Time Series: The Most Underrated Domain in Industrial AI

0:00 This podcast is presented by NXAI, your partner for time series. foundation models and physical AI. Hello everybody! And welcome to a new episode of our Industrial AI Podcast. my name's Robert Fieber... ...and it's a pleasure to talk to you. Peter Sieberg good morning afternoon evening to you, Robert and all of the listeners wherever you are wherever you are in the world.

0:33 You're sending out of the caravan Peter, right? Yes I'm in a country that yesterday celebrated their national holiday because it was... I don't know how many years ago. Fourteen ninety two Columbus arrived. I think what is today called Sun Salvador, i'm not there. But the country who celebrated their nationality yesterday that's the country you're in.

1:00 Oh so your still in Europe? That's a quiz now! Yes exactly and thats the quiz. Im going to be asking more questions actually... Me too because i caught up in trade show fever because Fakuma is running right which our main feature. it all about Fakumaa Peter. Oh, yeah very good I. Yeah the big news of The last is it a loss week?

1:25 I think at least as far as I could see was the fact that Schneider wants to buy PTC. That's what I saw me. so there were so many comments And i Think It's A Very Very. It Is Very Interesting. I think This Is More Than Just One Of The major automation companies buying consolidation. Yes, implicitly it was also about consolidation I believe.

1:50 so Schneider offers twenty two billion dollars. is that too much? That's one question amazing. Is this about keks or computer added x computer edit X for design computer design? This why come from what i studied engineering whatever uh I way of engineering. Is it about both? Let's discuss. so where i would like to start maybe very quickly is because, this is a big thing now for me and let see what for you Robert who was... And what is Schneider ?

2:28 I have known Schneider as French automation company mainly for building and lately also data center automation And the main supporter, and as I was looking in a formal role of my life at all different communication protocols. There's hundreds right? They are big support for this backnet. is building automation control network protocol.

2:53 Is that how you've seen Schneider roughly or... No, also see Schneider in our Automation business. so in our industrial automation business i think You can compare Schneider with Siemens. So it's French Siemens. They are not building trains, right? But in the whole automation business. And I think Schneider is also in a field of robotics evaluating stuff there.

3:15 so i think they're on their whole Automation loop and the whole animation circle when it comes to PLC two robots to data centers too energy. but When I was at GTC last year Jensen promoted Schneider as a data center partner and the whole stuff. I would have more thoughts here, okay? Yeah And i think that's a huge market for them.

3:37 they hold data center market too. But yeah, i think you can compare it with with the Siemens guys And there's also some management exchange between Siemens and Schneider. Yeah, if this is going to go through they're gonna get become closer right? So let's see PLC is you know... They offer a range of digital solutions computer design product life cycle management for industrial manufacturers.

4:02 I haven't been that close with them. i'm not sure if you have been. So, If we look at exactly what you said and maybe I've known only one section of Schneider so starting with...if they were..at least the word is Building Lately Data Center plus What You Know They've Been Doing Also in Industrial Automation.

4:22 Now they bought Aviva right? In twenty-twenty three. last week i had David Rodriguez interview we had in live, he said you know PI system is his bread and butter. And after started with OZ Softie explains that was bought by Aviva. so it's one of the things actually now is belonging to the Schneider group, right?

4:47 So all of MES systems maintenance in its portfolio. And then also June of this year only I think they announced a definitive agreement so i guess it has gone through them. off the Norwegian. we talk about little bit countries as well because afterall there's going be consolidation concentration at least European-based automation activities, right?

5:14 And they are acquiring the... What is it? Industry? Data management. Cocknight and they serve industries like energy process etc. That's huge, that was huge. And now they're going to add PTC. so yeah along this complete let's say life cycle management the stream They are adding more upstream I would say.

5:37 So it is all computer added design which maybe in the past did not have PLM stuff, right? Yeah. But also computer added design really so it's more like the upstream. before you start maybe designing whatever they... It is that they've been building. That may be what didn't have in their engineering but its interesting because You can compare Altea or PTC.

6:02 remember last year Siemens acquired Alteas The Engineering Simulation guys. So big automation players going fully into engineering cat, cam simulation stuff. That's very interesting I think. now what yeah and What for me disease? Well it is that they're doing i believe Is that They are not just thinking of.

6:23 you know if one product have two or three Or four And they come after each other. so that's always about integration which almost never happens So difficult. no It's Not only the integration Of The humans. I mean we as Humans Are The most Difficult Part. Typically Because we don't like, We are children sitting in the sandbox.

6:44 Sometimes were very happy but sometimes... ...we're not that happy and we do want to play with other children right? So we are the biggest! My point my personal perspective. The investors haven't seen it because their stock was down from three hundred to two sixty or something so huge. ten, twelve percent.

7:07 I looked yesterday. i'm not that sure...I do strongly believe what the Schneider people see in this is that In The End they're going to provide What? I will call in my words a Schneider personal AI engineer. and This Engineer which of course Is in the end some kind Of Model. it Will have extracted the core value out of all these different software pieces.

