Episode 8: David Ariens

Welcome to the Factory Ops Exchange podcast hosted by the US center for Advanced Manufacturing and by Guide Wheel. I'm Lauren Dunford, CEO of Guide Wheel and executive committee member for the US center for Advanced Manufacturing. Today we're joined by David Arians. He's a former educator, founder of the IT OT Insider, and former industrial digitalization leader at basf. His work sits squarely at the intersection of theory and reality on the factory floor. In this episode we dig into the reality of what it takes to connect systems, teams and data so technology actually works for the people on the factory floor. Let's dive in. David. So great to get to have a conversation today. I cannot wait to find out exactly where it takes us. You have such a fascinating background, really experiencing the IT OT challenge and the reality and also now learning about everything that's happening in the industry and directly influencing it. And I just have loved. Yeah. Getting to benefit from the framework frameworks you've created and I'm excited for the conversation.

Yeah, cool. Likewise. This is going to be really interesting, so just fire away.

Well, maybe I've been listening to some of your recent podcasts and I think the tension between theory and reality keeps coming up as a very interesting topic. Where are you on that right now? What are the things you're wrestling with when it comes to reality versus theory?

That's actually a really good question. I always lived in this, in this blurry zone between IT and ot. Sometimes I identify myself as being more IT alike and sometimes it's being more OT alike. Like and I think dealing with all the buzzwords around us, it's basically a daily thing. And the cool thing of being working in this intersection is that I guess that's really where you get hit the most by all those buzzwords. So a couple of years ago it definitely was blockchain and I don't know, something augmented reality, the things of a couple of years ago. Right.

Yes.

And then, and then, and then that changed with ChatGPT. So all of a sudden it's all, it's all gen AI. I think at this point, obviously we're getting hit now in our face by all things agentic AI. That's one thing where we've been doing some work now to clarify that. A bit more that domain, I think. Another one when we talk about data platforms, data systems, they think something like a unified namespace. That's probably one of the hotterms right now.

Yes. And maybe this is just me putting it very bluntly, but have you seen any manufacturers that are approaching this in A way where you say, wow, that's gold standard. So hard.

I wish I could say yes. Okay, let's back off. So if we talk about becoming a data driven manufacturer, obviously there are companies who have been investing in it for years and years. Something like a unified namespace is also nothing new. But I must say that when I was, the last weeks and months when I was trying to make sense of manufacturing AI, um, I actually got, I got some insights which I think like, okay, these, these, these are worldwide or worthwhile to, to explore further. And so the thing we always want to do with the ITOT insider is I want to avoid talking about technologies and I want to avoid just talking about buzzwords. I always want to, you know, I want to try to first understand myself and then explain what I think it should be about. So when we were discussing like, what is artificial intelligence right now? Because for the one, it's as I said, it's something chat bottle. Like, for the other, it's still machine learning. So we, first of all, we were thinking like, okay, what are the, I would say the big families in artificial intelligence? I have a mathematical background myself, so I clearly come also from the machine learning things and vision systems, et cetera, et cetera. So typically, yeah, we see, okay, it's about machine learning, let's say more classical AI used to be about generative AI, now it's about agentic AI. But that doesn't really explain the problem we're in. So we took a step back and we went like, okay, if we look to how a typical manufacturer is working today, then you could say that on the one hand, they have the physical twin. The physical twin is the actual thing, right? Yes, exactly where they actually make their money. And then they have multiple digital twins. A digital twin can be in all ways and shapes. It can be your data historian, it can be your quality system, it can be your engineering system. Maybe it includes some 3D visualizations as well. But you have all these different systems which contain a small or larger part of the digital representation of that physical asset. So the normal way of working is you put operators and engineers in between. They get data from one system and they manipulate it in the physical twin. Now, luckily, a couple of years ago, well, a couple of decades ago, we introduced this thing called automation. And with automation, what we basically do is, well, we do the repeatable work, which we can easily program into a control system and we take that away from the operator. So again, the automation system gets some data from the digital twin, make some calculations, and sends it to the, to the physical twin, to. To. To work on. Now if you look to what we are doing with AI right now, then I would say that AI are bolt ons on these different modules. So call it an add on, a bold on, whatever. But best thing we can do is, you know, now, now we have, I know on, on our, on our historian or on our trending tool. We now have an AI agent which makes it a little bit easier to get some, to get some data out of Trends or the automation system. Maybe you have something new and you have integrated it with Claude and you can maybe the engineer, you can use some LLMs to make the engineering process a little bit easier. But there are still add ons. So if we really want to make a step change and we really want to have this AI thing embedded in our operations, then we have to do two things, maybe even three. Let's see where we come. So the first thing is obviously all these different digital twins, they can't live in silos. Somehow we need to put a data platform next to them. Brings us a little bit into the UNS discussion. But okay, so first of all we need to connect these digital twins, which means where is your master data, you know, who rules them all, etc. Etc. Lots of challenges. And then secondly, you can only have a good AI controlling the physical twin if you also understand the physical reality on the other side. So that means that somehow you need that physical, those physical models as well. And if you have those two, then maybe we can start working towards something we now call a virtual engineer or a virtual operator. It is not really important, but then you have the physical operator which gets augmented by a virtual one. But we can't do that with just the bold ones we have today.

