Cemtech Live Webinar: Digitalisation in the Modern Cement Plant

Hello, and welcome to the Cemtech webinar. it's great to be back back in the UK after our trip out to Riyad. my name's Thomas Armstrong. I'm managing editor of International Cement Review. and it's a pleasure to host again, this monthly webinar. Here's the, the schedule for the year and we're, we're already onto the, the, the second month. if you haven't been to our webinars before then I think you'll, you'll in for a treat. you've have excellent speakers and some time to really go deep into a into a topic. and today it's gonna be digitalization Cemtech. Live webinars are organized by International cement review, and we've been serving the cement industry for 35 years.

bringing the best in market intelligence technology reports and, and everything about the cement industry into your hands through our monthly publication, ICR and also, of course, online. if you're here watching this web webinar, then this publication is for you. it's full of every aspect of the industry that you, you can think of but with a big emphasis on technology and with dedicated issues to digitalization as well. So that's icr r just go online to www.senet.com/subscribe. you can subscribe and get instant access, and you'll also have access to our archive going back many years. as I said, we've just come back from CEC Middle East in Riyadh a fantastic event.

300 or so delegates from 35 Nations joined us for that two day conference and exhibition. many of you on the panel today, and your companies were represented. also many of you tuning in today. I know were in attendance, so thank you for coming. And here's some news about our next event. we'll be in Bangkok on the 14th to the 17th of June for our Asia meeting sponsored by SCG cement. And it's gonna be another fantastic meeting. We're very excited. And it's always a pleasure to be in Asia in the heart of the huge cement industries that are, are, are in the region. So put that date in your diary if you if you can. and now onto our discussions for today. I won't take too much time.

We can get straight onto the presentations. There's a lot to get through, but really excited to be able to have such excellent companies here today to, to share their cutting edge technologies. digitalization is more than ever at the, at the heart of cement plants in, in every aspect. we're gonna hear how it's helping in logistics how it's helping, bringing costs down, energy costs down for, for forecasting, for real time optimization really it's quite amazing how many areas are now being improved, optimized in our sector. and we're gonna start now with Fuller Technologies of course, previously known as FL Schmidt. and I'm delighted to welcome our first speaker.

Dinish Sampath Dinish brings nearly two decades of global experience spanning cement, process engineering, advanced process control, and digital transformation. During his career at FLS, now Fuller he's held diverse roles across process design, field services, and a PC delivery, combining deep technical expertise with commercial and strategic leadership. today Dinesh serves as global product manager for ECS process expert for his flagship a PC platform where he sets the product direction and roadmap strengthening digital optimization. I think that's a, it's gonna be a great way to start our session. Dinesh, if you'd like to share your slides. and the floor is yours. Yes, do that here.

Yeah, I hope you can see my screen. That's perfect. Please carry on. Yeah. Thank you, Thomas. Thank you very much for such a nice introduction. and hi, all. I thank you for joining. I really welcome you for this session today. I'll take you through the easiest process expert in terms of the evolution being it an industry benchmark for the advanced process control, and then how we try to add and enhance optimization layer, adding the A based sub sensors to the, to the, to the ECS process export. So we'll see it in two important steps here. One, our self adaptive controller, which is already delivering the best for the real time optimization as and, and, and forms as a new foundation.

And then we will, I'll jump into the partnership, which is the most exciting thing that, that we are doing now to enhance the capabilities of the optimization. so first I'll take you through the evolution of the process expert, which we have which is a rich legacy for many tickets. And then following that, we'll talk about the challenges in the advanced process control. I believe without addressing the challenges, I don't think we can have the right solution on the table and into the, into the self adaptive control technology, what it has delivered to the market until now with the proven deployment across the worldwide.

And finally we'll finish up with the, with the complementing the, the, the advancement in the partnership in terms of partner in terms of how the PXP is getting value added with the A based substances through first thing first. So here it starts with with more than 40 years of legacy that we introduced PXP during 1980s as the first firstly control in the in cement industry transitioning into 1990s in the PC platform. And having a, having a value creation with the version seven, which had the possibilities of model controller, which is still being used in that once process control space during 2004, 2011 had been a big step in terms of delivering industry specific object.

Meaning we made an application very tailor made to the cement industry that built more confidence in the end users, especially the operator, that increased the transparency of using the application. So, so that, that got a lot of adoption and as, as has been seen as one of the user-friendly application with, with, with the V eight version in 2011. So we always strive for implementing the latest technologies that can be, you know solving the real challenges that we'll be speaking in the coming slides.

And one of it, a big step change that we did was in 2024 with the version 9.1, where we introduced the self adaptive control technology that is now seen as a industry leader in terms of advanced process control technologies. So now onto the challenges in the a PC advanced process control instrument man manufacturing specific to cement manufacturing, not all the, not all the cement man manufacturers really gets the benefit, the full benefit of a PC.

And we need to understand why instrumentation and data reliability any good, even a best advanced process control with the, with all the, you know, the technology that it has to, should really need the best foundation is to have the right instrumentation and the reliable data to feed in. And then the process variability is something inevitable as we grow and mature in the pro, in the, in the, in the industry, because we have different KPIs based on different regions, which we need to satisfy, so that, that makes the process more non-linear.

So the, the controller has to be that one process controller has to be more robust to take care of the process variability and non-linearity that comes along with the process. And then the most key thing is the adoption for the operator. Operator. If he thinks it's a black box, then he is gonna war, right? And kind of do a manual control. So the transparency and the trust makes the operator to utilize the advanced process control much in, in a longer sustainable way. And the other key aspect is about the ownership. So the lack of ownership is one of unfortunate situation.

The success of the advanced process control is, is based on a mutual engagement where the supplier, and also this, even manufacturers come together and not see advanced process control as a, as a one time project where we get that you, you deliver the benefits maybe, you know, once you commission within three to six months and everybody's happy. And then we don't see it as a system in a long term where we have to set the KPIs the relevant KPIs to, to periodically measure. And that creates an accountability to the end user. So, so that lack, that, that level of commitment and engagement is really required in a long-term success. And finally, the performance longevity.

Any, any kind of a best a PC will still have a degradation for various reasons. It could be process, it could be controlled by itself, but we need to address this at the right time so we have the sustainable benefits in a longer duration. So all these parameters, all these challenges makes makes a success of the real work coming. All these challenges makes a success of the real advanced process control technology. And we from Fuller in the ECS process expert try to focus on these challenges and provide solutions accordingly. And that is how we evolve our product and keep it very relevant to address the market. And that brings to the self adaptive controller.

So back in 2024, we would've shared about how we are introducing this new technology and so on. Now, it is more about what it has delivered to the market with the global installation that we had, the self adaptive controller, as it says, it can adjust to the changing process dynamics based on the rate predict control technology, so it can self-learn. So it can predict the rate of change of the variable. By predicting the rate of change of the variable, it can act proactively. So since we are acting proactively, then it brings the stability, no time avoiding the war shooting of the, of the measurements or the variables, basically.

So this increase in stability and kind of adaptive nature based on rate of prediction, makes it outperformed compared to a fixed model control technologies. When I say fixed model control technologies, more like MPCs and other old technologies with fuzzy and all. So this is modelless controller not any static model based controller. So that delivers a sustainable benefits. And what makes it to stand out, makes it an outperforming controller in terms of the technology or as I told the, the, the stability, it increases the stability due by the nature of, of its design to adjust to the process dynamics. And then the most important aspect is with with being a single whole controller.

So we, we, it can, the one single controller can handle the whole process which, which, which means we don't have to really have multiple controllers tuned for multiple scenarios. We know the process has any, any operations has different scenarios. So we cannot wait for the scenario to happen so that we tune a controller for a normal conditions and upset conditions and different, different conditions. So this is more intelligent and smart, that own controller can be smart enough to be aggressive and also to be sluggish or steady state, whatever it is required for the process. So we don't have to spend a lot of time tuning for the different conditions and and delivers stable operations.

