AI-powered simulation: a new era for industrial engineering podcast transcript

Chris Pennington: Hi everyone and welcome back to the Siemens Digital Transformation podcast series. I’m your host, Chris Pennington, global industry marketing leader for industrial machinery at Siemens Digital Industries Software. In these sessions, we’ll be looking at the impact of AI in industrial machinery and considering the implementation barriers along with looking at the future.

As always, I’m here with my colleague, Rahul Garg, vice president for industrial machinery vertical software strategy. I’d like to welcome today’s guest, Dirk Hartmann, head of Simcenter Technology Innovation. Dirk, could you start by giving us a brief introduction explaining your role at Siemens?

Dirk Hartmann: First of all, thanks a lot for having me on this fantastic podcast. My personal role is really twofold. On the one hand, I’m leading the Simcenter Technology Innovation Team as you highlighted. On the other hand, I’m a technical expert myself in the field really of simulation technologies in ML and actually also even a professor in that very specific field at the University of Darmstadt.

Let me focus in the brief introduction really on the role of our Simcenter Technology Innovation Team. Our objective is to translate technology trends into business expansion opportunities. The way how we do this is by what we call demonstrating the art of the possible. In the end, taking those technology trends, explaining them simply by demonstrating them by benchmarking, desirability and feasibility really along very concrete proof of concept and prototypes. And through this work together with the organization on future product strategies.

Chris Pennington: That’s great. Thanks, Dirk. Now, I’m sure many of our listeners are familiar with Simcenter, but maybe for those that are unfamiliar, could you just spend a few minutes explaining what Siemens Simcenter is and maybe who typically benefits from using it?

Dirk Hartmann: A tough job, but I’ll try our best. So I’m in NX. I hope more of you know, allows you to design in the virtual world. And when you then would like to test those designs, that’s when the Simcenter portfolio comes into play. If you want to know how a certain design behaves under a specific load, what is it? Is it your ability? What is the expected energy consumption?

Then you typically would use Simcenter to investigate these questions either using virtual testing or that makes us very specific with our real test hardware. Do that in terms of real testing and all within the same family of solutions. Now I’ve only spoken about virtual testing simulation, but of course, once you can do that in the virtual day world, then you could put that in an optimization loop and simply let the computer do the design and the testing and then really iteratively find out an optimal design, you can then start to build your actual designs on top.

So whose benefit thing in the end from that? I mean, the usual persona is computer aided engineers validating, verifying performance of the systems, optimizing performance of the system. But then if you look more at the future trends, it’s really to democratize that whole suite of solutions, bringing it closer to the cat engineer so that actually also the designers themselves can benefit from this so that out of an axe, you can simply run a CFD or a mechanical simulation to very early upfront validate your designs, which is mostly known under shift left.

Chris Pennington: You have looking at what technology we’re going to be deploying next and where the product is going to go. So if you were to just tick a broad view, what would you say are the technologies that are really at the forefront when it comes to innovation in Simcenter?

Dirk Hartmann: So obviously, I mean, we continuously do to innovate in our classical space. So also the solvers, the very specific tools are still innovating themselves. And actually, that’s quite unknown. Yes, whenever you speak about performance increase, everybody only has more loss somehow in the back of their minds.

But the algorithmic innovation there in the tools has actually a much deeper growth. Probably using a new algorithm on old computer is faster than an old algorithm on new computer. So that’s critical to what we do. Accelerating always the time to prediction. But the really two crucial technologies which would stand out here, I would say, is AI and machine learning on the one hand side and then all the new technologies around the industrial metaverse on the other side.

And that’s really kind of what we focus on, innovate in our classical domains and then see how we can leap rock some of the developments using AI specifically.

Rahul Garg: Yeah, I was just going to say, Dirk, that given your focus on AI and simulation, one of the challenges we have as an industry of industrial machinery companies is the adoption of simulation is not as prevalent, I would say, as compared to some of the other industries. A lot of it being not having the skill labor you would need to leverage simulation capabilities. And the effort involved in trying to take advantage of simulation versus the returns, right? Because a lot of the times we are dealing with equipment and machines that are perhaps sitting on the shop floor.

And if you’re trying to do a stress analysis, simple, just improve the quality of the steel you’re using, use a higher-grade steel and you’re done with it, right? So you don’t have to worry too much about trying to look at the stress or the durability of that equipment because you can just improve it by using higher quality, more weight equipment and trying to lightweight the equipment is not as big a challenge. The equipment is not moving. It’s bolted down on the floor.

One of the things that comes to my mind is as I’m hearing you talk about innovation, especially with AI and metaverse, is that an area you are exploring on how to use AI to drive the adoption and simplify the adoption and usage of simulation technologies?

