simulate it based on physics. This is important: Not a simulation in the sense of a rough approximation, but a truly physically accurate, physics-based simulation. Then I am able to train the AI model directly in this physically accurately depicted digital twin.

This means I no longer have to train the AI model directly with the machine's data, but I can do it in my digital twin, based on a physically accurate representation.

Your company also has a technology for this. The Nvidia Omniverse is a platform for the industrial metaverse and for creating digital twins. What advantages does the Omniverse offer to machine builders?

These benefits are brought to you by the industrial metaverse

What potential does the metaverse have in the industry and how can companies utilize it? Expert Prof. Bauernhansl explains in the interview.

Kistner: First of all, you have to understand what Omniverse is intended for. Omniverse is a platform for collaboration and development. Collaboration from the perspective that we bring together a wide variety of development tools. This means that when I develop my products today as a machine builder or as an industrial company, I usually have a multitude of tools that I use to develop these products or to plan and optimize my factory, introduce products in the factory, and so on. These tools more or less communicate with each other.

This is exactly what we want to overcome with the collaboration aspect by saying we are able to bring the data from the individual tools - mechanical, electrical, whatever kind of data - into a virtual space. This means I connect all these tools and am suddenly able to map the individual parts of my product or my factory in a virtual space. This means I have a much earlier understanding of what my product actually looks like. I see much earlier: Oops, in the planning of my factory, there is somehow a collision between a robot and another machine because I cannot recognize the connections in the individual tools. This means I become more efficient as a result.

We have examples from the industry where we see that just by bringing these tools together, we achieve significantly more efficiency and thus significantly more speed/time-to-market.

And the second aspect is the development side: Omniverse as a platform to develop physical AI. This means I also use the opportunity to bring together all the data sources to train my intelligent machine, my robot, in this virtual space. This means I can really work virtually from the beginning of development to the certification of my product, and this also brings me efficiency again.

In the industry, the first question often asked is, does it pay off or why should I do it? Do you have examples where physical AI has already proven economically viable?

Kistner: Absolutely. There are several examples. BMW, as one of our partners, uses Omniverse, among other things, to advance both collaboration and this physics-based simulation. One example is the new factory designed in Hungary in Debrecen: colleagues talk about a 20 to 30 percent efficiency gain simply because the factory planners can see together in Omniverse what the individual parts of this factory look like. Until a few years ago, we only saw sections of a factory in the individual tools and then had to put them back together across these tools. Today we are able to map the entire factory.

Another example from our partnership with Siemens: a ship from HD Hyundai, which was developed based on various tools and the data sources were brought together to visualize in Omniverse in real time: what does it look like right now? What do the electrical systems look like? What do all the pipelines look like? We're talking about 7 million individual parts. 7 million individual parts of this entire ship, which we would not have been able to visualize before. Today, every change we make in the tools is mapped in real time in Omniverse. This makes me more efficient and significantly faster, brings the development teams closer together, and gives me a better understanding of what the overall construct looks like in the end.

Nvidia also emphasizes the role of AI-driven robotics to increase productivity and reduce errors in manufacturing. In this area, they work with companies like Kion and Accenture to optimize supply chains with AI-supported robots and digital twins. What experiences are there already, and what specific benefits do companies gain from such optimization of supply chains and factory processes?

Kistner: We keep coming back to the approach of collaboration and development platform and the management of very complex processes. What we simulate there [at Kion] is, for one: How can we optimize warehouses accordingly? This means I can suddenly conduct simulations where I can look at a much larger scale of different scenarios. This is also an important aspect. Different scenarios that I would have had to painstakingly build by hand before, I can now generate with synthetic data to really find the best setup for my warehouse. That's point one.

The second point is, I am then able to simulate exactly the robots you mentioned - AMRs, AGVs - with much greater depth and accuracy and with a much higher number of corresponding robots. All of this puts me in a position to manage a completely different level of complexity in the end and thus become more efficient again.

What do you think are the medium- to long-term potentials of AI in the industry?

Kistner: First of all, we are convinced that little will happen in the future without AI. AI will be an essential part of every manufacturing process, every development, every product. It is important to start with it now. I think it is difficult to predict what exactly the impact will be for each individual company. Whether in terms of more sales of my products or in terms of the efficiency of my factory's products - that depends on the individual company. What is important is that I deal with it now. I have to start with it now, because otherwise, in the end, I will see a gap compared to my competitors. This concerns both artificial intelligence and the topic of physics-based simulations.

We have talked a lot about technology now, but people still play a big role. It is often a challenge to communicate the value and practical applications of AI internally. Do you have any tips for companies on how to make the benefits of AI understandable and create acceptance within the company?

Kistner: It is important that I start with it now. That means integrating simple things into daily routines - whether it's copilots, which come in numerous variants, or clarifying how to use them efficiently. The copilot is not necessarily just the next evolution of search. I can certainly use it that way, but it is not necessarily what makes AI usage strong. Rather, if I set appropriate framework parameters, I will also achieve significantly stronger results.

How digital twins and AI become innovation drivers

AI and circular economy rely on data quality - there is a need for improvement here. Guido Reimann, deputy managing director of the VDMA Software and Digitalization Association, explains which services are based on digital twins and AI.

That means introducing simple means and training employees: What can I actually do with it? And I think what is important then, especially when we talk about the factory environment, but also in product development: How can I use this technology? How can I really involve everyone? I can now involve every single employee at the beginning, but I have to start with a larger group. I have to train employees who then serve as multipliers to motivate the rest of the workforce to use these tools.

I think what we do see is: sometimes the beginning might be a bit difficult. But once you get used to it - the effect and efficiency you gain from it is quite impressive, and I haven't seen a project yet where we found that, well, okay, the employees are not enthusiastic in the end and actually don't want it anymore.

What misunderstandings do you encounter most often when it comes to AI in the industry? You now have the chance to clear them up.

Kistner: By now, I must say, many of these misunderstandings have been resolved and cleared up. Certainly, a question that has come up repeatedly was initially, what will happen with AI in the end? Is my job threatened in any way because of it? Again, it must be repeated: we won't have all the skilled workers we would need in the end, and accordingly, these tools are absolutely necessary to continue operating the company efficiently.

The question is not whether AI will take over my job, but whether it is the person who deals with AI and can use AI that might threaten my job in some way. It is important that we start with these topics now, that I deal with it now. It's not about doing AI for the sake of AI and that I absolutely have to implement the next project within the next five or six weeks. But I have to deal with it now: what are the levers I have? And where will I have the gaps in the future that I need to close with these tools?

This text is based on excerpts from the podcast Industry Insights. You can find the episode with Timo Kistner everywhere podcasts are available or right here: https://youtu.be/8yHTEEdEyrs. The podcast is in German language.