From Rule-Based to Goal-Based Automation – Transcript

Wrapping up the conversation on the transition happening in industrial automation right now, we brought Rainer Brehm of Siemens and David Humphrey of ARC Advisory Group back on the Future Ready Podcast. Starting off with how businesses are separating their expense budgets, we’ll dive into how automation is becoming goal-oriented rather than focussing on the rules that defined the automations.

Nick Finberg

00:09

It’s time for another episode of the Future Ready Podcast. I’m Nick Finberg, but the moderator for this episode will be my colleague Conor Peick. We are wrapping up this three-part interview with Rainer Brehm, the CTO of Siemens Digital Industries and the COO of our automation business; with David Humphrey, the director of research at ARC Advisory Group. In part two, we left off talking about the shift of business expenses from CapEx to OpEx as businesses implement more digital workflows for maintenance and operations. But let’s kick off this episode with how that change is shaping industry today.

Rainer Brehm

00:45

We see currently a big trend into centralized command centers. Which basically is now building on that. If you have a pop-up factory or different workloads you, can you control centrally, you don’t need to be there on site anymore.

So that command center thing comes. up currently quite heavily in different industries like in water wastewater semiconductor but also in data centers so that’s an interesting aspect and the other aspect is I it’s a shift to OPEX. The topic of cybersecurity is very relevant.

So, you will update the systems and so basically you buy it then also as a subscription and then everything is inside, yeah. The updates and everything are inside. But also reality is a big change in our customers as well.

Because still customers are not planning in this way. In the IT world it’s completely normal. If you go to the CIO that’s normal.

You buy maybe software as a service or you buy a subscription, so you have a quarterly or yearly fees, and that’s normal. If you go in the OT world and say, hey, I want to have a quarterly or a yearly fee, they say, yeah, but I have a CapEx budget. And my OPEX budget is low and I it’s not it’s easy hard in my company to shift now from CAPEX to OPEX because that’s not planned.

So how can we do it that we know we put it in the CapEx budget because I have that one? So, I think also there’s a learning now in the companies as you upgrade plans or build maybe new plans. What is a ratio between CapEx and OPEX?

Conor Peick

02:16

So, it’s part of that partnership, right? Is that it’s not just us developing new technology, you also have to have the customer understand maybe what might be involved in shifting to adopt that new technology as well.

Rainer Brehm

02:28

Absolutely. And again, you know, I think the operations world need to learn also from the IT world, but also the IT world as I said need to learn what it means to run something twenty-four-seven with a very high reliability. I think we need to learn from both sides.

Conor Peick

02:41

Yeah. And so, then David, as we continue to think about the business impacts of this shift, in your research are you already seeing data points or benchmarks showing productivity gains and efficiency improvements or return on investment for these changes that are being made.

David Humphrey

02:56

There’s some data out there already. For example, when I think of the engineering side alone, there are figures of twenty-four to forty percent reduction in engineering time due to the fact that my co-pilot is generating some code for me. I’m kind of funny about everything being generated. I just don’t want my have my hand in it.

But It’s it might be searching libraries for me to find a function that I programmed three years ago but I kind of forgot about, but it does exactly what I want, what I’m asking for right now. So, it calls it up again and says, this is what you this was your work three years ago, or somebody else did it so why don’t you just let me plug this into your program? So, engineering right off the bat, good productivity gains.

Ten to thirty percent is another figure that’s floating around for time to ramp up for new lines or changes to new products on existing lines. And in a lot of end users like to measure the effectiveness of their equipment in terms of OEE, overall equipment effectiveness. That’s a combination of the what’s the availability of equipment?

It’s the combination of the of the uptime plus how much the machine is actually being used when it’s available. And we see about five to ten percent improvements in OEE. And it’s such a common measure.

Every visualization software has a tool for OEE built into it, so it makes it really comparable. You can compare it to other machines, to other plants, to peers, even to competitors.

Conor Peick

04:16

And I mean thinking about those business impacts that you just laid out for us, how does that influence competitive advantage in manufacturing for the future

David Humphrey

04:24

Manufacturers that switch to software-defined architecture, software-defined automation, that implement AI-enhanced tools are going to gain competitive advantages because of the software centric mentality because of the mindset that they’re going to start pulling ahead in the race, in a number of fronts. They’re going to gain knowledge and learnings that they apply and that they continue to learn from that manufacturers with a hardware-centric mindset and control system cannot replicate.