Dirk Hartmann: Yeah, actually that’s a very clear focus of what we do in the team. And that actually not only has to do with enabling less skilled people to use the technologies, but at the same time also, the same set of AI functionalities can be used to make even the fluent users much more efficient allowing them faster to do their work. Whatever we do there on the AI front really benefits both fields.

And that’s why it’s really close, close to our heart because in the end, what we would like to do is getting the time to prediction significantly reduced. And on the novice non-expert type of user persona you have spoken about, I mean, the first step we did now a couple of years ago is realizing a CAD integrated stimulation tool, the NX performance predictor there.

But obviously you still have to configure that appropriately as it might run in your CAD tool, but you need to know what are the right boundary conditions, what is the right material parameters, et cetera. And their generative AI is an enormous lever who could help you to automatically configure those tools automatically suggest at least which boundary conditions to place with there and then to allow also a non-expert user of less knowledge to use these tools.

And indeed, at the moment it’s not as much used as possible, yes, because you could simply go with more steel, but then of course more steel means more cost. Looking a bit at becoming or many companies trying to reduce their CO2 footprints, yes, that means the higher CO2 footprint, et cetera.

And you don’t really want to play around there a lot, yes, because this very often means downtime goes to performance critical things, et cetera. And that’s really where simulation can bring significantly value to squeeze out efficiency there than just using the simple solution, go with more steel.

Chris Pennington: Rahul, I’d like to throw that question back to you as well. Just one point of view of your customer interaction and have you seen any good examples where AI is helping to simplify and extend the use of simulation tools?

Rahul Garg: Well, frankly speaking, Chris, not yet. We have seen a couple of examples, at least one of them is more in the solution call as the executable digital twin, where the customers have now looked at how to create a digital twin, that is what we call as executable, so it’s actually live and it’s running while the equipment is running or being used while the machine is working and then comparing the performance of the machine against what was designed during the initial design phase to see if there are any anomalies and then check for those anomalies and then kind of take any corrective action.

One of our customers actually is using that, for example, to see the impact of some coolants being applied when you are trying to do some milling operation and trying to look at the number of the chips that are flying out during the milling operation and seeing if any of them are causing any blockage during the cooling chambers to make sure that all of that flows through very easily.

So they are using the executable digital twin to highlight, hey, something is not working right and then see if there is a way to further improve that whole process. And that’s in the context of AI as well, because they are trying to use vision lenses to kind of see where those chips are falling and then trying to use AI to say, okay, hey, this is where there’s an accumulation of those chips and we need to figure out how to get them out of here during that whole process.

That’s one example I’ve seen in the context of AI and simulation. Of course, there are many more examples of AI in the industry as we speak around quality and whatnot. But in the context of AI and simulation, at least the adoption still has a lot of room for potential growth is what I would say.

Chris Pennington: It’s interesting that you mentioned the linking of the real and the digital world to get data from the real world into that executable twin. I think, Dirk, I’m very interested from your point of view where you’re working more in that theoretical upfront world. We need to train these AI models. Where do you get data from? Where do you get enough data that you can effectively train an industrial model for simulation? So it depends a bit on the various use cases where you deploy in the end, those models. Is it something more upfront you would like to use in the design, accelerating your design phase, doing more efficient design exploration?

Dirk Hartmann: Yes, that is then when you can build your AI models purely on synthetic data. Ideally, synthetic data you start to prepare upfront very specifically for this case. It has to save your simulation results, curate them so that you could reuse them. That’s a bit the design process or the design world.

And then speaking a bit about the digital twin, the executable twin in operations. The biggest challenge is that whatever data we get from our machines is good data. Obviously, we don’t run our machines into failure situations where anything could go wrong because downtime is super costly in most cases. That means, basically, if looking at the real data from the sensors, typically we can only say, hey, look, something is going wrong here, shut down the machine immediately. Which then again could lead to very long downtime.

Yes, because we could use simulation to appropriately prepare synthetic data, which then we can enrich those real time AI models, doing performance optimization, doing health monitoring or whatever, and take much, much more informed decisions. Yes, not to shut down immediately the machine, but taking from the simulation data collected upfront a bit saying, OK, this will still run for a week. Let’s continue to run it for a week. And within that week, we’ll have the spare parts on time. So we will reduce more or less the downtime.

So that’s a bit the way how I see it as bring in more simulation data, reuse that knowledge created during the design phase also to leverage that for the operation. And that is actually really kind of what is. Yeah, from my point of view, the core vision, what is behind the digital twin. Don’t forget the engineering knowledge once you go to the operations world.

Rahul Garg: I think that’s a very important point because it’s taking advantage of the engineering insights that have been designed into the equipment and bringing that to the operator to ensure that any issues that may come up are addressed in a more rapid manner as well, right?

And that’s where AI can really help bridge a lot of that gap as well. I’m just wondering on that same point, trying to take the insights during the operation phase and bringing that back to the engineer as well, right?