So this leads to a path about high utilization, high utilization, if you ask me that. Is the foundation or any, any justification on the KPAs? So we talk a lot about the benefits. We talk a lot about the improvements on and so on. But it, it'll be justified only we have a very high utilization, even the announcements that I'm gonna talk about, adding a substances to the, to the layer, and then trying to add more trying to increase the optimization even for that, the foundation is high utilization. So we should have a high utilized, high utilized advanced forces control to add anything to that to improve it further.

And this happens because the controller by nature is, is, can handle the constraints so very well that it is more flexible in handling the constraints, which can bring back an upset condition to a steady state, a stable condition in no time because of the red predictive actions. And then this leads, because of the superiority of the technology itself. We have a higher consistency and scalability in implementation and the, and then indirect way that we have a lower load in the, the, the system. So, so overall, I would say we could see from the last one and a half years of our implementation that we saw a very robust control actions that made the utilization much higher.

And that has become a benchmark in the industry now. So what we see in a standard A PC and what, and what this self adaptive controller has enhanced, as you could see from on the, on the KPIs from a two to 5% of benefits in terms of energy consumption and the productivity, we could move to to 47% of the benefits and relative improvements in terms of CO2 and alternate fuel usage. Of course, these numbers will vary from plant to plant, but you must remember that the trend, what we are talk talking, is the trend. The trend is proven that we could see that it can outperform a standard a PC with the model based control technology.

And now coming to the benchmarking here, so the case stories we had a real world case stories. I mean, we, in fact, we had more feedback and case stories from customers. considering the time, I, I don't, I cannot, I cannot present all of it. So we just picked two of these, which we already published in, in our forum in the webpage. one of it is the GCC Mexico here. It is an interesting story where the customer had to replace non A PC non PXP advent process control to A PXP looking at the technology that they were seeing it as a very positive thing. And then they could see the benefit, as you could see in in the trend, I'm sorry, in terms of capacity.

As you could see, there's an improvement of 4, 4, 4 0.2%. And also on the energy consumption, the power consumption and heat consumption also as well. And the important thing is we could see there is an improvement in terms of the alternate fuel usage. That is one of the key things which we see in all of the plants that we implement being a focus on the green fuel usage. And with also improvement in the product quality here. it's being a k it is a, it's a failure, standard deviation for the clinker. So this is one of we could see the case stories already available. you know, with page, another interesting story. This is one which we started the customer, which, which had much interest in.

And, and, and being an early adopter, I'm, I'm talking back in 2024, where they were using our PXP 8.5, the version before the 9.1 for say six to one, six months to one year. But seeing the, the, the, the, the advancement, the technology, they wanted to upgrade to nine one. So there, you could see from the manual, there was an improvement even with the 8.5 with the, with a fixed model based standard A PC approach. We had a benefit of about three to 4%. And then upgrading it to the nine one with the self adaptive control technology there was another three to 4% summing up to summing up into around 78% of total benefit.

and, and customer are very happy, as you could see from the acknowledgement that we received here. So this shows a clear difference of how a manual to a standard A PC, and then from a standard to the self adaptive control technology here. So now, what have we delivered from July, 2024 with the implementation on this? The, the, the, the controller, a strong controller that I've been explaining till now. So we have sold more than 140 applications now, and we have executed around 35 applications or 35 applications now for around 40 applications under execution. And we have plans for the 65 plus applications in the coming period.

It could extend even 20, 27, of course, and we could see the split across the globe worldwide implementation. And all of these 35 applications are running with a high utilization and a happy customer. So, yeah, so now we talked about a very strong foundation, a proven technology, and and a controller that is robust. Robust. So while we have that now, we are try, we just address the measure of the problems, challenges that I've been explaining. So now we have to turn into the other constraints that is towards having the critical data on the right, on right time to enable, to enhance optimization. And for this reason, we have to, we have decided to we made a strategic partnership with ibit.

So what are we delivering here? So we, we are integrating the P XP nine one with the, with IBIT based, the AI based substances from ibit. we call it a best combination because Fuller ECS process expert comes with a deep process knowledge and control knowledge over decades. And now we are adding it with ibit, a robust platform with a high reliable and high accuracy substances. And this combination makes it an optimal closed looping with this optimal closed looping. We create a value to the end customer by filling the gaps in the spaces that is really required for the A PC to enhance its optimization. So now, again, I will go back to the challenges. I always like this challenges.

So we, we understand what we are bringing and what we are solving here, and how it is really helping to the end, end customer. So the current situation is we have an advanced process control with the quality data that is coming from the lab here. We talk about specifically about free lam, C3 s, free Lam, or CS for the kill, and blame for the cement mill. And then also from the process side, we have inputs from killing it oxygen, 'cause the killing it environment. We know environment is more harsh. I mean, in the sense it's very tough with the sticky dust and also high temperature, which, which could lead to outages and maybe unavailability of the signals. So this is the present situation.

And, but what happens because of this, we are left with blind spots. So basically we, we, we have limitations in the optimization, even the best, the technology, the best optimization engine that I explained will have the blind spots, so that we call it a lost opportunity because it'll be more reactive that you have to wait for a lap signal, say for two hours, then you have to adjust the control actions, or you have to, you have to run a kill probably for some, some, some period without a oxygen signal due to the maintenance issue. So what we do, we add the, the substances from ibit as a complementary solution to the, to the PXP, both for, for example, in the lab.

And I mean for, for helping out on the clinker freelance C3 s or the blame in terms of quality or in terms of the killing oxygen. And the outcome is we have a high frequent data high frequent, high accurate, and reliable prediction. for the, for the quality sensor free lam CT threes the blame, and that gives a quick response to the process. So now the advanced process control has deeper insights of what is happening in the process based on the predictions coming from the a b soft sensors. And that can ensure that the actions are more proactive.

So basically, we can adapt the target much frequently than to wait for a lab data, which is, which is two hours, or even later four hours, or even further. So, so this, this makes it more proactive and kind of have the control actions well in advance. And for the cases like where we have an outage, maybe a less availability of oxygen due to other reasons, now we have a continuous data coming from the predictions. And that can really add an input to the process optimization, which can optimize the kill, much better increase improve the energy efficiency of the plant and the product quality as well.

So from the limited limitation through by the blind spots of missing these critical datas on the right time, we move towards a greater visibility where we have chances to maximize the opportunity, we reduce the gap and we maximize the opportunity space. Yeah. So this is, this is a logical approach, which makes a lot of sense. And in terms of our enhancing the optimization with what we have now. And then we are now have to talk about the real experience that we had on site where we did a trial in in GCPV Gios in, in Spain with a, with a freelance soft sensor, integrated closed loop into PXP. And what we achieved an improvement in the product quality in terms of reducing the off spec linker.

That is a lot of improvement and also improvement in energy efficiency. So we have a quick, it is, it is a trend to show you the worldview to understand how, how it was with the, with the real time experience on site. So the brow line, what you see is a predicted free line, which is at every five minutes interval. And then we have the lab prelim, which is a two hours of interval. And then we have the talk target which is now adapted every 15 minutes. So previously assume without the prediction of free prelim, we have the target. So here, the primary parameter to measure the K process or the kill status is the talk of the kill.

So it, it, we previously, we, it used to be adapted for the two hours based on the lab data. Now we are able to adapt based on 15 minutes, that is eight times, at least within the two hours, based on the free and prediction. So if you see the accuracy and the directionality of the free lamp, we could see, for example, we could already sense that well in advance that we get a deeper insight that the product quality, what is the, what is the product quality that you're producing already in the K. So by knowing that the product quality is changing continuously, we are also changing the target of the primary measurement.

