Can AI Run Your Plant ?

David: When I say AI and manufacturing, what comes to your mind? Willem: Fully automated, no people present, twenty four seven sides that look like magic almost. David: Can AI run your plans? Willem: Take a good look at at what's really changing. Things that robots can do now compared to five, six years ago. It's really moving fast thanks to AI. David: also not just what they can do, but how we program them, how engineers actually create the software to do what they have to do. Willem: So, David: Here is a simple question for you. Can AI run your plans? Willem: If you believe the marketing, yes, it can. but you know, at the IT/OT Insider we need to go past that bling message that you hear everywhere, David. David: Yep, that's what we do. That's what we will explore also in this video of the IT/OT Insider Now, before we dive in, so we are recording this video exactly a month before our book is going to be released and we need your help in beating the Amazon algorithm overlords. So for those who don't have experience writing and selling a book and the only way to be visible is if you pre-order, if you leave reviews, that way we shoot to the top. So please, it would mean a lot. If you do so. Having said that, Willem, when I say AI and manufacturing, what comes to your mind? No backflips? No backflips. Willem: Only if it's useful. but yeah, that's what comes to mind. Those fully automated, no people present twenty four seven sites that that look like magic almost. I think I think an automation engineer's dream basically. David: Yeah, yeah, yeah, kind of. But is it really true? Is that happening? Willem: I think lights out, I mean, exists already for a while. you just need to look at the right places. We see examples of car factories that are heavily, heavily automated already for a long time, and then they're going even more in that direction. But more simple warehouses. Yep. That's something that's been around for quite some time, is fully automated warehouses. And it's the same thing, you have those robots going everywhere, picking stuff and even packaging, all happens automatically. David: reorganizing the warehouse, making sure it's fully optimal. So those are indeed very interesting examples. However, in a pure manufacturing context, there are a couple of, I would say, a set of factors which makes this possible. First of all, we need to have a structured environment. We can't just keep on slapping AI on everything. Processes typically who are repetitive in nature, where there is not too much variation. And I think that's really an important one with automation, digitalization, and then maybe even the very advanced Lights Out concepts really taken into consideration from the start, which means it's more applicable to Greenfield. type of plants, so the ones who are newly built, are newly engineered, than the ones who already exist for decades, maybe even a century. Willem: Yeah, yeah. I think if you think about a warehouse example, a warehouse that's been there for like fifty years, always been operated manually, you're not just gonna transform it with software into a fully automated warehouse. Impossible. David: On other hand, we started talking about AI. The examples we gave, isn't that really AI or is that just really, really clever automation and digitalization? Willem: Some of it is. I mean, let's be honest. I think everybody needs to say they do AI for a couple of years now. But take a good look at at what's really changing. If you're looking at robotics, things that robots can do now compared to five, six years ago, it's really moving fast thanks thanks to AI. David: also not just what they can do, but how we program them, how engineers actually create the software or have AI create software to do what they have to do. Willem: Vision systems, David, I think I mean we still worked in a time where this was much more deterministic. Now with AI, vision systems are vastly more powerful than they were before. And and that's something that's accessible to everybody, I would say. David: It's clear that in those cases, things go faster, things become more easy to implement. However, most of the existing companies, so going back to those brownfield plants, they don't have the luxury of having fully digital engineering files, digital twins containing all data. In new plants, have simulation data of what is going on inside the factory. Brownfield plants typically don't have that data set. That also means that having AI run a plant might be a step too far. However, current AI technology is very, very good at something totally new or untouched until a couple of years ago, and that is working with knowledge, putting knowledge to work for us. So let's explore that a bit further. Willem: David, we've been working on data in manufacturing for a long time now. but if we're honest, we've been neglecting and not looking at a huge set of data for decades. Because it was just too difficult to handle. I'm thinking manuals, procedures, shift reports, incident reports. B and IDs can be digitized, but how easily are they integrated with everything? all those types of Data would take a huge team and even then give mediocre results, let's say ten years ago. So we just it wasn't worth the effort. And I think now that that's changed. I mean, we're all so used to it by now, but handling text the way that AI and LLM specifically do opens up a lot of opportunity for us. David: Yeah, let's just take a simple example here. So suppose I'm an engineer. Well, actually I did that type of project myself as well, but like I'm an engineer and I want to understand why my pump P101 has been losing performance over the last so many weeks. Like what are typical steps an engineer would take? Willem: Okay. maybe he would take a look at the pump