Yeah, yeah, it's super interesting because also I think, and by the way, I love the multiple digital twins because you're so right. I mean, maybe this is a dumb question, but if you have a representation of an asset in your ERP that's also a digital twin, even if it's just of course, of course in your erp, your mes, you know, you may have a different one in your maintenance system, whatever.

Yeah, but just think about the level of the application we also have because typically in all of those systems you have some representation of a part of the plant. And there is also so many, so much duplication going on. I used to work also a bit on the engineering side. How many? I'd say things still live in access databases, Excel lists, in copy of a copy of a copy of a P and id, there is so much application that it's. Even if you have the data, it's hard to go like this is the master and this is how it all connects to each other.

I was my typical prep for any conversation where I want to know a little bit about the person that I, that I won't find just by doing a quick Google search is. Is to do a deep research through ChatGPT and then upload it into NotebookLM and have it make a podcast for me. So in the podcast about you, what ChatGPT came up with which you're gonna have to tell me if it's something you really said or if they. They made it up. Oh, I guess this was Notebook LM coming up with it. But they said while you think of data in a manufacturer as hopefully being like a nice library where you can go to the right shelves and find the right book and there's one of each where makes sense. Duplicate it. If you, if it should be what it is is like a bunch of file cabinets, some locked, all with different ages, like types of data duplicated all over the place and then a bunch of piles around all as well. Plus some missing data.

Well, it sounds like something I would say.

Yeah, well, anyway, I just, yeah, the, the, the duplication and then the still missing spots. And when you were describing the physical twin, the machine itself, let's use a piece of equipment as an example. And then the digital twin. The thing I was thinking of was not only the instructions coming with some automation from the digital to the physical, but also the physical to the digital, that even a manual record of what was the runtime on this machine, what was the downtime, all of that, even if someone is writing it down on a clipboard and then giving it to their manager to enter into an Excel somewhere, which is also maybe a digital twin or one piece of it.

Of course.

Yeah, it's all the same type of let's keep the digital and the physical in sync. But that it's been so painful in so many ways to do that.

Yes, I'd say the Itot insider, you know, our work, we don't talk about. Well, we don't really talk about specific technologies or specific choices someone can make. But I think the most important thing, I also say I don't know if your notebook LLM extracted that, but that's typically like there is enough technology. The technology is for this particular problem, not a problem. It's how companies organize themselves. How. Yeah, making the business case to solve This, I would say this data foundation is a very, very hard, very hard case to make. Having said that, the problem is if you don't do it, then yeah, good luck with all your other AI initiatives. They will fail.