And because of that, the stability has increased, and because of the increase in the stability, we are able to improve the product quality and also reduce the energy consumption improve the energy energy efficiency of the plant. So in that way, it was a very positive outcome. And both in terms of the, the soft, soft sensor ability and the reliability also in terms of the closed looping and the overall benefit that we could, which we aimed at first place in this combination of the PXP and the soft sensors, and now into the, the, the structures, the product structure of overall the layers here in the automation portfolio.

So it starts with the advanced sensors that goes through the the process control, gas and sampling, and then to the optimization layer. So what we do here is the layers are the same, and we augment and complement these layers with adding a soft sensor that is in close loop with the PXP to get the improved enhanced benefits. So this means that we still rely on all the physical sensors and and soft sensor act as a complementary to the physical sensor, and creates value by, by integrating into the process expert to, to have a proactive approach in terms of adapting the target much in advance and kind of improving the performance.

So, so this is how the overall structure in terms of the adding the soft sensor to the PXP looks like. And then we move on to the value on the performance which is very important in terms of the different layers that we can, we can imagine in, in the, in the optimization world. First thing, of course, the basic thing, we have a manual com combination of PID. It is very limited, high variable, and and has more scope to improve. And then it comes to standard a PC with, with the model based control technologies and so on, where we could expect around you know, two to 4% of benefits depending on the site.

And then we have our strong foundation that is the process expert, nine one with the self adaptive controller which can add another no two to 3% to make it four to 7% of improved benefit compared to a standard A PC. And now in combination of this strong optimization engine, plus the soft sensor can add another one to 2% to make it up I to 9% or improvement. So these are the steps.

So we have to see these in the steps on how we come take it through from an standard A PC to to the next level of superior control technology, and then enhance the performance by adding soft, soft sensor to it to create a better visibility and deeper insights through the AI based predictions so that we can, we are trying to fill the gaps that that is in, in, in existence. So, so that is the overall narrative that I wanna give here. And that brings me to the end of this slide. Thank you very much for listening to me. I'll be happy to answer you guys on anything regarding the strategy of implementation or the soft sensor integration and so on. Thank you very much.

Thank you very much, Dinesh a really intriguing presentation. there, there are a few questions yeah, that, that people are asking. one, one that I would just like to to start with is you opened up with a slide and you, you said, what, what are the, what, why are these benefits not being fully realized? yeah for, from the different software and, and, and two of the reasons were seem to be kind of human. It was operated trust and adoption, and it was lack of sustained ownership in government governance. and this was something that we talked a bit about at Emec, in Riyadh, okay.

there seems to be a lot of importance on these, these on, on, on ownership of these systems once they're installed in the plant for them to be truly effective, they, they the human element is actually still very important. They're not just plug and play, you have to have people buying into them and knowing how to operate them. How do you approach that that challenge? So yeah, this, this is a very crucial thing because for the success of the A PC, the, the trust and the buy-in from the human onsite is very important.

So the approach here should be like it that the A PC should bring a lot of transparency that, that the operator must feel that he's able to understand what is the action that it is taking and why is that taking, and he is able to clearly see those things. So, so bringing transparency and having a, a UI that he doesn't feel like it's a black box, that he doesn't feel like it doesn't know what it is doing. So then it brings a buy-in that he is able to understand, okay, I know why it is happening, I know why this action is happening. So that level of transparency and trust can really increase the engagement of the operator. And that is what we've been trying thriving for.

Like I said, from the previous you know, we, where we, we completely changed the application because of the reason that the operator should be able to clearly see what is happening within the A PC, if not at the back, I mean, if not the technology and in the backend. So you should be able to at least visualize and see what is going to happen, what has happened, and why it is happening, and what is going with the extractions. So the transparency and the trust is what is going to improve it. And we are, we have worked a lot on it, and then we could see that that could really help bring in the buy-in from the operators. Okay. Okay.

So buy-in and then onboarding the technology with the right people is, is really important. a couple of related questions. What is the most important input information required by the system in order to provide the best results? and alongside that, I would like to ask this other question. For advanced automation and control of horizontal ball mills, which process data do you primarily use or prefer? Do your control strategies rely mainly on temperature readings, pressures, power draw and feeding scales, or do you also integrate additional instrument instrumentation such as vibration sensors or acoustic monitoring to monitor chambers?

So just look if you could just talk around the the sensors that in the inputs. another question is is around the soft sensor working principle. So I think there's a few few questions that come together there. Yeah, so, so we must understand very clearly that the, the dependency is heavy on our reliable signals, the reliable instrumentations, because if you don't have the reliable instrumentation, and we cannot have a reliable soft sensor at first place. So when I say reliable instrumentations, there are key instruments, key signals that is required, for example, in a kiln process and also in a, in a ball mill process, for example, when you asked.

But I think the list, we, I, I don't think we can talk about the retail list of the parameters that is required, because it is, it can be 10, can be 15, or it can be even more than that. But there are certain critical thing, like for example, in a kill, if you, if you have to have a burning zone temperature, you have to have a killing oxygen, or if you have to have a dark measurements. So these are very clear, very crucial, even to understand the calci temperature and so on. Like in a ball mill vibration might not be crucial, but the ball mill power and you know, the elevator load and there are other parameters that could be very crucial here.

So, so we, how the approach is like we, we will get all the required data that the required data set from a plant, and we'll try to see which resonates or which is more relevant or more key in the prediction. And of course, it is just not, like I said, the, the, the, the whole idea is just not going by the data itself, trying to understand it. Does it make sense? Because it must be a crucial data or the instrument or the signal that reflects the process, just not, that is something which is having a good relation just because it is it is, it is a you know, only data driven thing.

So, so we identify such parameters, which is more relevant and reflecting, and then we build a model and and kind of deploy it and train it. So in that way, the crucial instrumentation is very important for both kill and the mill to name it, I, as I said in the mill, it in the kill, it is more towards the burning zone conditions, and in the, in the mill, it's on the filling percentages and the kilowatt and the elevator power and so on. So these are very crucial to, to build the model and to see that relationship and the reliability of it. Mm-hmm. Okay. Thank you Danesh. And and thanks to John Klein and Jan Lec for those questions.

there are some more in there in the q and a if you'd like to look at those after the presentation Danesh, but for now, that's all we've got to for Thank you. Thank you very much. Sure. Thank you very much. And I would appreciate if somebody wants more, they can also write to me to the email address that you see in the screen. Sure. Okay. Thank you, Thomas. Well, that's got us off to a great start. thank you. And we're gonna move now onto our second presentation. please Francisco Iglesias share your slides. Francisco is he's calling in from France.

it's nano like the company and Francisco's a, a civil engineer with degrees from Argentina and France, and completed his executive training in data analytics and artificial intelligence, at mean Paris PSL he specializes in cement logistics, industrial IO ot, and data-driven operational improvement. his work spans Europe, Latin America, and the ME MENA regions supporting cement producers in deploying modern sensing technologies and predictive replenishment models. it will all become clear at man alike or Silo Connect their product.

he leads international initiatives focused on silo level digitalization and AI enabled vendor managed inventory helping organizations strengthen logistics reliably and overall supply chain efficiency. So a very interesting product here that we've had featured at some of our Cemtech conferences. But please over to you Francisco, for more details. Hi, well, hello everybody, and thank you Tam, much for the introduction and to send a review for the invitation and for all of you that you are really numerous people attending today. So thank you a lot. I'm Francisco Lecia in general.

Unlike today I will share with you how leaders in the cement industry are already using AI to make logistic a performance advantage that unlock new sales opportunities more precisely by connecting ReadyMix cement silos directly to the seven dispatch. unlike it's a French company that has more than 5,000 silos installed with all the leaders of the cement industry over the world. and when we think about innovation in cement we often picture kills and meals and automation at the plant. Today, I want to focus on something else. Today, I want to focus on something more invisible, yet just has critical, which is the downstream logistic.