7:36 They're adding one, right? And this Schneider personal AI engineer my term not their turn. it will be offered to existing and the new customers as well. It'll offer a new way for creating high-level value by phasing our individual software in my perspective The individual physical software very soon over the next five years.

8:03 The value is going to be going two zero. That's that's my perspective. it can only be like that I believe if It was a good old i buy one. in My line there's One missing a by-one and that's gonna take you know, Five to ten Years. no way would that ever have been? had the value of twenty billion. I believe it is only with this very, very much higher level new way off doing things through AI.

8:32 my perspective from my perspective its closing a gap. right. so they have to get when it comes to engineering Ed Schneider and how's it called the PLM stuff from PTC this name? Windchill. Oh, is that one? yeah I'm not in detail though. Yeah And Creo...I think Creo has cut stuff right. Now when they come to design They have now a solution on their market.

9:01 But to the engineering field, right? Because they already knew that how to run machines and what customers are doing with machine. They were not involved in design process of a machine. And now they're closing this gap because Siemens had this engineering capability. Right! Now Schneider has also this capability.

9:21 Yes I completely agree but as i say... ...I do not believe within next couple years their customers are going to be working the same way that they've been doing. We see so many new kids on the block, young people just doing the engineering a complete new wave by saying what is my job now? To buy an electrification house piece factory whatever?

9:53 and I say my LLM do it! And we're seeing that, Davey De Maio you did the interview with. He's showing now every day how he is building a plant and How Is He Building It? I think this guy has been what was it like ten or twenty years in China? Whatever! But he can do it With The Tools but Now He Isn't Doing It With The tools.

10:22 it is impossible to believe, understand that Schneider believes the day their customers are still going to be doing in an old way. No way! That cannot... It must be as you say filling a hole. and then if they wouldn't have that whole filled than what I call this AI person engineer by Schneider would not been capable of helping Their customers out in three of the four segments.

10:50 Yes, but not an in-the-force one. Yeah exactly and you can see that That they whole engineering world is changing for example with the Emmy guys acquired by Mistral And not a quiet by Siemens or buy by Schneider. They're acquired by mist trial. right there are new kits on the block as he mentioned and these New kids will change how industry?

11:11 How engineering we'll work in the future. our physics acts. That's another startup in the whole engineering world. So yeah, maybe you're right? Maybe it's all about data. It's all About getting access to date again and access to customers. Maybe that's more important than windshield or Crio something like this.

11:31 Yeah, it is the axis through The data but the data its intelligence In the data rights. so Each of these, it's not the software. It is NOT the line! There are millions and millions of lines of code in whatever language. That´s the software that does things for you. but by the fact that these pieces have been running for ten or twenty years... They have been gathering knowledge about the environment of each computer edit, whatever they were.

12:07 That's the value! And that is the knowledge being extracted by a large language model. So this is what The New Kids on the Block... they need to all of them find for solutions. If we talk robotics, then we have the robotics. go-to-the-gym is one way. We know that Nura's doing it. There are many other ways.

12:31 But if you are a company like Schneider or Siemens, You know...you have all exactly that data and you can have the AI The models go into it And learn about them. I believe this is what they're talking about. What else Peter do you want to say? First of all tell us at the exhibition I will join the exhibition today.

12:56 We are recording on Tuesday, we'll join that exhibition to day because as you record an episode with Bastian Simcon right? So i recorded and episode was a our book guys! I was very impressed. so in the main part Werner is waiting. Very impressed Because he has really interesting AI strategy when it comes to the our book or model.

13:17 so they're very interesting. but Um, yeah. The whole industry injection moldy is very fast forward when it comes to AI for example angle. that's what the exhibition Is about right? Yeah exactly Exactly. It's About Injection Molding. Stuff Right and as Werner always said its a Very Complex Process. Its Very Difficult.

13:39 You Can't Find People Anymore To Run The Machines. you need A Lot Of Domain Knowledge And They Are Looking For AI experts or not AI expert maybe for a solution that helps the normal in quotation marks, Normal people to run an injection molding machine. So because of labor shortage etc. and I'm really looking forward To meet The guys there at Fakuma Because i think it's A very innovative branch And um...I hope will see a lot Of AI solutions.

14:11 Okay, looking forward to maybe so. Maybe Dan is talking about Simcon at it now. but just another example of young guys. Young people You know coming up with a complete new way of doing things right? So we started talking about which is now may be the big four Big five still most of them Now With a European head office.

14:34 some of them you know within the glow the conglomerates they'll being run out of the United States. That's one. and the second, as you say when we need to people maybe just has a quick reference. A sort of entropic investing money in training what they call frontier deployment engineers right? FDEs When I was In One Of These Tech Companies There Was The FSE Right Field Support Engineer And that Was More Like You Would Have.

15:03 Your Customer Would Start Designing Something And if they would have a problem, then needed help. Then you know in this case the chip company will send out person who knows most and that's something similar. They call it now The Frontier Deployed Engineer. but basically those people get into companies.

