Learning faster on an adapter shop floor
From the outside, the factory floor and production environments are not known for change. But that is not the case. And even small changes can have a large impact of production. That is why learning the new processes and getting up to speed is so critical. And we had Alastair Orchard on the podcast to talk through this topic.
Nick Finberg
00:09
Welcome to the Future Ready Podcast from Siemens. We have another great guest to talk about the transition to software-defined, Alastair Orchard. He’s the vice president of digital enterprise at Siemens as well as the CTO and co-founder of dimax.cloud. Thanks for being here, Alastair.
Alastair Orchard
00:23
Sure. Hi Nick. Pleasure to be here.
Nick Finberg
00:26
So, I think for a lot of people when they think Siemens, they think of the hardware we make and deliver. It might be the trains, the turbines, the medical equipment from our sister organizations, or the programmable logic controllers in production environments but it’s often hard to see the software, whether it’s the software aiding these hardware products or what we do with design manufacturing tools. Um this part of our business can be easy to miss if you’re not interacting with it on a daily basis. But some of that’s the point to get out of the way of the healthcare professionals, manufacturers, engineers, designers. But there’s an evolution taking place and we’ve been talking about it a little bit on the podcast, and it’s super important to producers, especially. Um software defined automation and the broader growth of software across industries. Alastair, would you mind painting a picture of what this shift looks like? Um for our audience.
Alastair Orchard
01:13
Okay, so I don’t know how old you are, Nick, but I was born in the 70s. I think I’m a bit older than you, right? I’m basically this uh a Gen Xer, so I was born into an analog world that you probably kind of never experienced. And then I grew up during the digital revolution. And now I you know I live together with you in this uh software design defined society. And there’s nothing particularly unique about that I think this probably describes a billion people. But in some ways, it provides me with um something of a different perspective. If I’d been born earlier, obviously I’d have missed this transition to digital completely. But I think younger generations and I was thinking of my kids, but this may actually apply to you as well. Um I think um I think uh They often find it difficult to grasp what actually living in a hardware-defined world was like, um and just how dramatic uh this transition’s uh been. And in fact, You call this transition a shift as though it was a sudden thing, and I think it’s exactly right. Um but I like to uh equate um that shift to um maybe a compressed evolution of life on Earth. It’s slightly strange, but um if you go back far enough then all the animals on the earth basically had kind of hard-coded behaviors. And that’s where the analogy uh comes in. Um they all they operated, let’s say on just inbuilt innate instincts. uh that were hard-coded, defined in their, you know, in their genes, the ones that they inherited from their from their parents. And they had brains obviously, just like our modern equipment has brains capable of kind of interpreting simple sensor information, reacting to input. But those actions were really I’d say predetermined. So it was, I guess, a predictable world. Species evolved to fill a niche. Animals uh kind of lived within that niche. Um and any abrupt change really, you know, in the conditions, it would cause actually an extinction. Okay. And then humans evolved and everything changed. So, we were and we still are kind of slower, weaker, and squishier than uh average animals. Um But as you know, our big advantage was our was our brains, right? So, in the place of pure mechanics, um emerges this operating system that could generate a world model. And um that world bundle had predictive power. Um allowed us to think this substrate was actually programmable as well. That’s the real advantage uh we could learn. So, if you fed that brain information then it could uh learn to use tools, communicate, plan, and um, you know, adapt in real time to any challenges we’re were presented.
Nick Finberg
04:10
Yeah, you didn’t need the generations to evolve that tool use. You could learn it from one of your fellow humans next to you sitting at the campfire.
Alastair Orchard
04:17
Exactly. Exactly. So it’s um the key here is adaption, but it’s also speed. So you’re absolutely right. And then we have memories of obviously So we can store and retrieve these software programs that we’ve learned and that we learn from other people. And those collaborations actually allow us to perpetuate this learning over kind of multiple generations. So s uh slightly strange analogy, but of course when we start to look at if we jump into industry, let’s say, then we have this industrial revolution. It created this Cambrian explosion of specialized machines. And that’s what we find very often filling our factories today. They evolved to operate in a dedicated niche And honestly, they’re all subject to extinction if uh conditions uh changed. Um and uh digitalization, this this transition It’s again helped transform that substrate and so we’re now able to kind of embed compute and communication capabilities and just about everything. What we have is a kind of industrial operating system. And uh we can use that to create these world models, um, just like our brains do. Uh we call them digital twins, and we can use software logic to program adaptable behavior at every rigid juncture and uh really start to transform all of that specialized hardware to become repurposable, software-defined, uh, and universal.