Everything that happens once that your cement that you produce left your site. And because even in the best cement in the board, the lucit value, if it cannot be delivered on time in the right quantity and without cost re disruptions. So get work has a cost and it's massive. We can see everywhere that, depending on the country and depending on the company itself, we see different challenges that are cement. Could be cement shortage at the ReadyMix plant, or slow stop at concrete production, lack of drivers, high costs, and the leads gold zones. But all these challenges highlights one blind spot and only one, which is the semi logistics. And it is the same inventory at the concrete plant.

Today, we do not have visibility of what is happening at the, at your customer concrete plant. and that's creates all those, all those challenges. currently each concrete plant operate its own ary train system, but none of them share this data with the logistics teams or the sales teams in your cement company. every group needs this information to ensure a smooth operation, but basically you wait, or you guess what it is in the silos, in inventory of all of your clients, you have to wait for a customer to call or place an order in order to know what are their reality at x at the moment.

S so you operate reactively rather than proactively all the time, and that's takes you away some opportunities, and that takes time and profit are lost. So we are talking about how can we fix this? Why do you, why don't you operate differently? Because basically you lack of action, actionable data to drive decision, you need data that must be reliable, real in real time, and accessible everywhere. If you do not have this, the data is not usable in the industry. The data is in silos, like figurative and physical silos. So you, we have a lot of manual entry. We have a lot of batching issue approximation with your batching dispatch system that the concrete plant used today.

Some safety assets related to the physical silo checks that are risky and emphasized the lay data that maybe you have someone working in the concrete plant and then has to call someone at their management, and then someone who calls to the cement plant in order to have the cement or, or to place an order. All that noise is not ideal. So what if you could stop fi flying blind, because really today we didn't, you do not have this video. What's happening in your market at, at the end, they are their, your customers. I know you want to know what's happening to them. So I just want you to imagine what it will be.

If you can prioritize version deliveries at critical site, what it will be, your logistics every day, if you can predict and prevent customer stockout what it will be. If you can negotiate smarter transport contracts with all the data that you have, or ensure 100% production uptime for your clients, or protect key accounts despite constrained supply. So today I want to show you how to stop reacting and start leading and take total control at your cement to supply chain, okay?

so it's possible with the actual technology we have today, but not with data whatever the technology we're talking about, radar, laser ultrasonic, I know that here, there, these people who has years and decades of experience that saw everything coming and passing by, and in theory, they are really accurate, but in practice, its technologies are intrusive because they're inside the silo. They're really hard to deploy on larger scale. They are expensive, and they require a lot of maintenance. Why? Because they're installed inside the silo, which is a really dusty environment, which is really hard to, to, to work on. So on top of that, the, all the data stays local.

these systems don't provide good data, okay? you don't get, you don't get global visibility. You don't get any dashboard to think about the supply needs and when it exists normally it's hard to use and not user friendly. So how, how can, how can we solve this? Imagine that you have a, a new AI system, and we will see how it works. That will make your silos talk. The concept is just this, make every silo in every and every one of your client plans, concrete plants talk to you and tell you, Hey, this is my state day. This is vaguely what we are achieving. we will, you will get 24% visibility at your ready mix.

Seven silos, silos you will get remote access wherever you are, you will have great accuracy, zero of physical maintenance, which is key for and a smooth development. And you will get actionable insight from field data all the time. So basically, the technology, thanks to the AI exists to give you full operational visibility with zero impact on your customer daily routines. And you'll be asking, well, okay, super nice, but how it works let me get into that. The system is quite simple, but complex at the same time. basically it's based on an extra engage. it's a proven technology that has been used for nearly a century.

We all know extra gauge that are going to measure the compression in, in this case, in one leg of each of the silos. we have refined and patent one of those. we measure the compression of the silo. So the more s there is, the more that I compress, the less s there is, the more it relaxes quite simple, right? We measure the mechanical def information of the leg, of the silo and little, and also temperature. And I will, and you will see why. so we have a patented solution that this is going to process this data, okay? this data is going to be processed, but a patented AI algorithm that are going to turn raw def formation data into reliable insight, to be honest with you.

and string gauge, you know, you can buy it in whatever you want for a couple of pennies. our value is not in the, in the string gauge. It's really in all the process we have to make to keep you with this roughly, really roughly signal, a really clear data about the level of your silo. so how we make it, we make it because we train one AI model per silo. So there is no generic behavior. This AI model per silo is going to learn from each silo, structural and thermal profile. And it's going to continuous signal, doing some continuous signal processing that are, that are going to drift correction and do a lot of anomaly filtering in order to give you a really clear data.

so, but at the end, we will show all of this once the data, this process is going to be displayed in a platform that is designed for you for some logistic, for cement sales with features that are tailored to your needs. you will have real time field levels. You will have quantity of free space inside the silo, day by day consumption and history average and maximum consumption. You will be able to anticipate and evaluate remaining stock duration of your customers. you can make all the adjustments you want with us. it's super user friendly. any training is required to be used is really easy to use and is you can get, for example, automatic and customizable alerts in order to, to get notified.

So even sometimes you don't even have to go into the platform. okay, the system itself is the whole system is designed to be really, really easy to deploy. it's a system that, from his concept itself, it's assigned to be easy. So we, because why, because you have a lot of silos to monitor and to get the most of the value you need to really monitor the entire fleet, not just a couple ones. So our, our system can be installed in 30 minutes. you see drill two holes, you screw the sensor on, you connect the transmitter, and that's it. It takes half an hour, no downtime. It's non intrusive. There is no Calvin work, and that's it. It comes with this home battery that lasts at least two years.

It can last up to seven years depending on the connectivity. And that's it. So as you see, really simple simple design for mass installation, okay? One thing that is great is that basically you can choose about the level of digitalization. Normally we will see that every customer, every company, every different country has different way of working and we can adapt to all of them. normally it's always about giving visibility and turning reactive, logistic into proactive planning. So normally we like to go little by little. we learn a lot in the last eight years, we are working in this industry about how to implement this digital transformation process into big companies like yours.

We know that it's not simple. We know that trust is fundamental. So we like to go baby step by baby step. So first baby step is visibility. We just give you the platform itself. so your logistics and sales team can't know what is happening across all your silos of all your cloud customers. So that eliminate manual checks and phone calls and every goes smooth. Why? Because you have a smart guy working in logistic and sales. So if they can do their work without knowing what's happening, imagine if they know what's happening. second maybe step is out ordering. Here we out. We give you automatic deliveries that are triggered by free silo space, improving fleet and driver utilization.

we will see how it works. and the third and last step is when we put a full vendor management inventory service in place, which is means that you as a provider take their ownership of assuring that to your customer, that you will take the inventory and the stocks responsibility for them. So knowing what they are, they have in their silos, you're going to manage. So basically you promise your customer, hey, doesn't, doesn't, don't worry about your stocks, I take charge of it. we are talking about a really strong relationship with your customers. This opti, this is going to optimize a lot your seven flows. here we integrate directly with your ERP and with your end-to-end distribution also.

So everything goes out full automat from A to C. And we have a couple example of this. Please, if you have any questions, just write it down. We are going to answer it at the end of the presentation. So once you have all this in mind, basically you will see what's happening. You will imagine that you have some ReadyMix plant, internal plant, or you have a precast plant, or you have a, an external plant. Well, you can have, for example a ReadyMix internal plant where you only have the classic mode. You only see you only the only KG ability of what's happening there.