15:23 maybe what is going to be happening In what you just mentioned as well. And those people can then work with whatever tool, Whatever maybe AI or large language model they have chosen and know the complete environment. so they know the injection molding machines inside out and then they learn to use the tool.

15:47 And I'm going to come with a specific quote from Gitta Kütnjak, you recall later on that is gonna bring it all together as well. very interesting Perfect. I have one news because things are about to change around here too, the podcast is going to change a bit. we will be producing more episodes but with this specific focus.

16:07 i won't reveal any more than that for now But it'll all kick off in September. No November sorry at SPS. So We Will Have Something New Here With The Industrial AI Podcast But We Will Reveal That In November. so I'm Really Looking Forward. It's not surprise for you. What is it going to be talking about now?

16:35 Okay, let see we have a new co-host. so Peter goodbye. That was your last episode. greetings To the human or an AI as a humanoid One that can walk or one that can drive only speak. all these. I only need persons that can speak. Yeah, very good. Okay I have...I'm going to extend Bogdan Kirste. He had a conversation with a tactile sensing founder.

17:02 they make gloves. you know there's hundreds of these sensors on the finger pads and their work is they wear them right so that it's same topic almost like as what i just said again away producing the capabilities around, in this case, robots. And they would endure normal shifts in electronics or automotive factories and then capture pressure temperature vibration whatever.

17:30 right now what day found out? I'm going to ask you after collecting data across thousands of hours real factory work analyze which fingers. First I'm going to ask you how many fingers we're doing the actual work during this high value manipulation. How many fingers? Five, four or three free? yeah you're right.

17:54 and which ones were there? You think oh I don't know. they are to one two Three. so at a thumb of the index in middle. okay well i Think if you go To your right hand And then you look At it for the thump we all Know what the thumb is. Second thing The One that Is next to It. That's What We call the Index finger.

18:10 i guess In the Middle One. This Easy as Well because it's the Middle one. So long story short Why is that interesting? I think it's as far as this company, doesn't matter who they were. They now start building instead of a five-finger robotic hand thing, robotic hand. Not sure that you have heard about that or if you've seen it maybe already before?

18:35 I haven't seen it before. think yeah You seem to three finger ones for years of not sure that we are. you and i Have already the experience in In knowing That The Three Fingers you know represent whatever ninety Or Ninety-nine percent Of the use cases. yes so And I guess in the end It's is another Confirmation of the fact that the human form factor If I may talk that way about our human body is not the ideal for working in a factory.

19:07 I guess, right? So there's always things that can be better than you know. i think In the end The Human Body was nested Not Necessarily. What Can I say? made produced I don't want to Be too specific There but arrive as the form As we have For work or Work Only Right. Well That's Peter. so Vanna Is Waiting.

19:30 Yeah, I have a close-off. A close off by Gitta. Gitta called Tunyork. We had an interview with her i think one year or two years ago? No it's three years ago. Three is running when you're having fun Robert. She's the chair for mathematical foundations of AI at LMU in Munich that's where you studied as well and she asks why should accompany hire to do tasks an AI agent can already do faster, cheaper and sometimes better.

20:03 What is her answer? Because it was gonna ask you first! You're a very small robot...you immediately want to know her answer. well okay yes I left that at the end. so Gitta increasingly things we should not ask whether student can outperform an AI agent, but where the student plus AI can achieve something that neither could have achieved alone.

20:33 She's looking for a sum of both. she saying like you know what is it? That one plus one makes three You know, rather than two I guess. Is it a hope or is...I'm not sure? Did we discuss that last time and never? no because there's always so many messages when Jensen maybe you should invite her again. yeah And its exactly on this topic Because she the mathematics chair.

21:02 right And was it that Jensen, Jens Jensen said you know we don't need to study mathematics again. Is there what? Yeah right yeah We discussed last week or not? I think That's exactly this very topic. now she is not saying that Right i mean She the cheerful mathematical foundations of AI. There You go.

21:21 but What should I say? it's not is brave? It was almost the wrong word, but But asking this question. I didn't see what there's gonna be. many Many people said oh what's happening? they are you to chair from that thing. let's have a tata. I think it's very very very correct That she exactly in this position asks exactly that question because i think at all of our universities All over education Is changing needs to change And I guess we're all part of that.

21:53 You just mentioned maybe Werner is going to be telling us, like now he needs people who know not only the tech... No! We talked about a token budget for his employees. Well there you go. so looking forward listening to you talking to Werner The Ur model. So I have no translation for the ur-model, what is it?

22:17 Well maybe its one of these terms that are going to stay you know as yeah because you recall that Siemens came with this word Eigen. when they come in their Eigen solution and there's certain words will recognize the word UR model is like, original. The original of all and that's what the UR Model is right?

22:47 I guess i'm not sure. It was interesting because there has one quote in this very nice book. Because I asked Stefan do we underestimate time series when it comes to the URA model? Right And then he laughed and said yes everybody is underestimating Time Series a lot When it comes To This So Very Interesting Episode.