Nick Finberg
05:53
So you mentioned digital twin there. Usually I kind of think of that as a mirror held up to reality, kind of fed by IoT data, um useful for transparency capable of predicting maintenance conditions, right? Is that what you’re talking about?
Alastair Orchard
06:05
Uh sure. Uh that’s certainly one application. I think you’re that’s a limited view though. You’re kind of presuming that a digital twin is just a shadow. You said mirror that’s the same kind of thing of an existing asset, of a real asset. And that’s certainly that’s certainly true. And it’s certainly useful in brown fields. So where legacy or existing machines need optimizing. But you know, a digital twin’s at least as important when it’s also used to optimize a design before it’s built. Uh this is where I this is actually the Siemens special source. It’s this idea of shifting left and starting in the in the virtual world. So we spent really decades creating these um multiphysics environments that uh that really accurately model the dynamics of the real world. Okay? And this means our customers can design, uh, build and test really anything, uh an aeroplane, pair of sneakers, personalized medicine, a computer chip or a production machine. And they do that in this simulated reality so that they’re sure of what it’s gonna cost, whether it’s manufacturable, what the performance will be, what the maintenance intervals will be and everything before they ever actually commit resources in the real world. So it’s very powerful. And if they’re careful enough to make the behavior that they model adaptable, I mean this is the exact opportunity at design phase. then what they can do with that digital twin after is to actually um embed it in the machine itself or the product itself. It kind of acts as the code behind the product’s operational behavior. so that it can actually continuously adapt. And then of course it can also act as that digital mirror you mentioned in the you know in the question. Okay.
Nick Finberg
08:03
Okay, thanks for that. So when we were when we were talking prior to the recording, you framed this idea in a really interesting way around a self-driving car. How does that autonomy in a vehicle translate to manufacturing and a software-defined mentality?
Alastair Orchard
08:18
Uh yeah. Do you mind if I go back through that analogy or that um example? Um yeah, software defined and actually the whole concept of this virtual real duality. And this this is this is really a story of when companies first started to look at building self-driving cars. And actually they couldn’t really imagine how difficult it would be at the time. Um you’ve probably heard of Elon Musk talking about this. Getting the basics right is actually quite easy. Um but there’s this 80% very quickly and then there’s this long tail of edge cases that are really almost impossible to kind of to predict. And it actually takes literally billions of miles of driving on public roads. to gather all of this experience and that takes just far too long. So the answer is actually to step into the to the virtual world, to create, to generate myriad kind of uh virtual environments. Um and again these are multi-physics environments so um and then we fill them with literally um millions of virtual cars and then they play out um this interaction um you know complex traffic scenarios and everything in what’s really complete safety ‘Cause they don’t really know their virtual cars uh or that their journeys are kind of happening in this what’s really a video game. And the reason they don’t know they’re virtual is that we kind of equip them with uh digital versions of all of their sensors so they can see um the street ahead of them, they can see the obstacles, they can see uh other cars. Um and these sensors, although they’re virtual, they’re fully operational. Then we use like real-time ray tracing to ensure that the light sources um bounce off objects in in really natural ways. Um and this makes a big difference if there’s water on the street or if there’s low sunlight or whatever. This this creates reflections that actually do affect the sensors. And that would create chaos in the real world if they hadn’t um already encountered it and learned how to deal with it. And then we model friction, momentum, gravity, inertia. It’s an incredibly accurate world. And then we can We can turn on the rain, we can make it foggy, we can have kids run out from behind buses, and uh at the beginning you really get a lot of accidents. Um, but each one improves the model And then each of those learnings is actually replicated to all the other virtual cars. And so the swarm, let’s say, learns at a really incredible accelerated rate And the key to that, so that’s all in the virtual world. And then once the error rate, let’s say, once the incident rate drops bis beyond uh a threshold. then the model is actually compiled and sent into the brain, into the AI chip in in the real cars. And then it uses this executable digital twin. to drive and guide itself in in real streets. So we’ve transitioned from virtual to real. And again, we’ve done it faster and safer than it could ever be done in a purely physical world using physical experiments.
Nick Finberg
11:45
Okay, and then that links to what I thought digital twins were where you have all of the extra data happening during operation impacting the model of the design of the processes. Um can you talk a little bit about the like operation side?
Alastair Orchard
12:02
Yeah, so you raise a you raise a really uh important point there and that is those um those models uh are not static. So we talk about this life cycle of the digital twin and we’ve discussed a couple of these things. So we’ve discussed the you know, when the digital twin’s born in the virtual world, when it transitions into reality and starts um helping operate and guiding and optimizing the operations of a of a machine. But um I think you said you used the IoT word at some point, this Internet of Things. This is essentially the ability to pick up data signals from sensors, in this case in the car. and um feed the results back into the um either into the model so that it can react directly but also back into the training uh data so that these models can continuously evolve. Every accident that then occurs in reality is going to improve that and you see the see the error rates fall even further.