So you can customize, alert, and send the trucks, but keep working like you always did, just having visibility and being more proactive in other cases. For example, precast plant precast is really perfect for ma for making BMI. We have really nice examples of that. in, in precast, for example, you can implement a full BMI services with 100% automated orders. And where we have placed orders and tracks are sent when the site have enough face. So if a silo goes down using send the track automatically, you can do something in the middle from some, I don't know, ready mix, external plans or clients. It depends to you, it depends really on, on, on, on your behalf.

And what do you prefer and what is the relationship, the relationship with everyone of your customers? All the information is going to be visible for your logistical sales that are going to communicate really smoothly with the seven plant and with the seven transform transportation in order to really make everything easier. So where are our impact? Well, our impact are going to be in the sharing savings and the logistic departments and both your sales, and we're talking about how it works.

Basically when we're talking about savings in the logistic department you will, you will be, you will be able to generate savings and logistics, like eliminate blind deliveries through these planning and stress, prevent shortage, reduce idle time at the rhythmic plant, prioritize the right plant at the right time with zero manual checks and analyze delivery time and rate, negotiate third party contracts. Why? Because you can distribute all the deliveries throughout the day. normally we know that we have normally all our clients that call us for having their stock ready for the day. So we know that it's quite normal to have really hard morning rush.

So we like to attack that by knowing what's happening and be more proactive and preventing that. In the case of the sales, normally you will be able to sell more with the same team if all the time that takes you to manage the 80% of the processing, all the orders that are just regular orders. instead of doing that, you can spend all your time to focus on that 10, 20% of session exceptions that really make the difference. So your sales team can really sell more with the same team. You can place order even before the customer calls. You can offer a new premium service, bring your customer peace of mind and increase production. You can really be customer focused and not just in regular stuff.

You can focus on building trust and loyalty. When you ultimate orders, you free yourself from all the routine specialized in the special cases that really need your attention. So more sales with the same resources are less minor work, and that takes higher profit. So a lot of common key concerns that we always have. What do we have? First question that always came, it's expensive. Well, no, it's not expensive compared to a single waste to track run ate cost the equivalent of 20 cent of dollar per seven ton delivered. So if your price is, I don't know, $100 ton of seven delivered, well this cost 120 cents going to be your final price and you get the system paid. so it, it is cheap.

So when do you see returns? Really in a few months? It depends on, on your specific case, roughly between 4, 6, 8 months. Not longer than that. It's hard to deploy. Well, we saw it. It's really a really tiny piece of tech that you just need to screw in one leg of the silo and that's it. So no downtime and dedicated support. So another question I have quite often, will my customer accept it? Well, yes, because your customer, and they will love it if they are really stuck out sensible. For example the data is used to improve service, not to control behavior. So basically we remove frustration and deliver measurable measurable outcomes.

Thanks for our technology and making a point about how much has cost, how we arrive here, basically, it depends on the country. Of course, it's not the same divide and France and Korea and Canada than Mexico, Morocco, Greece, there are all different countries with different companies. But we see we see really a roughly same cost between 2 cents and 33 cents per ton delivered with range around 20 cents. So we just took all the clients we have from different cost, from different customer and clients. We check how much did they pay for the system itself? We did just a division meeting, one and the other. And we get to 27% of seven. So I really want to focus on cases studies.

we have some really nice cases study because if doesn't work in the field it's just, sir. So one really nice example about VMI or vendor managed at inventory is the case of all in France. they are operating a model where the seven supplier manages the stock levels directly at the client's site. so based on the real time silo data, based on the real time silo data, they they can reduce the fleet thanks to shifting deliveries from later in the day, from early in the morning to later in the day. Basically at the beginning, they had a lot of stress at the beginning of the day. after using our services, they really flat the demand curve in order to work in a more practically way.

You see all this demand that were at the before 9:00 AM now, it's going to switch to the end of the day where the fleet where the logistics is way cheaper. So you stop sending trucks when it's expensive and you start sending truck when it's cheaper and how it works. Well, pretty, you get the idea every time that the silo level drops at one of their clients below the red field threshold silo, it detects it detect that there is enough space for a refill that's notify the customer that the refill order has been placed. This is an optional thing that we made for them.

And the refill order is created directly in their a RP, so no human interaction automatically, and the track deliver a cement to the customer plant. So customers are automatically replenished it without any manufaction. We have more than 700 I think that now like 900 silos connected in this manner. and it's working quite fine. also Mexico is one use business case. I love also because I don't know if you know Titi of Mexico, but you know, the traffic is quite terrible. they had a, a problem that it was idling time. After the plant, they send the truck and the truck stuck there for hours. Why? Because customer just ask for seminar place order in order to be safe for their operation.

But they ask for a truck when they are not sure if there is enough space in their silos to unload the chart they bought. So the truck arrives, there is not enough space. The truck has to wait. Your logistics teams and company has to pay a another charge just for, because the, the, the logistic the truck there is, is useful is, is just stuck there and wasting time. So just by using our system, by having our visibility, now they can know if the customer that is asking for seven, he's actually the place to receive the load. So they only send the truck once they know that the customer can take the load. So that's why we really slam the idle time at the plant.

we have a minus 50% idle time at the plant. with that they win 24% of Trumper productivity and they win also a 10% reduction of transport cost per ton, all thing to just seeing what's happening and active acting practically. So that really cut out the transport cost. They really reduced the onsite delays. it's a kind of semi-automatic BMI with WhatsApp alert. And Juan, which is the logistic and distribution director for all in Mexico, said that what began has an innovation quickly transform in a new logistic standard. So it's not just an smart upgrade for them, it's really now the key driver for a sufficiency at scale. Another similar case is for Spain. Same case, same idea.

You have really morning rush that's happening before using our services. Now that they use them, they really can flat down the, the man curve and have a more homogeneous one. So again in this case, the man Fernandez, which is really supportive for this project said that is there is no longer a question of if we have to adopt or not. If, if they can afford to, not to, because in Europe in general, there is like the perfect storm about high peaks of demand that concatenates with a really restraint offer of logistics. In Europe, we have lack of trucks and lack of drivers. So we cannot afford to have this concentration of demand. We need to really flat flatter out the curve.

So the results are pretty clear. They cut it in half last minute orders. the average transport paying lead time reduced in 20%. with the same fleet, they are able to transport 6% more of of charge per per truck, per month. And once they install our system, they used to have some stockouts in some of their clients. in every silo that our system is installed, they have zero net zero stockouts. I asked to my data colleagues about, no, no, I cannot put zero, like real zero in a presentation. But they told me, no, no, it is real, real zero. There is any stuck out in all the silos that have installed cyonic.

So just for ending this presentation I think that it was really conceptual, a little short presentation, but I don't want to take too much time. The idea today is show you how we can go from daily pressure to operational cer certainty. we saw how for the logistic, we can win predictability that scales. We saw how for the sales, how re reliability can build trust. We saw for the customers how transparency can be achieved without disruption. So in just in one phrase to you to remember, AI here exists for create visibility. With this visibility, we create predictability, and with this predictability, we create growth for a company. So that's all I want to share today.

thank you for listening to me, and if you have any questions, Thomas or the rest of us, I'm happy to answer. Thanks very much, Francisco. Yes. a great presentation and really interesting to see the case studies and how this very simple technology is, is, has really big, big impacts once you multiply it over many silos and a bit of time. and like you said, especially in countries where there's a lot of traffic and where transportation is a premium. so thank you very much for that. there are, there are several questions I'd like to clear up.

One thing though, I think there are a few questions around whether this technology works on concrete silos, and I understand that it's it's only for metal silo structures that have legs where you can, you can attach the, the equipment. Is that correct? Exactly, Exactly. Yes. we work in the downstream logistics. So we do not work at your seven plant. We do not work with your big concrete silos. We work at the tiny 7100, 150, 200 maybe tons steel silos that are in all the concrete ready mix plant or precast plant. Yes, exactly. We work only with silos that has metal legs as we saw in this picture here. really quick, this kind of silos. Okay. Just tiny. Yeah. Steel legs, silos.