23:08 Yeah, looking forward to thanks a lot Peter. greetings. To the country where Columbus were. Columbus was born or very way left? Where he left? Okay in fourteen ninety I don't know when how many years it took. a couple of years you talk before then forty ninety two believe right here I've been perfect.

23:29 Thanks a lot bye-bye. Thanks guys. Bye-bye, Robert! My name is Rubel Wieber and today I invited two guests Stefan Lamp und Werner Faulhaber from Aarburg. welcome guys greetings to the Black Forest. Hi it's a pleasure to have you with us. before we start talking about AI, Aarborg & The Uhr model We need to explain that little bit later.

23:52 what is the Uhrmodell? Please introduce yourself briefly to the listeners. Stefan please start. I'm the team lead of the data analytics team within our work and innovation technology department. And my main topics are AI, chatbot everything we do on a product which has to do with AI. that's on my table integrating their great ideas coming from Silicon Valley in Europe and China.

24:18 basically yes perfect Vanna please introduce yourself briefly. okay My name is one of all our vice president research and development at our book. And AI, it's one of my main topics in the last four years. yes so we always get together at our events and that where you share your ideas. but The Last Time We Spoke publicly was four years ago.

24:40 So let's talk about division from Four Years Ago. Can You Evaluate a Little bit? What Became Reality? What about the UR model and the federated learning stuff? We will go into deeper in a next question, but what has become reality. Okay our vision was not simply to add an isolated AI function to the machine.

25:03 we wanted to look at the entire injection molding process. The idea was to develop reusable base models. Today we would call them foundation models for different processes of material classes. Customers would then be able to adapt these models through their own production environment by training them locally.

25:21 An important element was federated learning, allowing customers to benefit from knowledge gained across installations without exchanging actual production data means using anonymous status. What is Federated Learning? Is it that? No not at all I would say. Federated learning currently still on the bench but out of game and we're focusing more new trend with the foundation models explicitly, with a time series.

25:48 Foundation models like Tirex too because they allow us to move much faster and create value more quickly. but is there still market for federated learning from your perspective? I think it's in the long term cause again we get through federated would be still there. So like learning across customers, across production plans customer has.

26:13 so the idea is not that but it will be a little bit later than we expected in our initial vision. You mentioned foundation models. right and when you record episodes four years ago? Maybe that's not the right translation of UR model, but what is it all about? Is an UR-model now a foundation model at injection molding.

26:39 Foundation models are still the goal for our book? Yes I think they're still the goals. Federated learning remains actually interesting and we learned first requires useful base model with clear customer value proposition And without that we run into a chicken and egg problem. That's interesting, so when talking about a UR model, do we talk about time series data or is it more material data?

27:10 What else as part of this URModel. Is it also text related? maybe let's dream a little bit what could be in the URmodel? so In this vision I would say its like a time-series prediction model and back then We always wondered where does that base model come from? The initial vision was to get customer data across different lands And just drain a model on there.

27:33 The first step would be to get customer data and that's a little bit harder than we expected. Because I think the technique was not yet ready for that, right? So We saw huge increase of foundation models two years ago or something like this And it was earlier then that It wasn't late twenty-three Yeah.

27:51 so we knew we couldn't use Customer Data but there is no end-to-end simulation For injection modelling machines. What we did was, um... We looked at a different approach using synthetic data and the idea behind it was okay. so we don't have a model that does understand injection modelling explicitly but maybe we can broaden or generalize the model.

28:20 So it could at least predict patterns on time series data because obviously uh in the injection modeling machine also uses time-serious. And so back then in late twenty-twenty three we trained our first own base model or foundation model for time series data and that worked pretty well actually. Up until now, um you still use it for product stability assist which predicts when quality relevant process parameters are likely to drift outside the tolerance limits.

28:54 Okay You always solve the problem? or do you see improvements there when it comes to the model? Technical improvements, new architectures. New ideas how to run models etc. Sure that was a very simple approach we had. The model is still really small I think about five megabytes and as soon as We saw more and more time series foundation models would be emerging We stopped basically our own development And looked at those Within our machine control.

29:27 isn't that Werner your own model trained on you're own data? Isn't there a competitive advantage? because you said AI itself is not the competitive advantage, but now You gain something with an own model, maybe you can use the existing models out there to fine-tune them. Isn't that a competitive advantage?

29:49 The rapid development of foundation models shows that AI technology itself is becoming increasingly accessible. That means the competitive advantages. not have your own model and It's more than knowing how to use the right model in the right industrial context. For us, that means combining time series foundation models with injection molding expertise.

30:13 We have machine parametrization and we know about process understanding and ability to judge whether a model output is technically plausible or useful. So you mean only to say, oh now we are running an AI model on a machine. You need to have the deeper understanding about the process and you need more details?

30:36 You need domain knowledge And you can need to partner or transfer domain knowledge into the model and take the power out of the model for domain specific data The domains plastic material the knowledges something you have to combine with the right model. When we meet each other, You always tell me We are The most complicated industry in the whole world So our processes of a more complicated process.