first, understand where it sits. David: terribly. Willem: Yeah, or maybe sometimes during maintenance they might take pictures. Maybe he can find out what's happening if there's some kind of corrosion happening over time. So he would look at pictures maybe. He would look at the P and ID to get a a feeling like to what is connected, what sensors do I have. he would maybe query the PIM system to find out. How much power is it drawing? What are tank levels? Those kind of things. look for an all of these there. He might look at OE reports to find out if there were downtime events linked to this. Search the CMMS for maintenance history. You would then pick up maybe the manual of the manufacturer to find out where the safe operating points are. Are you operating outside of the margins? Take a look at the PNID again. So if if you miss something, read read shift reports, talk to people, and all that would go into his mind to come to an analysis. And if you would say ten years ago, let's build a software to support that, I think we would have said like just don't start. David: difficult. Absolutely. Willem: while now it's perfectly feasible for AI, have a couple of agents to process those shift reports, look at the OE records, get that SAP data, start to correlate all those things, even look at the pictures from the manual with the curves, and maybe combine them with some basic knowledge and formulas on on pumps. I think you can build quite a performance system already to support that engineer in his in his mission. David: That means that we are basically taking steps not towards autonomous control, towards autonomous investigation or just a very clever aid. Willem: I think you start you start with a clever eight. I think that's the first step. Yeah. because it's gonna be hard to go from zero to an auto autonomous system looking for data. But once you've noticed that it really works, I don't know if there's much stopping you from letting the AI read your data and looking for correlations on its own. And then coming back with findings to to the plan, to the operators, to the engineers. So instead of Asking AI to control your plant, lights out fully automated. I think this is a thousand times easier to achieve and more realistic for brownfields. David: Well, actually, I would even say for every industrial setting. Willem: That's true, that's true. David: Now, before people start going like, yeah, but that's also what the marketing because what we're now saying is basically what the marketing also says, right? Willem: In a sense. David: How do we achieve that? What steps are needed in order to have those agents start running and start investigating and start giving us advice? Willem: There's a spectrum on being able to execute on this and not. And I think one of the biggest factors to be on the ones that can execute this is gonna be your access to contextualized data. It's what we've been saying before. No matter how good the AI gets, if it doesn't have access to the data, it's gonna be useless. So all the work we've been mentioning about structured data, time series data, contextualizing it. getting an idea of events happening in your plant, that remains absolutely critical. We're just going to have an add a la we're gonna have to add a layer on top of this. And that's gonna be how do we manage the knowledge? It could be generic knowledge, knowledge about heat exchangers, knowledge about pumps, knowledge about whatever equipment you have. Could be site specific knowledge, like we process these things this way. I think That's what you would put other context management or ontologies. That world is is relatively new. but I'd say without these two elements you're not gonna get very far. And if have those, I think you can do really cool stuff. David: That's the technical side of the coin. Coin has two sides. The organizational one is equally important. And I think for these types of innovation, neither IT nor OT can do this alone. Cooperation is absolutely essential. Now we already recorded the video which we titled, what is IT OT convergence? And we talked about that in that video as well. for those who would like to take a look on the more organizational side of things, go to that video. Can we end this one with a statement that starting today is always the better option? Willem: I think the alternative is other way to put it is like there's two plants. One started today and one just didn't start. in ten years. Which plants do you want to be? Well David: That's a no-brainer. So with that piece of advice, collecting data, contextualizing data, yes, it's a gigantic effort. Willem: I thought you were gonna say something else. David: And I'm not gonna cut it from the video. It's a shitload of work. No, it's a gigantic effort, but it's like eating the elephant. So piece by piece starting today with this video, we try to explain you a little bit about industrial AI. We have all the industrial AI videos on the YouTube channel as well, as well as data platform videos, UNS, et cetera. So you can explore all bits and pieces on our YouTube channel. Now, if you want to go pass buzzwords, If you really want to dig into digitizing manufacturing, go to itotinsider.com where you find all our free content. Go to itot.academy if you want to explore these topics together with us, which is always, it's a lot of fun. Willem: It's a lot of fun. And I love those. David: And also the interaction with the other people in the core. Willem: That's why we do it. Absolutely. It's not a it's not a remote e e-learning course, by the way. It's very interactive. It's not recorded. David: Absolutely. Or finally, we already mentioned, grab your copy of the ITOT handbook. to itotbook.com or your favorite online bookseller. Thank you very much. Take care. Bye bye. Willem: Bye bye.