And just to, to clarify with the sensor on the legs can that, can the sensor give false readings for example, if there's any maintenance, if someone is climbing on top of the silo if there's a I guess an earthquake that's for a more extreme concern, but I mean, how, how sensitive are they and and how does it work in that respect? Okay, well, it's really sensible. We achieved to measure roughly plus five or plus or minus 5% of the total capacity of the silos. So you have a silo of 100 ton. Normally we give you that the cell has 80, you can have 85 or 75, which is a really good precision for a device that is, it is not into the silo just attached to one leg of it.

we do not see if there are anyone climbing. we do not see if there are earthquake. What I can tell you is that this device, it's important, but what is really important is our capacity to filter all the noise that the rough signal has in order to achieve a really clean signal of values. we have a lot of more than 19 algorithms that are going to running in parallel for each of the silos that are going to be learning all the artifact and mechanic deformation problem that can have we will have in this manner, like in conceptually a tiny AI brain for each silo.

so yeah, we, we are kind of accurate not to the level to know if someone is coming up, but yes, to the level to know if someone takes one 10 tons of cement. Very good. And there are still quite a few other questions. if you look in the q and a or in the chat people are asking o other things. So please take a look and, and, and answer those. that's great for now. thank you very much, Francisco. We're gonna have to move on to the next presentation, but greatly appreciate that. from Francisco, from Nano, like with the Silo Connect product. Yes, hesitate to, to just send me a WhatsApp here, whatever it needs. we keep the good discussion. Okay, thank you a lot Thomas. Okay, thanks.

Okay, so we're gonna move now on to our next presentation which is from Sami welcome Mattia Mattias KES who is the machine learning tech lead at smy where he's responsible for developing, guiding, and scaling the intelligence software, which is currently optimizing cement production in 40 cement plants across 16 countries with a PhD in computer science. And over 15 years of experience ranging from process optimization at a MD to large scale recommendation systems at Zalando, Mathias specializes in bridging the gap between the theoretical ML and robust industrial systems.

His goal is to help cement industry utilize data-driven technologies to optimize quality while saving energy and reducing CO2 emissions. Mathias, if you'd like to share your slides with us the floor is yours. Thank you so much for the introduction. Thomas, are you seeing my slides? Yes, we can see. That's Great. Thank you so much. So, hello everyone. today I want to give a bit of a different kind of presentation. So a colleague of mine in a previous edition of this was already talking about alami and also cement product or predictive product.

And today I want to give a little bit a peek behind the curtain how basically we make this kind of predictive product work in practice, how to make it basically deliver value consistently. And for that, I am gonna have the following outline. So first for all, which don't know yet, I give a short introduction about the company and what we are doing and the product. And then I will talk about three things. I will talk about how do we do reliable, reliable predictions, how we translate these predictions into actually outcome in the process, and how we then scale this whole solution to multiple plans efficiently. So let me start with TME itself.

TME was founded seven years ago by Robert Meyer and Leo, and it's basically a software for predictive quality control and its cemented concrete industry. Today we focus on the cement and the idea is to both optimize the quality and minimize energy and clinker resources and thereby make it more cost efficient and reduce CO2. Now, some numbers for the companies, were roughly 50 colleagues. We have about 40 cement plants and hun 150 concrete plants. And so far every one of the plants that we started working with also stayed with us. So now, very shortly as a context, what we are actually doing all starts for this particular use case with the cement cement mill.

And there we have production samples or spot samples, and we have shipment and composite samples. And for all of them we have a subset of this device measurements. So P, s, D for fineness or C, we have X-R-D-X-F, and particular for the shipment and the composite samples, we also have strengths. So now all of, for all of these samples, these measurements go are exported into our system. And what we are basically doing, and what also this presentation will be partly about is we predict the strengths for this particular sample. So early strengths, like two day or 28 days strengths.

and based on this prediction, so we, we say like the 28 days strength is on target or above target, for example, we then optimize we calculate basically a set point, for example, for fineness. So if we are above target, we say you should grind corer to get to to get to the target strengths again. So that's the, the, the principle. And by doing this, by basically having 28 day strings, for example, in real time and not have to wait for 28 days by, by having that you're basically able to react quicker to changes in the material or the process and thereby eventually reduce the variance of your process or the production.

And when you reduce the variance, you can save energy, you can reduce buffers, and by that then also reduce clink up ideally. Now there are a lot of parts which are going into that, and there are some called here. Like we, we always retrain our models, but some of them we actually want to get into in this presentation of what happens in the background to make this whole process work reliably and bring actually cus a business value out of that. So on a side note, this whole, the presentation I give now is kind of a, a subset of a bigger presentation, which for time reasons I cannot go on all the points here, but I'm sure we can provide the slides then on request.

But now I first want to get into the first part. So how do we actually get reliable predictions for the cement samples? Now what you see here on the left side is schematics of the pre-processing pipeline. That means each it's a cement sample before we do a prediction, goes through this whole pipeline. And there are certain kinds of transformation done there. And as you see, it's like around 40 steps of transformations, which are done even before the prediction. Actually, predictive model is, is, is done. And you might wonder, okay, that looks quite complex, what do you need this, this this complex reposing pipeline for?

And the reason is simple, like in, in, in every industrial process that's a cement plant is no different. The data are simply tricky and it can be messy. So we have missing values. They can be, for example, based on different measurement programs for different cements where some have some values and some not for cement types. There are of course measurement errors that can be differently strong and that are data drift, the measurement devices themselves maybe in the, in the process. And of course finally you have things that are not actually seen in the measurements, like a strength enhance or, or some, some reactivity, for example, of slack.

So all of these things you encounter basically by working with this 40 cement plants or they were encountered there. And for all of these kind of challenges which appear in, in, in the real, in the industry, the, the, this pipeline or this models were extended to cover them, but for all of them there were experiments done, different version tested until something was found that then is able to cope with them, these challenges. And when is this important? This is especially important when you actually want to have a closed loop. So you want to steer like automatically with this. So then you actually want to take care of these kind of issues.

And now to make this point even clearer and want to show this slide. So in general, now there is a big hype around ai. So AI like with starting with the chat GPT moment and with the belief that everything can be done now with so-called LMS with these chat bots. And so we put it to the test and the question was, what happens if you actually let a, an one of these chatbot of these AI do strengths predictions? And it turns out internally, of course, the big, the chatbot itself one to the predictions, but internally it just writes a little program, which then does this kind of predictions of strengths based on the measurement data.

And so we compared what the, this, this model, this AI would use against what we kind of handcrafted over time. And then what you see is here on the right are examples of that maybe to say, if you look from a high level, the prediction accuracy or the prediction error don't seem to be too far apart. But the pro, the, the, the the problem is that if everything if the, if the production's very stable, there has not much to predict for, you could also use a predict simply the mean of the production. but in this examples on the, on the, Sorry for that, but in the examples on the right here, you basically see the green dots are cement strengths.

The the black lines are basically predictions from our pipeline. And the yellow line are predictions from the LLM or the AI version, let's say. And what you see here is especially, I mean, when everything is kind of stable, they are roughly the same, but especially if something happens like here, then the yellow versions or the the basic version doesn't cut it. So that would, wouldn't predict the strengths and you would basically lose out or on this in the middle part you see that this basic version would, would just, there were likely some measurement errors in the basic version would simply overreact to them.