31:06 In the word why is it so complicated? Why Is It's so special? can you Can you give Me an example? what is so difficult when it comes to injection molding? I think this not so different from other projects. sorry Well your way you Always say we Have very Different Very Difficult Problems To Solve. Yeah, the problem is that actually a machine controller doesn't know about the process.

31:29 And it's something we want to change and that's the idea of bringing the model on the machine combined with three-domains plastic molding and processing. So I think this step has to go in other branches where maybe little bit easier. Stefan, can general purpose AI models as Werner said time series foundation models understand injection molding machines?

31:57 I mean sure if you're asking GPD an injection moulding question it would be definitely able to answer that. But If your are asking a specific question for like our book machines, so how would I change which parameters? Would i have to change to increase the quality or reduce the cycle time. The answer wouldn't be that great because it doesn't know the parameters It does not know what machine works.

32:23 Because if you're looking at all these models are trained they using publicly available material and If your googling Our Machines on How To Use Them There Won'T Be Much Material About That. That's our domain knowledge. And that's why we see it as own task to put this the main knowledge into these. agents or a chatbots, okay?

32:46 Oh agent and that's that's The interesting thing. now agents everybody is talking about agents. are bookers also talking about Agents? Now sure because As you said everyone was talking About us. Okay What Are You Doing With Agents? I mean We Already Have A Chatbot Solution Like We've Asked Our Work about asking and getting an answer.

33:07 And with agents, the next step would be on the shop floor or where's the chatbot? Next to the machine. basically everywhere you can use it on your smartphone tablet also in a machine itself. we have an innovation for that. yeah and its maintenance topic there. what is the main use case multiple use cases.

33:26 I mean, it's the great thing about large language models that you don't have to explicitly have one use case. Sure maintenance use cases. so it knows... It tells you what do step by step when there is a maintenance needed on your machine but also when there are like alarm at the machine. What should i Do?

33:46 So By That we're Also solving The problem. There are less and less exports on injection molding within the companies of our clients. Now this is a chatbot application, right? So what's your idea when it comes to agents now – to have swarms or autonomous agent? because there we see I think deterministic machinery applications, i'm a bit skeptical when it comes to the whole ancient thing in a short term period.

34:17 Maybe ten two five years. but what is your opinion on that Stefan? Sure we are also sceptical because do you already want an agent which can change the parameters of your machine? Not sure! It could be... In first step it would make recommendations at least and give reasons why its recommending these parameters So you can read it and then, okay that makes sense.

34:42 what's actually recommanding. And the step to getting there would be if we're looking at ask our book now uses basically a classic retrieval augmented generation. so into regression. it retrieves the corresponding semantic correct data from training manuals or machine manuals giving the agent to have different tools, access a machine.

35:10 To look okay how is the machine configured right now? What happened in our last... Our signal's looking right now? Is that because you're also very ambitious when it comes to OPC UA Werner and whole standard of communications And we see Asian communication going via MCP, new ideas by Antropiq from an industrial perspective.

35:35 Do we underestimate these new communication layers or will there be an industry-grade communication layer? Good question! I think uh... We see a good foundation model and need high quality data for this And we are also working on the tool. bring up the quality of data out from machine to a level we can use on agents, for example.

36:04 We have one on Aurora, our books world means in the customer portal to support their customers and every need they will have, For example there maintenance problems where you have quality problems too. They are parts And so The idea is to have a solution. Okay, it's interesting because I started my career in the whole automation business and there was a chief editor-in-chief And he told me yes when i was young.

36:32 There were this Fieldbus wars and everybody who is talking about which field bus? Which communication layer? and The thing is still not solved right. so we are still struggling When it comes to communication layers and now We see new kit on the block coming agents And we talk about a totally new concept when it comes to MCP and stuff like that.

36:56 Stefan, do you have maybe not the solution but feeling or idea how will handle this? So I think what's missing most currently at OPC interface is semantics. What does that parameter actually do ? What happens if i change value of it? in which context needed be used? I'm not sure if switching to MCP would be the solution.

37:22 You could also go ahead and give an agent database so it can retrieve that semantics about these parameters, then just talk with the OPC interface by itself. Yeah, I think we still need to discuss this topic in the whole industry because We are challenging. We have challenging problems when it comes to agents and everybody wants to run agents In the factory but them very efficient and smooth with the right information to fulfill.

37:54 That's only, that is my point of view of this discussion. but let us talk a little bit more about use cases because Fakuma ,the most important trade fair show coming. you will present some AI news, some real AI use case or new add-on updates. talk a little bit about that, because it's the most interesting part for all our listeners.

38:18 What is this? You mentioned stability assist. what does stability assist do? So as I said, stability assists uses basically or foundation model that we trained ourselves and thus in first step if you're running It always needs a little time until it stabilizes. And looks at the quality parameters and after awhile decides, okay your process is stable.

