How AI brings DevOps to product design – Transcript
In a recent podcast episode, host Spencer Acain is joined by guests Boris Scharinger, Senior Innovation Manager and Technology Strategist for Siemens Digital Industries, and Dr. Justin Hodges, Senior AI/ML Technical Specialist in Product Management for Siemens Digital Industries Software, to explore the applications of AI in simulation and design, including applying AI to unify different elements of the digital twin and deepening the connection between the virtual and physical worlds.
Check out the full episode here or read the transcript here.
Spencer Acain:
Hello and welcome to the AI Spectrum Podcast. I’m your host, Spencer Acain. In this series, we talked to a wide range of experts from all across Siemens about AI and how it’s applied to various technologies. Today I am once again joined by Dr. Justin Hodges and Boris Scharinger as they continue their discussion on the impact of artificial intelligence across the full breadth of the product, design and manufacturing process.
Last time Justin ended on an interesting idea that the market would be kind of gearing toward wanting smaller models in order to accommodate the need for less training data while still receiving a good result that they can actually use. What do you think of that, Boris, as a place to start our discussion?
Boris Scharinger:
Yeah, absolutely, absolutely. And then there’s so much more to simulate. If we move on a bit to the most interesting topic of the industrial metaverse. Some of you listeners know that Siemens really drives the creation and the roadmap of the industrial metaverse. And basically, there are two sides to it. One is the side of immersion and 3D glasses and whatnot. Okay, let’s park that for a second. The other fundamental concept behind the industrial metaverse, it’s an aggregation platform for thousands of digital twins.
So today what you do is you have a digital twin for a component that could be a digital twin for a turbine, could be a digital twin for an electric drive. But what we really want to simulate, we want to simulate systems. We want to simulate a full machine. We want to simulate a full airplane. We want to… Production wise, we want to simulate production lines, even full whole plants or maybe even a supply chain network. And we only can do this if we can kind of aggregate all the component digital twins into one platform that allows us to do a holistic, sorry, holistic simulation. And because we have so much more to simulate, if we go to the system level, we need speed. And AI is really instrumental in helping us to accelerate simulations to a degree that these types of simulations become computable at all. So that’s why I really welcome the trends that Justin was just talking about.
Dr. Justin Hodges:
Yeah, it’s like a on-demand model store or something, right? Or instead of models, AI based reduced order models. How did you phrase it? You said you park one of the aspects of the-
Boris Scharinger:
Well, exactly was the immersion aspect. Right?
Dr. Justin Hodges:
But the stuff that I really liked was an aggregate of many different individual digital twins.
Boris Scharinger:
Yeah. It’s an aggregation platform, right? It’s an aggregation platform for digital twins so we can simulate on a system level and the system can be as complex as you want it to be and as our compute capabilities allow it to be. And let me just add one interesting topic from the kind of later phases, production, production engineering and production operations. Today if you really look into the market of design space exploration and how that adoption of design space exploration is, so design space exploration, I take a product and I have an algorithm creating 1,000 variants of that product. And then a simulation identify me the best top five of these thousand variants. So that’s design space exploration. But today it’s a sophisticated art form and maybe we see 2% or 3% of companies, of the larger companies having this set up as a discipline design space exploration, particularly just in your home turf, to machinery and aerospace, for example.
However, none of them is totally simulating how things can be produced today. So design space exploration in 95% of all cases stops at product features and doesn’t explore the 50 different variants of a product with regards to, or the 1,000 different variants of a product with regards to how efficient can I produce it? Can I produce it without changing my production line? Can I produce it only with small changes to my production line? Would I be facing a higher waste and lower yield when I start the production of my new product? Or when I choose a different product variant, can I already by simulation, determine that a certain product variant is causing less waste, thus higher yield? So these types of… This field is totally open yet, and I think there’s so much potential that once we manage to bridge the disciplines of product design, product engineering, production design, production engineering, and finally operations, and we cover all of that by simulation, accelerated by AI. What’s your view on that?
Dr. Justin Hodges:
Yeah, I think somebody knew what they were doing when they matched us up to talk because we’re speaking to it at two different very meaningful levels and in alignment. But I just think about it from hands-on simulation point of view. I mean, this is taking how it is now and making it further refined until, like you said, it’s an aggregate of models that you can basically use and get results for on demand. And it makes me think of, well, it’s really kind of painful normally, to have one group for a turbo machinery organization, for example, one group do all of their work within CFD for aerodynamics, and then another group do the structural aspect and then another group do the system aspect. I mean, it is very different set of knowledge, and tools and all through workflows, everything. And basically you consider how do I integrate those teams centrally so that the information is available and kind of passed from team to team very fluidly?