So all of these things happen to say like that a lot of experiments and handcrafting are reacting to these issues, went into this pipeline, and they're necessary to actually have a stable system that predicts successfully. Now taking even a step back to do this kind of system you first have to understand the data, it's always about if you have bad data in there will be bad predictions out. And so here, similar to this pipelines that we created over time, we create the tooling and with experience a different views on the data that then show you exactly relevant patterns or problems and, and make this very efficient to spot them.

And so one example is like we have a hundred page report, which we can create for every plant and which we usually look at, especially in the beginning to see like are there availability issues between different product and pro product and shipment samples, for example, or composite samples that some measurements are there and some not. There is looking at drifts in some of the data or inconsistencies, for example. And this is of course the the starting point for everything else to then also with this kind of insights improve the models. Now kind of a subset, but it's not exactly the same is what are you doing if there are ob observed problems happening?

And so one thing is clear these plans and then that the context always changing over time, the process changing over time and there will likely never be the perfect predictions. And there can be multiple reasons for it. So there can be normal noise. So we have measurement errors on the, on the strengths measurement itself. So even if we would predict the exact right strengths, the real strengths, the measurement error on the strengths would lead to some prediction error. But there can of course also be model issues. So there can be something wrong there actually that we have to take care of. And there can be issues with the plant itself.

So for example there's a new person who does like a strength test and they're less experienced and suddenly the strength tests are a little bit of, or there's a calibration, or the samples are the sampling is done differently. So all of these things can happen and the challenge is to, if something seems off, to identify the, the, the right cause and then act upon it. And so to do that, you need at the same time a deep understanding of the plant and it's a met knowledge. And for that we have special colleagues which called customer success, which work closely with each plant to then have to bring this knowledge.

You have to understand the machine learning where the engineers like me are the experts. And to make this efficient, you have to have tooling which shows you where are actually problems and what might be the cost for them. So now that we talked about like how we get reliable predictions in the first place and how they should stay reliable by managing problems and reacting to them. The second part is how to translate these predictions actually into changes in the process. So the steering of it. And there, there's two aspects to it I want to go into, into this presentation.

And the first one is like, first of all, there are all different kinds of setups in, in every blend in terms of what you want to steer, for example, there are different strengths. so there some, for some cements it's like a two day strength is important for some 28 day strengths or some kind of combination of those. We can steer with production of spot samples of the composite samples. There are restrictions that are important for different cements of plants like plane that have to be met. And of course you can steer with finances or recipe or different control parameters. And all of these combinations have to be met.

And so you need a platform basically, which allows for all of these combination to be configured. And having that then you actually start to set up and let's say successful steering. And what does it need to do? It needs to handle outliers for measurement. for example, measurement errors. It needs to handle campaigns so that there are different batch sizes where the start and end samples might not be representative because they come from from from influence from a previous campaign, for example. So you have to handle that.

And all of this you have to do so that of course you stay in all the parameters within the boundaries, but you also react basically fast enough, but you don't overreact and you eventually go into some kind of rollercoaster what you also want to prevent in this dynamics of the steering. And so now that's a tricky thing to do. So there's also internal tooling for us where we can play around of different kind of parameters to adjust the steering and, and find the optimal spot for the specific plant to, to be successful. Now, that was very short on two aspects of the steering.

And finally now if you have the, the predictions and you steer and that works for one plant, there's a whole different challenge to do this on scale for multiple plans. And then aspect like lots of models. And that's what I'm gonna talk about now. So there's one very under underappreciated part to it, which is the whole infrastructure. So once you do a while, you do a prototype, you may just do on your laptop and train something and then do some predictions in it and maybe send them by mail. But if you want to do this for like 40 plans, hundreds of models or hundred plans, then you have to make this scalable.

And for that you need infrastructure to basically to all these parts from storing the data and having compute for the ingest of data, the training doing the forecast. And all of these parts need to be covered. And this is a non-real feast. So just to give an example for us, they're running roughly about thousand of so-called jobs every day for all these parts of the system, which to them, and that has to be scaled and has to be scaled in a cost efficient way as well.

And then when you have so many plans and you have hundreds of models for them, of course you cannot look at them all ly or manually, and you cannot look at all the moving parts which are in the system, like the, as we start infrastructure data getting in. the model quality that the API sends, like everything correctly for the closed loop, that the front end is working correctly, all of the, the, the dev app is working correctly. So all of these parts require monitoring and some automation to the automatically detect when there are issue issues and being able to react to them.

And we use a combination of existing tooling for that, but also have specific tooling, which are some, which are often required for this domain to make it again, more efficient to spot problems that are relevant for us. And of course, as I said before, everything is changing all the time, so there will always be problems and that have to be caught. And for the nearly last slide in the scaling part that's, that shows basically our web app. And here it's important that this is gives an interface for, for, for the plants itself to ideally, I mean, of course get trust in all of the predictions and the steering, but ideally also be enabled to improve the data and the processes themselves.

As of course, we are limited capacity to do all of that. So ideally the more we scale, the more of these capabilities come to the plant to see are there problems in the data, is there something off of the steering? and can we trust the model and what does the model currently doing? And all of this basically gets reflected in, in this web application that we have and should enable the plans to improve the whole product or the whole the, the whole predictions and the outcome themselves to some extent. And with that, I am nearly at the end.

So what you see here is a slide from an old Google paper, which is roughly 10 years old, but it's about this kind of predictive system or machine learning system and what makes them successful or what is necessary to make them successful. And what you see in this slide here is in the middle of this small black part, which is the actually machine learning, the predictive model part, which is very small and only a small part, and making actually this kind of system work. But you have all of the things that we talked about which are required.

You need some serving infrastructure, you need monitoring, you need like this pre-processing pipeline to this feature extraction, which also does data cleaning. And you need the configuration capabilities to make this kind of work. So with that, I talked about how to make the predictions reliable and keep them reliable, how to turn them, what is necessary to turn them into actually a process outcome and then therefore business outcome. And then finally what is required to scale this kind of solution on a big scale. And with that, I'm done and happy to take questions. Thank you very much. mateis that's fascinating presentation.

and I, I'm, I'm, the question that comes to mind is what, how long does this take to deploy in a, in a, in a plant if you are, if you are setting up and optimizing the system, and do you have some kind of base level of technology that, that you sit on top of? Or can you be, can you install it from scratch? So maybe to answer the second question first, so we can install it from scratch because it lives in the cloud. So we get connected to the data and then we connect via an API or the app.

And for the first question, it ha it, it very much depends on what's the state of the plant, so how good are the data, how, how well can they be exported to us how, how quickly we can integrate into the system to steer. so it can be from very fast, so from a few weeks to a bit a bit longer. Okay. Okay. So it can be, can be very quick. and there's a, a question really around the economics of the software. can you talk a bit about the initial cost or the return on investment? Yeah, I mean, so I mean, therefore ourselves probably is better guided for that.

But in general, it's like we have a different case studies which we can provide maybe afterwards, but it's, it's like, there are two, two ways basically for the return of investment. The one is saving energy is a big part by basically if we are above, if you are above target, then we basically can become closers and thereby saving energy. And the second part is of course, if we are unable to take out clinker, then that by itself saves costs, which is the biggest part of the, of the cost of the cement for the detailed, for the detailed numbers, it's better to maybe follow up. So happy to if contact for some case studies.

But those are the kind of two main reasons I guess people are using the, the the software. It's yeah, it's the, it's the energy savings from drought and grinding Corsa, but also with new products coming onto the market exploring ways to reduce the clinker factor. I guess that the software can be deployed across a, a range of products to optimize that clinical reduction. Yeah, I mean, in the first place, maybe to, and maybe to make myself a bit clear in the first place, it's to optimize the the, the, the cement quality. So we choose variation. And then as a side effect of this, of course you said ideally energy, and then you can, by reducing the buffers, you can also reduce lingham. Yeah.