38:50 now go ahead. all parts are marked as good parts. Okay its predictive quality or what can we frame like this? Yeah exactly that would be also in next step. so we're stable And we want to be stable in the future, so it predicts where their quality parameters are heading. So if they're maybe is there a shift upwards or downwards?

39:15 It will predict If We Will Leave Our Tolerances Within The Next One Hundred Two Hundred Cycles. Okay okay! This Is An Existing Product and you will update your product. You Will Evaluate What's Also Possible When It Comes To New Models etc. Yeah, definitely. So what it doesn't do currently is reacting to changes of parameters when you're setting a new parameter.

39:43 so It will start at basically zero and look at the progress anew. That's definitely a field where we are looking currently To be better. And uh The new foundation models which were released in last month I think Will give us answer for that problem. Okay Perfect Yo. also Demonstrate something with the guys from plus ten, but we already had an episode was them.

40:06 I think maybe one year ago or something. greetings What do you do with them on the shop floor? So basically what plus ten has is their product. plus and hopper which just said in simple terms this parameter recommender. so it basically looks at your process And after a while It knows how are. your process reacts to external influences like change of material batches If parameters are changed by the operator and later on, for example in a night shift when there's an unexperienced operator.

40:38 And again situation which occurred in the past system knows how it was handled back then can recommend these parameter changes to low level operators. basically okay? Do you integrate this solution into machine controller? How do we do that? What has to be said is the model still running in the cloud, but we're integrating their web interface into our control.

41:06 So no additional phone or tablet is needed. and also normally when you would use plus ten hopper on auto machines You'd need a plus-ten data collector which runs an additional server so that it's set up at your plant. This one isn't necessary because this one runs directly geistical control system. So it's much easier to deploy in because you basically only need the machine with option enabled and connected.

41:37 today internet, how difficult is it when out to sell cloud connectivity related to a product? In certain in this month or since one year something big cloud cloud cost are exploding. And How difficult Is It? or how easy is it? Or Is it a discussion cloud or edge? Or is it both, what does the customer feedback?

42:02 that's really different. Some customers they want to have the data on their own. That means no cloud version as possible. but for that we also solution. where then? on prem We are our ALS system and can support them also on Prem. So we can react in most directions And we have there solutions. Also Is it more a topic?

42:26 I don't want to have my data in the cloud or is it a cloud cost topic from your point of view, because when i talk to machine builders they say uh we cannot cannot sell our product. When there's a cloud connection for their customer and different problems with cloud one thing sure that costs but also connectivity because some plans are not really good connected to the internet.

42:53 We have examples in Italy, for example... In Germany too I think yeah! Not only in Italy not only in italy. greetings. What else, Stefan? So cloud cost is a topic. Infrastructure as the topic and data security sovereignty Is that also a topic? exactly That was the point I wanted to talk about. The infrastructure is still so.

43:19 the solution works with IT of our clients for example, have machines with five G directly enabled. So they just can connect it without integrating in their own IT systems. and with cloud you have the same problem because I had that connection via VPN or whatever into the cloud. It has to be integrated.

43:44 The machine has an internet connection. That's still a huge problem which i personally didn't expect when I got into the manufacturing branch. Yeah, interesting right? So you already mentioned predictive quality and there's another huge topic. nobody wants to talk anymore about predictive maintenance.

44:05 it still not solved yet. everybody is still struggling. And it's a boring use case, but I think it's in very unsolved yet? What is your topic when it comes to the whole topic predictive maintenance? because you said You have solutions. When It Comes To Predictive Quality The heuristic objective is to supplement established maintenance strategies with weather condition information.

44:32 That means we have predictive methods that can identify deviations earlier, improve planning... And avoid of unnecessary interventions. That's the topic we want to work on, but that means safety requirements for example component specifications and operating experience. they remain important when it comes to detecting anomalies.

44:55 is that enough? because everybody now talking about classification stuff. I think thats next step or am i wrong Stefan? I have same opinion because detecting an anomaly isn't that hard and you don't even need client data for it, just like from a few experiments but classifying the problem. So what does that anomaly actually mean?

45:20 Is this good or is bad still somehow normal? because it happens every I don't know one year... Or which part is at end? yeah exactly. we can only find out using field data And That's why i personally think Predictive maintenance is more topic of processes. So getting that data from the field like when you have a service case at your client, something broke... Getting that data.

45:48 so You see what happened before Something broke? topic in my opinion. When I talk to people who sell components, right? Pumps or something like this they always tell me how. it's nice to know the anomaly. It's nice that there is some thing. but they want classification. as Stefan already mentioned How important is to really know from a service perspective From your service technicians To really know which part has a problem.

46:20 As already explained data, high quality data is the key for that. Means we have a bill of material data from our machines since thirty years ago means if you can combine this Bill Of Material Data with a service ticket knowledge and bring it together then your possibility to find out right part. on this specific topic here.

46:49 Peter recorded an episode, I think three weeks ago and published the episode. And he did a fantastic episode with Bastian also quite well from Simconn. He talked about simulation. We know all of great stories like Johannes and Mistral. which role does simulation play in injection molding business for you?