Conor Peick

04:55

Moving on maybe to talk about strategy and positioning a little bit. We’ve talked about regions or maybe different industries that are leading this adoption, but David, what do you think in terms of a region that is best positioned in the in this transformation? You mentioned Western Europe obviously earlier, but my short answer is Germany, US, China.

David Humphrey

05:13

So, Germany because this is where things are invented. Europe in general is the I’ll call it the intellectual and architectural leader in in SDA. Even if adoption later on becomes moderate rather than fast like we might see in China and the US. The US because of its strong relation to software and its greater flexibility in adopting new things and then then China as the that landscape of greenfields. There are not as many greenfields in China as there used to be. I wanna I want to keep it things down to earth here, but the Chinese are very, very efficient and very flexible in terms of terms of their experimentation and the kick in the butt they get from the state.

Conor Peick

05:55

Do you think Siemens is particularly well suited to driving SDA? What benefits do you think customers can gain from partnering with Siemens as they, you know undertake this this adoption.

David Humphrey

06:05

All suppliers are going to go in this direction. All suppliers are working to some degree on their solution portfolios for SDA. Siemens has a very complete portfolio.

I don’t want to do a commercial for Siemens, but I like what you’re doing, Rainer. Let me just let me just put it that way. I think it’s pretty impressive.

What I like most about what Siemens is doing is your messaging. You’re making it really simple for us to look into the future. I’ll give two examples.

One is the simple equation, SDA, plus AI equals autonomous operations. And the other is what you presented at the SPS show in Nuremberg, I forget the terms you use, but it was the transition from the fixed architectures the rule-based to goal-based Rule-based to goal automation. That’s what it is what I find really compelling. It’s not saying come buy our products, it’s saying come change your mentality. Let us help you by supporting you on that journey.

And then as we’re talking about some of the advantages that companies can gain or the benefits they can gain from taking on this transformation, Rainer, of course there’s also some amount of maybe homework that that companies might need to do. Many companies struggle to scale due to fragmented data that they have. Is building a data fabric the real challenge behind making industrial AI and SDA work? How is Siemens helping to tackle that challenge?

Rainer Brehm

07:25

Well, you always talk about data silos but take a step back. I was personally surprised that in the engineering workflow and we show this Eigen, you simply use it. So, you don’t need to prepare any data whatever.

So, you use it and you get the productivity. And maybe that itself is one process step only. You do the automation engineering, which is an important one, yeah.

But you don’t need to kind of now break up silos. You take it, you install it and you use it and it creates great results. So, it’s not like get your data straight before you haven’t done this, you cannot use it.

So, use it, yeah, and I think getting data straight is then very important. And maybe another topic, we see in more and more agent coming up which are basically can connect to different data sources and then you know use that data. So, you know don’t need to pull it all up and put it into a big data lake and then work on it.

I think so you can you need to untap those silos, but it doesn’t mean you need to put everything in a big silo. Yeah. So, I think that’s also, especially with agents which are very flexible to connect you to different topics, will help here overcome that hurdle but then the other topic is then around the operations phase.

So not talking about engineering, but now you’re running a factory and you want to do the OEE or you want to understand the root cause, you know, if you have a big line and something happened, uh, you know may maybe a day earlier in the in the process and now you have a defect here, how you bring things together here. I mean that’s a different topic, and this is really where we need to connect the workflows and not from an engineering side but really from operation side.

And you need to get then also some knowledge in, yeah, how things is connected. And there we supporting the customers to building a kind of unified data fabric. That’s what we call it.

We are building a semantic layer because you need to understand what is the though what is the asset, what are the values in there and we are using here a product which is Graph Studio from Rapid Miner, which is building this ontology, building this knowledge graph. Which is then a great basis for AI to access the data in a structured way and go to reasoning what’s going on, what are the relationships between and where maybe the root cause is.

Conor Peick

09:33

And I love that you brought us to that topic of agentic agents helping to connect some of the data that and that doesn’t mean that you have to have one gigantic silo, like you said. These can help us connect data from various silos.