What do you say? So say I were manufacturing leader. I've got 50 plants. 10 of them are a recent acquisition. The other 40 are mixes of different types of manufacturing. We have some vertical integration, so lots of different things that we do. Lots of different ages, makes models of equipment. My oldest one is 1952. How would you. And we've got some plants, especially the recent acquisition. Four of those 10, I'm making stuff up here, but four of those 10 have an M. Yes. The others don't. How would you walk me through the process by which I should be thinking about what to do and what to do first?

Yeah, you're not making something up. This is like the definition of manufacturing. Because also how many greenfield plants do we do? We really like pure greenfield. I'm not talking about an extension somewhere, but pure greenfield plants. We're not building so many and even.

When we do, it's often there's second or third hand equipment in amongst.

Yeah, yeah, yeah. And even if it would be new, new equipment, then typically we would build it with I would say the ideas we have at that time which most likely will not be the same as what we did when we build our previous plant a decade or so ago. Yeah, that's a really good one. Actually my latest article was actually all about this and the. Coincidentally so I think the most important thing is that we shouldn't change the say the automation foundation. We have to really treat it as.

A given by automation, as the foundation. You mean the instructions flowing from. Oh, let me keep consistent sides from the digital to the physical.

Yeah. So I would say the. Yeah, yeah, your PLCs, your SCADA, you know, the HMIs, the things which are actually the real OT, the OTOT part of it OT I've seen so many projects fail where at a certain point in time a certain team says like you know, we have a good idea going to standardize, I don't know, our naming convention, our PLC vendor, you know, whatever. So that's bound to fail. Even if you would start such an initiative, as you said, a new M and a coming in very long life cycle, so you really do not want to touch your pure OT layer. So that means that at a minimum we should start by building a layer on top of that. A layer where first of all we need to figure out what is the Most important business case, most important use case, we have to or we can solve right now, and that's not so easy. But in order to find that use case, that means that you have to have this collaboration between typically some kind of a central team and combining that with local experts, probably those plants who are, I would say, most eager to start a change. Now this now becomes a really interesting trade off because obviously we typically say like, yeah, it should be something particular for a site, but you also don't want to customize too much. So you need to find those use cases where you see that there is a value to run it across. I would say multiple plans. Where you see like, okay, there is scaling potential, where you see that there is a business case probable and where you have at least some first site managers or you know, whoever is involved who go like, yeah, this actually makes sense for me to step into this. To do so you need to have some level of standardization in your approach. This can't just be like, I'm going to try this and this and this. You have to have some kind of standardization, standardization in everything. So your software stack, your hardware footprint, you don't want to go to every single plant and then decide on a new hardware footprint or a new cyber security footprint because that's also not going to scale. So you need to have some kind of standardization. And then the difficulty, and that's really, really difficult is how are you going to build that namespace? How are you going to, I would say, tie everything together in a way which is scalable on its. And that's, I'm going to be really honest here. It's a challenge because yes, we have something called ISA 95, which is a good foundation, but still you need to start somewhere and you need to make decisions and who is going to make those decisions and what is the layout of our plans, etc. It's not that easy, but it's possible to do that. And then after, if you have these early successes, then you have to do something which is very, I would say, not really a common thing to do for engineers. And that is to be really proud of what we've done and to market the hell out of it. Right, because you want to create traction from your other sites as well. But yeah, it's like this engineering thing.

To do completely and I love that. So thinking about, okay, what's standardized and then what's co built together with the sites and finding the right ones who are going to grab and pull, two things are coming to mind. First is I really liked a conversation that we had at one point with a great customer who was trying to figure out okay, where, where do we deploy first, which sites? And they said all right, we want to look for sites where the leaders a know what they'll do with data. They know how to use it.

Yeah.

And two where they want it. So they know what to do with it and they want it. And that's the winning combo. And that makes everything else a lot.