47:16 Because we talk alot deploying models to predict quality, etc. But when it comes to engineering or to simulation which role does this topic play for you? Simulation allows us to explore a process virtually and identify promising parameter combinations before running the trials actually on machine. so those selected experiments can be executed systematically using feature of ours on the DOE assist, which automatically runs a set of experiments one after another.

47:53 This will again generate lot high-quality data obviously and that data can be used then to improve the underlying AI models. so there's connection from simulation basically over actual experiments. Okay, that's interesting. So is it a product you're working on there? Yes definitely It's basically also part of the vision.

48:21 we were not presented in two thousand and twenty-two. okay Can you evaluate a little bit more on this topic because that sounds very interesting. The first step I think Is something We will present On the Fakuma next over Next week. Me. That's called the First Trot Assist helps us if you have setting up a new process.

48:44 that requires time normally. You need material, plastic materials and normally I'd like to have an experienced person in there who brings the process stable. The target for our first shot assist is intended to shorten this phase And help users reach their initial process more quickly. That can reduce trial and error loops, make the expert knowledge more accessible.

49:12 It means we use a simulation here on this topic to find the right parameters... ...and bring in the first shot for metadata.". But is not AI related? or am I wrong Stefan? The first shot assist isn't implemented using AI models. it's basically classic simulation but after that finding good initial parameters setting up a DOE, then doing the actual experiments on machine and getting high quality data for AI model.

49:45 that's where we get connection to AI. Does simulation replace machine trials? Shorten time. normally you have hundreds of cycles to fight the perfect condition in the machine if your simulation is behind less than ten times to find a good solution for the first production. Okay, so do you accelerate at that time?

50:16 You mentioned cycles because cycle is also huge topic and cycle-time optimization especially in your industry. How can AI help to optimize cycle time on your machine? Cycle Time is main topic, but cycle time must not never been considered an isolation. that means a shorter time the shorter cycle time.

50:38 it's beneficial only if quality is good The process is stable and machine availability And the component life remain within acceptable limits. That means, um... The objective is not maximum speed but best overall production result for customer and therefore we have to a good model behind. bring this parameters in limit that I want to have.

51:02 Can you evaluate what are doing there or what's idea? Idea also basically time series foundation models. So we have a multi-object target problem. Well, what means multi object? Can you explain that? for our business? That would mean... We have multiple objectives. so low cycle time higher machine availability high process stability.

51:24 there are also hard margins like the quality tolerances I talked about earlier which the stability is just looking at and that's promising area because predict or you could basically simulate different parameter sets and what effect those have on the cycle time. And objects I was talking about, You are very fascinated when it comes to time series?

51:49 It's not. we do not often talk about this topic. Do We underestimate The Topic Time Series in the whole industry a bit When Talking About AI? Definitely Yes Okay I think it's one of the time-series foundation models, is one of their enablers because as i already told back then with disability assistance was also they enabled to have that product at all.

52:12 and without that first Foundation model we trained ourselves. The product wouldn't be there. so It's about getting about opening a door for our customers. We can already deliver an added value to them Because we have running model which works. And so the next step would be when we finally have to put in a door together.

52:36 Customer data and to basically find you in these models too for improved the performance. Do your inform? Your customers? that's more management strategic decision, do you inform your customer? so listen this is AI driven. we have new technology. This is a traditional model. This is more foundation mall.

52:55 are they interested In this technology? Is it part of a discussion when selling a machine or is It? I'm not interested. i've only interested in thee. any ass Good question. Depends on the customer, we have knowledge of customers and they are very interested in. is it AI enabled or standard? And what kind of an AI?

53:18 there I'm interested. how are the limits of the AI limited? In future will also be having a security discussion with the customers in AI. so their borders our limit And other customers, it's more okay. They doesn't care? Yeah Okay So that's interesting right? so It's a broader brand bench when it comes to AI because in some industries I see oh we don't want To get in touch with AI but you already have customers and say yes is their AI driven approach?

53:53 Do You Have an AI Approach In Your Machine? i think That's A Good Sign For The Whole for the whole community when it comes to industrial AI, right? Four years ago there was a site discussion where we had and It wasn't very interesting site discussion because you mentioned project. You are running with guys from Insa in Strasbourg University with Danny And There is topic Patents.

54:18 and do research with AI patent research. Can you do. short recap How AI now changed the whole thing when it comes to patents and research? I think. how is AI used in R&D, that's a question. Yeah! We do not have to talk about Claude or Kitapkit lab... That's clear. we use that also but we have a good connection to the university in Strasbourg and that's the idea from Dennis Cavallucci, professor at Dennis.

54:50 Dennis? Not Dennis, Danny Dennis here exactly! He is a guy who worked on. it was a combination of Dritz methods... ...and he had an idea to extract information about patents for bringing them together. with this step used models of LLMs that we got steps forward now. And it means, um... We have now a tool which is possible to link together the information of the patents uh.. Which are not so readable and make better connections.