And rather than metaverse thinking of some, like you said, parking the 3D goggles sort of idea. How I think about it is it’s taking these teams and letting them work better together and passing this information via consolidated information in models like ROMs that are AI-based or AI models. And then it’s making this information available to everybody. And I think that is a very, very relevant headache that could be solved from the different teams and making the overall product in a multidisciplinary way much, much faster through the teams having this information pass freely back and forth. So to me, that’s very attractive. That’s how I think about it.
Boris Scharinger:
Well, if you take that idea and accelerate it or think it to the extreme, you end up in a DevOps concept for hardware, isn’t it? I mean, the DevOps concepts in software development have exactly managed to integrate very different teams, very separate teams, even very separate cultures like the stability-focused IT admins running the systems on the right side and on the left side, the kind of change-oriented software developers adjusting their code according to the latest and greatest in algorithms. And then DevOps has built that bridge. And that bridge is a data-driven pipeline modeled process. And it’s a decade ago, but a decade ago I looked it up, Netflix was deploying 17,000 IT changes a month and it is actually a non-event to deploy such a change. And now if we do one step back and look at product development, solution development, including hardware, but then test concepts for hardware, which actually translates into different types of simulation. So you can do that in the virtual space, you can virtualize your testing, and then if you really spend this as a pipeline across your different development teams, you can have changes going through that chain of command very, very quickly.
So one week after the product design engineer has performed a change, the production people could give feedback and say, “Is this producible?” And you can only do this by creating a chain of simulation, a chain of virtual commissioning, of PLC code adjustments, and then checking that whether that works and whatnot. But it is as a vision, something that is very appealing to me from a agility and also from a resource consumption perspective.
Dr. Justin Hodges:
DevOps for hardware. That is such a good analogy. That is sticking with me.
Boris Scharinger:
Maybe, right? Now-
Dr. Justin Hodges:
It’s a cool thought. Yeah.
Boris Scharinger:
It’s great. I mean, we have this conversation and actually, believe it or not, there is no THC involved here having all of those great visions. Could we maybe turn the conversation to where are companies today in terms of the use of simulation? Project-based versus pipelines and then the use of AI? And where do you see Justin the most difficult challenges on a company’s journey applying AI, applying simulation, working towards that vision that we just articulated?
Dr. Justin Hodges:
Yeah. Well, part of this is cross-pollination, and that takes time when some new type of neural network is invented for within the Googles and the Metas of society, it takes time before a flavor of that is adapted to scientific applications in engineering. And then from there, that cross-pollination is done. And it takes time for people to research and build trust that those types of machine learning models are accurate, reliable, and able to be understood. Because if you’re designing an aircraft, I mean, you have an immense responsibility on your hands to make sure that you do it in a conservative way. So naturally, when these leaders in AI create such new models and advanced architectures, then it does take some time to really be shown that it can be used in a controllable, safe way. And then we realize all of these benefits, and I would really say in the last year and a half, there’s been major things happening in the market and in industries that are proofs of investment and people trusting and realizing that they can trust these sort of approaches and using them in the ways that we’re saying.
We’re not replacing simulation, we are using it to augment and to help in the early design phase. So to me, the theoretical proposal, I saw that in 2017 when all of the other people that were interning there, we would get coffee or hang out after work. “And so what did you do today?” “Oh, I replaced all of these numerical models for modeling hearts with AI models.” Like, wow, that’s quite a proposal and research. But then now you’re seeing that this is maturing and now it’s actually being used in industry settings and trusted in industry problems. And it does take some time to vet, I think is the answer.
Spencer Acain:
That’s a very interesting perspective, Justin. Thanks for sharing. As with anything, of course, AI as a technology isn’t going to find its place in manufacturing and everything overnight, and I think it’s exciting to see that we taking the first steps there with bringing AI and machine learning into the early stages of the product design process. But I think that is going to be a great place to end this episode since we are just about out of time. So once again, I have been your host, Spencer Acain on the AI Spectrum Podcast, joined by Dr. Justin Hodges, and Boris Scharinger. Tune in again next time as you continue exploring the exciting world of AI.
Siemens Digital Industries Software helps organizations of all sizes digitally transform using software, hardware and services from the Siemens Xcelerator business platform. Siemens’ software and the comprehensive digital twin enable companies to optimize their design, engineering and manufacturing processes to turn today’s ideas into the sustainable products of the future. From chips to entire systems, from product to process, across all industries. Siemens Digital Industries Software – Accelerating transformation.