This is not really data platform project, but many years ago I did a digital operator project where you mentioned the paper. So where we basically just got rid of the paper and went to a tablet for manual entries the way we started. First of all, indeed we selected a plant where we know like they were super hungry to change something right now that happened to be the biggest one on the side. But that doesn't really matter in this case. Good thing is they had money. So that's always easy because you need a sponsor. What we then did is we selected some key users from the shift teams so the operators themselves and we really started identifying like okay, if we would be changing paper to digital.

No.

What would that actually mean for you as an operator? Because when, when, when I got the, I would say the mandate for management, David, we need this. Then the, the, the, the idea the management had was if we change paper to digital, we're going to reduce the time it takes them to write something down, which is obviously crazy. But if you, if you go to the operators always do the same, the same checks and it doesn't really matter in this case or we only want to do this if this happens, etc. Etc. And then even somebody said like you know what actually would be really really easy? If the previous value would actually be pre filled. Management went like no, no, no, we can't do that because if you pre fill the previous value then maybe they are lousy and they're just clicking next, next, next. But the operators went like David, no, because if I immediately see the previous value of the previous shift or whatever, I can immediately spot that there is a change and I can act upon it that different mindset from between the operator and like the top down management mandates. So to say those differences were really really big. But because we really worked, I would say day in, day out with the operators, it became a success because they were at the impression like okay, they actually understand us, they make our work easier, better, more fun maybe. And then so when that project was a success I was asked to, to. To. To to internal promotion to Roll it out to the others, to the other sides. And I said like not going to do that. I think that we can only do one thing. You know what, we're going to have operators themselves talk, talk to each other, let them talk and then let's create a pool. And the years to come, every, every, every plant want not because we said so, but because the operator said it to them between, between themselves. That's what you want to, that's what you want to create.

I love it. How did you just very tactically to create that plans create forums where operators from the plant that had had the digital solution and kind of co built it could ship. Did you use teams? How if you were putting myself in the shoes of a manufacturing leader saying I want to do that tomorrow, how did you do it?

Yeah, that's pre teams but I think important thing we did is we actually tapped in into the existing internal communication channels of the site. At that time that was Physical magazine which was given to everybody who worked on the site. And you know we really showcased in the pictures we took. I wasn't on those pictures. It was an operator doing his story with a tablet. And that's just how it went. And then the discussion of the conversation became easy. And the fun thing is that it actually took us a couple of years to start building actual business cases for this tool. And the reason is because in the beginning you're basically just, I would say replacing what you have, you make it a bit better, maybe a bit faster. You won't be able to find that business case immediately for these types of applications. But if you give that a while by just the simple fact that people start using that data, that after a while it should be able to get that business case rather easily. And that's true for everything for the project I did so many years ago. But it's also true now for a data platform. You need this integrated digital twin in order to actually get value on your data consumption or your AI site. Yes, it's going to take you a while before you get that first tangible roi. But, but you need a management team who goes like yeah, we do understand foundations are important, so let's invest in this.

Yeah, yeah, completely. It's, it's that kind of here's what you need in place before you can start, you know, achieving, achieving benefits. When this is me being very. Asking a very basic question, but what do you mean by unified namespace? If you had to define what that means for.

Yeah, very, very simple. Centralized data accessing context. Okay. So that means that centralized data access, that means I have some kind of a platform approach where I can get all my data. Whether it's historian data, whether it's MBS data, whether it's quality data, it doesn't really matter. All the data is available in this centralized platform. So that's the first part of the definition. And then the second part of the definition was in context. In context means I have an asset context, ISO 95, I have my production context, for example, I know what I was producing. I have a quality context, I have a maintenance context. I know what maintenance problems were active at that point in time. And that is my simple definition of a unified namespace.

One of the things that I'm curious about is we've got the erp. Most manufacturers have one pretty good. Generally up until the moment a production run is scheduled, pretty good. Also, once a product exists and is ready to be sold, what we see a lot is there's the whip black hole in between production unscheduled and product ready to be sold. It's the hodgepodge. That's where hodgepodge becomes the worst. And, and the problem in manufacturing is WIP is what a manufacturer does. You are literally using machines to transition raw material into product ready to be sold. So when you.