55:21 Patents contain extensive information about technical solutions and mechanisms and also their development history but then knowledge distributed across many documents. So its' not easy. very hard extract that. And also, is it in a specialized language? It's not easy to read but AI can analyze this information semantically and instead of extracting keywords the engineers now could search for technical problems on functional principles or similar solutions.

55:54 approaches means from another domain. For example you have look at the domain medical for example find version for physical problem Mm-hmm. So you are already running agents that will scroll through all the patents on the market, or how do handle that in R&D? Is it really agent-agentic driven approach ?

56:16 Or is it more ... That's the future because You can run Agents right and say please every two weeks check all the Patents engineering in the field of medical and feel off. I don't know what this could be a way to go, right? It's not actually done. here at our work we look weekly up there new patents but it normally its problem driven.

56:42 that means so We have a problem or customers have a pop has a problem And so we look inside of the possible solutions about this tool and is not screening The actual patent market. Do you use these fantastic tools also to generate innovations or help you scope something and define the scope of work, defined projects.

57:12 Define new approaches? How do you use GenAI in the whole R&D field? I think AI can support creativity by finding connections as already explained. Support creativity, that's interesting okay? And also identifying related solutions and making knowledge more accessible. innovation is not simply the automated combination of their documents.

57:36 engineers still must understand customer needs. first step define problem evaluate feasibility take responsibility for result means the AI experts Expands the engineer search and solutions space. And that's idea behind this. so what about tokens? So it's a huge topic for you to write. burn tokens? Yeah, absolutely yes.

57:59 I think That question of the future. The companies they spent billions of dollars in the development off the models end Must be now reading They want to earn something at the end right? yeah i'm not sure What is going on next year Exactly. Yeah, interesting do you run your own models on a private environment or use existing providers like open AI or Claude?

58:28 Or Do You Use Maybe A Quen or Something Like This? So Mainly We Use GitHub Copilot. so basically all of the Models Are Available to Us but we Also are Experimenting with Local Models. good old for the ninety outtakes. Yeah, yeah but it's a token topic right? That is a huge topic of our developers because they yearn for tokens!

58:56 And I think its also KPI Topic Because nobody has KPIs when it comes to Tokens Output Engineering Output Token Output. What could be done by the engineer alone without the system or without a co-pilot. We have no KPIs for token generalized, token engineering stuff right? So I think it must be a discussion because at the end you have to pay your bill and is efficient or not efficient and nobody has any feeling.

59:27 was that now efficient of using it. Really okay how do we measure this? Good question! the usage of different AI tools. The guys have to describe what they've done and how many tokens are needed for that, also explain how much short-term time we had or result. was it good results? We collected the data and then that was a good base to discuss about new token system for our IT, R&D department with our financial department.

1:00:10 Okay what was your result? More tokens more budget on tokens. I will not share it with you. Okay so we have a topic at Arbok. i would say So let's do an quick outlook. What is on your AI agenda Stefan, what are you looking into in the next? let's say three to six months? As I said through time series foundation models will be still a huge topic.

1:00:37 In the next month we also have a project now going on with NXAI. We're looking into... i'm familiar with the company. yes yeah With that where were looking at multi-object optimization and um Also we're looking mode. What about robotics, Werner? You're also using robots next to your machine as a part of the process.

1:01:05 A lot is happening when it comes to robots. Do you see there are some topics on the horizon? or do say oh no we have classical automation that works pretty well. We don't have any topic here. You know, we are in discussion with Professor Stratas-Gavis of the University of Amsterdam. We have an idea to use a world model there.

1:01:30 that is something we are discussing and also work here at local university together maybe with Marco Hugo who has some ideas. So we work on that. That's some of the new topics, yes? Perfect! We will record an episode next year in Robotics Topics, Werner together with Stratis... ...that could be an option maybe.

1:01:50 so at the end I have one more question because you presented a lot interesting use cases and approaches ideas. We need to talk about money. We talked about tokens, we're talking about budgets. How do you measure whether an AI project has succeeded? I think industrial AI succeeds when it delivers reliable and measurable operational benefits.

1:02:14 So we are not yet there. Depending on the use case that they indicate does include the reduced setup time That we had already discussed. Fewer rejected parts. so here also solutions. More stable production Also discussed. Shorter troubleshooting time. we have discussed, think of the maintenance time.

1:02:36 Improved machine availability and better maintenance planning also reduced specialist workload which is constant faster access to technology knowledge. so all this and A plus for AI We already addressed. not a solution but working on that. Do you need new KPIs? AI or is it still the old stuff OEE maybe?

1:03:04 I think a big part of that would be adoption. So how many, so we have to same topic inside development department. How many developers are actually using AI and they're using Agendic in? the same goes for our customers too. That's I think a good way to measure it. Yeah, absolutely okay. thank you very much guys!

1:03:34 It was a pleasure to talk to you. greetings to the Black Forest. all the best for the Fakuma. i think you will have a fantastic show there present. uh The latest and greatest when it comes to AI in injection molding. thanks a lot and greetings To the black forest bye-bye.