Nick Finberg
13:05
Okay. So um going back to manufacturing, kind of what we want to talk about today, is this pointing businesses at possible solutions on the production lines like slower unit times or vibration reading? indicating a match to previous maintenance needs or is it something bigger?
Alastair Orchard
13:22
Um oh it’s certainly that but it’s also something bigger. Um when we refer to software defined Uh you m yeah, you might think that this is uh kind of happening on a single level, so a single a single use case. But it’s actually it’s more powerful than that because it’s recursive. So in the same way that you might think of her hardware components being I don’t know, assembled or combined into subassemblies and then a physical system like a car. So software can and really should be used to operate and optimize these nested processes at many levels, if that makes sense. So maybe one level is that vibration of a component. But then that component is actually sitting in a machine. And so our digital twins, our software is helping optimize the parameters being fed into the into that machine. But it’s also orchestrating the journey of You know, the products being manufactured in that fact facility, uh orchestrating the journey of the product between the machines in the factory. And um if you think of that factory as one of a network of facilities then software is also optimizing the dynamics of the complete supply chain. And these things, they’re nested, they all interact, it’s modular, it becomes a big living dynamic system. And Really the point is that when you define something in software, whether it’s that component, the behavior of the machine, all the way up to an entire network. It can be infinitely adjusted and repurposed. You’re never painting yourself into a corner. Although I have to say it’s kind of ironic because we’re generally eagers engineers somehow, maybe the older ones. To make the box. To make the box, yeah, to permanently fix this into hardware so that you can automate it and forget it. But that is the old way of thinking. We obviously need hardware, right? Hardware is our interface to the to the real world. It’s what we use to transform raw materials into finished goods. But um let’s say the mindset shift here is that we need to ensure there’s as many degrees of freedom as possible left at every level. So whether it’s the component or the machine or the line or the factory or the supply chain remains adaptive and programmable
Nick Finberg
16:00
Well it’s really interesting that you’re talking about looping in more of a business into the digital twin, um, like the unified context for products and processes. How far does that does that thought extend? How far are we pushing that workload workflow?
Alastair Orchard
16:14
Well, yeah, it’s pretty persuasive. You know, we call it software-defined X for a reason. Um yeah. Um and it’s not actually unique to exactly manufacturing. So, you know, ERP companies, they’ve sold their platforms as these software-defined processes for years. Um you might have heard of some of these order to cash, hire to retire. These are these are processes. Um and uh we really do the same along the innovation access axis. That’s what that’s what Siemens uh really does. So, um Our business process, let’s say, is idea to cash. If that makes sense. And we can break it down further. So, you can go if you start with the idea. Then we we’re talking about a software-defined process that turns or transforms the idea into a requirement. uh turns that requirement then into a specification, a detailed description of what we want to achieve. uh then that specification that gets turned into a design. Along the way that design gets validated, tested in the virtual world and fully validated. We then turn that design into a plan and then the plan into a product and then the product into revenue. Okay? These are kind of milestones. along the life cycle of the digital twin. We talked about this life cycle before and it’s kind of form formalized into this into this software-defined business process. Sometimes we call that the digital thread that links uh what’s basically a market signal um to revenue on the other end. And the more the more hard-coded steps there are, the more manual interactions and the more disc disconnections that you find along the way, the longer and more costly that journey will be. And that’s why the That’s why it’s frustrating actually. That’s why the process of bringing an idea to market, it should really only take twenty-four hours for a fully defined um software-defined and executed um uh process. But today it often takes some of the biggest companies in the world 24 months or more. Um Yeah. And of course they’re using software. So obviously software isn’t a panacea here, it’s not the only factor. If you take that software and write inflexible code with it then that’s just about as bad as building it into flex inflexible hardware. So, it’s adaptability the whole way.
Nick Finberg
18:50
Yeah, if you hard code a number. If you hard code a number rather than using a variable, that’s just so bad.
Alastair Orchard
18:56
Yeah, that’s actually best practice. So, it’s not hard-coding numbers, it’s not hard-coding interactions, and it’s not hard coding uh behaviors. Yeah, good point.
Nick Finberg
19:05
I think that is a perfect point to end on for today. So, I still have so much to talk with you about, Alastair, but let’s save it for part two. I’m really curious to understand what it looks like for our customers from your perspective. And thanks to the audience for tuning in. We’ll be back soon with Alastair Orchard to talk about Software Defined a little bit more and its role in industry. Make sure to hit that subscribe button to get a notification for future episodes. And until then, check out the episode description to learn more.