Rainer Brehm

09:44

And by the way, it was also important because I know companies say I pull out the data from everywhere and then I have a silo and then I can do some reasoning. The topic is if you pull out data somewhere, they are already old. The minute you take it out, they might have changed the second afterwards. So, you want to get to the real sources where the data are processed and changed and not pull it out once and then doing some AI on top of it

Conor Peick

10:10

That real time up to date that’s so critical in a manufacturing environment. So then as we think about, you’re looking ahead, what role do you think agentic and physical AI as well will play in autonomous production in connection with software-defined automation?

Rainer Brehm

10:23

Well, David said it first. I think SDA is really the basis. If you look now in the operations phase to bring AI on the shop floor we do today with AI inference servers, I strongly believe in this what we are building as we speak, that when we talk about physical AI, it brings AI intelligence into the control layer. Yeah, currently as I said, when I when I want to go from a rule-based to a goal-based automation, automation itself needs to get much more intelligent.

So, because the rule is I’m writing a program. So, if we are rule-based automation, I’m writing an automation program. If this happens, you do this If I do this, it’s rule-based, but it’s fixed because if something happens which I haven’t considered, the system will not work anymore.

If the system needs to decide on a situation which has been not pre-thought, it needs to have a built-in intelligence. And now we talk about AI. So how of can we combine classical PLC workload?

Because you still need that. A lot of deterministic processes ongoing which write a PLC program, and you can maintain it. But can we add now this kind of intelligence into that control logic and that you can describe it as skills or add a skill which is AI-based and that skill could be AI-based or could be also a visual language action model. And that vision language agent model could then maybe control a robot. So, for me, SDA plus AI plus maybe a special kind of AI, which are VLAs.

Or robot foundation model will add together to go into the future of physical AI, which is the basis to move from rule-based to goal-based automation. And for me, that could be a big inflection point for SDA. Because every customer probably will understand Now if you want to go in this direction, this cannot be done with a classical PLC.

You can always do something, yeah. But it’s much more easier to integrate all these different workloads. into one platform sharing one data.

You need to connect them by the way, real time, because it doesn’t make sense if you if you have a controller robot, it doesn’t fit to the line. So, this needs to be synchronized and real time. And there SDA is the perfect platform.

And it could be the inflection point really say now, okay, if I want to get these benefits, I’m gonna shift over there.

Conor Peick

12:36

And what do you think are the role of humans in that environment of increasingly autonomous operations as well?

Rainer Brehm

12:42

Working in the automation space is always about productivity. Uh so and we always embrace kind of neutral technologies how we get to the next productivity level. I think it’s a proven thing.

Companies which are working in productivity are growing and surviving and that we need to work always on productivity. You know, there was if when I mean I started at Siemens at ninety-nine, the automation guys were kind of you are destroying jobs I think currently automation is one of the main enablers to, you know, produce close to the consumption, to produce more sustainable, to produce more energy efficient. And still there are still people in the factories.

Yeah, I don’t know how many people want to work in the factories, but I think it’s a big enabler for the main challenges we have as a society. And therefore, humans will play a major role. They will play a different role, but that’s normal.

It’s always evolving, yeah. And therefore, I’m confident that a factory will have humans. And you might run a shift unattended.

And maybe that’s a night shift where anybody no nobody wants to work, but you still have people you know operating a factory.

Conor Peick

13:47

Yeah, interesting. It’s just because a factory maybe can work operate sometimes as a dark factory, right? But doesn’t have to all the time.

Rainer Brehm

13:56

I don’t know how far men robots will go, but I think there’s always the maintenance, you know. There are things you need human. For me, may maybe there’s also a topic where we want to move into AI powered and autonomous production.

Does it mean it’s a hundred percent a dark factory? Probably not. I wanna have a flexible factory, adaptive factory, uh, which will act more and more autonomously.

The good thing is every step we do there is creating a customer value. And that’s what I say it’s different than autonomous driving. Because an autonomous driving is a step function.

You know, either you can drive autonomous and you can you know take your hands off the steering wheel. You don’t might have a steering wheel at all anymore. Yeah.

All steps in between where you have some assistant functions, yeah, with a cruise control which is adapting, it’s a comfortable feature, but it’s not kind of the big thing, it’s autonomous or not autonomous. In factory I see it differently because more and more we can automate putting intelligence into the automation system the more value we are creating. So, it’s more a kind of a linear function rather a step function like autonomous driving So I believe physically AI in the industry is continuous delivering customer value compared maybe to a self-driving car.

Conor Peick

15:13

Yeah, fascinating. So, and then David from an analyst perspective, how do you think the SDA and industrial AI market are going to evolve over the next five to ten years maybe taking into account this linear progression model that Rainer’s outlined for us.

David Humphrey

15:27

I would split it into five years and five years. I would say in in the next five years we’re or the first five years we’re going to see selective deployment. We’re going to see a growing knowledge base.

We’re going to learn more and more about what this stuff can do. And we’re going to see lots and lots of proofs of concept. That’s what manufacturers really need.

They need to see it work on a test bed for a long time. They need to see it to test it in production in parallel with the existing system and then after a long time they can they can release it to general use in the factory. In the second half of that decade, so you know, starting in about five years from now, I would say that whole architectures are going to consolidate around SDA.

It’s going it’s going to be “We have no revolutions in industry. We have evolutions only,” but it’s going to evolve into architectures that are software defined that are IT-like, a favorite word of mine, saying that the IT OT convergence is going to come to an end. We’re going to have perfect convergence.

And then we’re going to move on merrily along to the visions that you just painted before that I that I liked very much about autonomy, but also about humans still playing a role in making important decisions, making complex decisions that AI today can’t always solve. But I changed my mind on that two days ago when I saw I saw your laundry sorting application.

Here at the fair there’s an application where two robots sort some laundry. There’s a shoe, there’s a t-shirt, there’s a cap, and they put it into sort of a floppy basket. And the application or the demonstration is very simple.

The laundry is just thrown down, the basket is put somewhere, and these two robots very gently pick everything up. My favorite part is the t-shirt. They take the t-shirt and it sort of falls halfway into the basket and then it goes back and it picks up the back half of it and then sometimes it makes a third motion to put the tail of it.

Um I see so much of the future in that very simple demonstration that it really opened my mind to this idea of coming away from fixed stepwise functions of going from one step to the other to this adapting to the to the current situation, solving problems as they present themselves in front of you.

Rainer Brehm

17:43

Yeah, absolutely and that is basically goal based automation because it gives a goal, you know, put all this textile into the in the bag. Or clean it up. Maybe somebody said, you know, the goal is clean up and then the automation system is cleaning it up.

And that’s exactly what we see by goal-based automation. And here we bring physically AI in the automation area in order to fulfill that task. To add on, you know, I think maybe on the on the topic of a future outlook, you know, I think architectures will be more software-defined, absolutely, but also the engineering will be more IT-like as you said the word. Because you know where classical automation comes for it more comes from a perspective of an electrician.

So, ladder logic is basically the programming language or kind of how an electrician thinks. I think in the future you have more kind of IT minded automation engineers. So, you also gonna change the way how you program your automation, whether it’s with agents or not, but it’s gonna be more IT like how you do a how you connect repositories, how you are having a test pipeline, how you do a deployment.

All that kind of workflows will be much more like you know it today from IT rather than you know you know it from what electric electricity nutrition is doing.

Conor Peick

18:59

Excellent. And so, as we’re coming up to the end of our time here today, guys, first of all, thank you so much again for joining. So, what do what do you think is one key opportunity companies should focus on today to fully unlock the value of industrial AI and software-defined automation as well.

David Humphrey

19:13

SDA and AI should be treated by industrial users as a strategic asset. In other words, not just technology you can implement at any level that can be chosen. by project managers to maybe improve a process.

It should be considered enterprise wide as a as a valuable tool, as a collection of the learnings that we’ve experienced over the course of our journey. from the very beginning stages where we looked at the technology, decided to do some tests to the point where we decided to actually implement it and to the point at the end of the journey where it’s up and running and it’s producing measurable positive outcomes to manufacturing.

Rainer Brehm

19:55

Maybe I do I start with a promotion. Use the Eigen Engineering Agent because it’s there, it runs, it’s proven, it delivers return of invest. So don’t wait. Everybody that has a TIA portal should look at that opportunity absolutely and then again start where you have the biggest return and biggest pain point and then building from there this new architecture

Nick Finberg

20:17

A big thanks to Rainer and David for making the time to have these discussions and thanks Connor for making the trip to Hanover to record them. This has been such an interesting progression from early adopters of AI tools to scaling digital workflows across a business. If you missed them when they first came out, you might want to check out our other episodes on topics like AI, semiconductors, pharmaceuticals. or even our deep dives on automation and operations with Rainer Brehm and our other co-host, Mark Hindsbo. We hope to have you back again soon.