But the main point is to reduce improve the quality. Yeah. Very good. Well thank you very much. fascinating presentation. all of these slides will be distributed to everyone who registered for the webinar later today. but for now, Matthias, thank you very much. Thank you so much. Okay. and onto our fourth and final presentation welcome manic manic Low from Grid beyond in the uk. he's a global director of sales and operations at Grid Beyond where he's played a pivotal role in scaling the company's commercial success over the past seven years with deep expertise in sales strategy and operational efficiency.

Manic leads to global initiatives that align sales performance with business growth objectives. His leadership has been instrumental in optimizing processes, enhancing cross functional collaboration and driving revenue acceleration in a rapidly evolving energy technology landscape. So to learn more about that energy technology landscape in the context of the cement sector over to you manic please share your slides. Sure. Thank you very much, Thomas. Glad to be back. and yeah, it's been fascinating, you know, seeing all the innovative technologies and applications that have been presented today conscious of time as well. So I'm gonna try and keep a good pace here.

but we'll obviously send out the the deck after, so I'll just share my screen here. Okay. Sorry, just one second. Okay. So yeah, so today I'll be talking around I guess how cement companies can take advantage of Dig Digitalizations specifically in the energy transition. So, so a little bit of a different topic but something that's very relevant. And I'll be talking around, you know, a little bit about Grid beyond and who we are for those that don't know what we do. but then I'll be going into opportunities and challenges that are currently being presented by the energy transition that we're currently in.

and then I'll also go onto a solution and how we're incorporating flexibility services, but also utilizing AI and machine learning to be able to optimize production production schedules against energy prices you know, deliver in in optimization programs and grid grid services to generate additional revenue, lower costs, all the, whilst being conscious that we're moving towards a, a net zero type environment and, and helping companies to be able to achieve those milestones With regards to decarbonization. to begin with though, very, very quickly, a little bit about grid beyond who we are, what we do. So we're an energy technology company.

We sit at the intersection between energy regulations and policy, energy technology and also the energy markets. and we operate across four continents. we, we, we operate in many of the key deregulated energy markets across the world. So UK, Ireland, us, Japan, Australia, to name a few. And our, our mission really is to achieve a net zero future, but while at the same time uncovering additional revenue, lower en lowering energy costs on opex and driving towards sustainability targets for our customers. we were founded in 2010.

we have now we have a global portfolio of around 2.3 gigawatts of, of load across multiple customers ranging from industrial, commercial public sector also with, on the front of the meter side in terms of independent power producers and renewable asset developers to, again, help, help using our technology and help uncovering additional value. we also have some pretty key institutional investors, financial investors, and also strategic ones as well that you can see at the bottom of the screen there, like EDP A, BB Yoko Gawa Constellation. So we've got, we've got the right sort of partners to be able to help us to achieve our, our mission and vision.

we're, we also operate a very large portfolio of energy storage assets. so yeah, over, over 900 megawatts now of energy storage operate globally, and that's transmission connected. Large batteries, like a hundred, 200 megawatt assets coming all the way down to commercial industrial level energy storage, and even now looking at, you know residential and very small commercial. So in terms of what we do, just very quickly, so sort of four areas that we, we kind of operate in, in terms of solutions, so intelligent demand side response, and that's basically energy flexibility.

So it's looking at how we can find latency within various processes various different sectors to be able to use that energy flexibility and, and put it into various grid grid services. just a very quick overview for demand response. For those that don't know what it is, it's essentially a means by which the grid grids have found that they can be more economical in the way that they're trying to match supply with demand on the grid. So rather than increasing the supply during times when the demand peaks by reducing the demand, it has the same effect of, of matching supply and demand.

and, and that's really where grid services have come from, or, or demand side response, and companies get paid to be, to do this across the world in various different growth grid programs. and that's where we come in really. So we use our market expertise, but also our technology, our ai, our digital twins to be able to optimize sites with regards to the energy markets. so the wholesale energy market, various grid services also looking at more local services down to the regional level. so yeah, and, and we also provide our services as a software as well. So we have various software packages that we provide to, to help again, enable customers into different energy markets.

And in terms of onsite assets, yeah, we, we basically provide an end-to-end solution for funding energy storage and so, and solar as well for industrial commercial sites. So this includes everything from asset funding, connection management and all and helping customers to, to deploy energy storage for resilience you know, and, and net zero benefits as well. renewable energy purchasing is all around PPAs power purchasing agreements, so helping to advise customers on the, the, the right sort of PPA deals to, to be able to secure renewable generation and transition towards you know, a hundred percent clean energy purchasing.

And then finally, a more recent one is 24 7 carbon free energy, which is where we are now, again, through digitalization reaching a point where we can match cons, you know, an energy, the energy load or consumption of a particular site down to the hourly level with renewable generation, whether it be on site renewables or whether it be through power purchasing agreements. So it's getting really granular in terms of, you know, tracking your, your carbon emissions right down to the hourly level. And again, this is being enabled through technology and digitalization, few of our customers there.

So I, I'm gonna, obviously, I'm focusing here on like our industrial commercial customers and if so we work across multiple sectors like food logistics, glass and metals, chemicals and water data centers crypto mines. But of course, the key key sector here to focus on is aggregates and cement in which we work with companies like CRH cx wholesome, zi, unison, central Plain Cement Company, and, and Iris as well. so yeah, we work, we work within the sector. We've been operating with with, with cement mill for, I wanna say around about 10 years now providing optimization and energy flexibility services.

and yeah, cement, the cement sector is sort of a high a sector which has a higher degree of energy flexibility. And in many of the markets that we're in you know, cement plants are usually the sort of first movers when it comes to looking at energy, energy, flexibility services. So we are just trying to take that to a different level now by incorporating our digital twin platform, which I'll show you in a moment. this is a key one here. this is basically showing, sorry, let me just hide this here. this is a key slide here. the three Ds of the energy transition. So decarbonization, so we're currently in you could, you could argue the third major energy transition.

So we went from, you know, biomass to coal, coal to oil and gas, oil and gas, and into, in a big way, into renewable energy. And now we're at the advent of moving into what I would call smart grids and energy storage, energy flexibility. So decarbonization itself is presenting various challenges, you know, for companies. energy costs is obviously paramount, but carbon emissions is, is quickly becoming a very important consideration with a lot of cement organizations, obviously having some pretty you know, tight and key milestones for decarbonization over the next 10 years decentralization.

And that's basically a shift from that traditional grid system set up of having a supplier going into the transmission network sorry, an energy supply a power plant going into the transmission network and then going into various businesses and, and houses, et cetera. That's being, that's obviously been decent, decentralized completely in in many you know, deregulated grids across the world. You know, customers have their own onsite generation these days. Solar combined heat and power plants, waste heat recovery units, energy storage is taking off and are taken off in a massive way, you know, with a lot of plants, again, investing into batteries and grid service participation.

So yeah, demand response and flexibility markets where, as I mentioned earlier, the cement sector is a key player and there's a lot of value to be had by being more flexible with the way we operate from the viewpoint of our energy consumption. And finally, this is all being enabled by digitalization. So to manage the rising complexity of, you know embedded generation coming onto the grid, you know more and more renewables coming onto the grid, you know, it's very important for us to have an operating system that can, that can manage that.

so this will, this includes things like real-time asset monitoring, digital twins with regards to energy prices automated energy trading, integration and visibility, increasing the visibility that is across your production, your energy and your grid constraints. And without digitalization you know, decentralization and decarbonization just would not be possible at scale. So it's absolutely paramount. a little bit around our solution now with regards to how we help companies within the cement sector. So our solution is called Flex Pilot, and it's a digital twin, which optimizes production schedules against energy prices and various grid programs.

So there are a few, there are basically two key components of that. One is AI based price forecasting. So obviously, you know, the presenters before me have all mentioned probably some level of AI integration. and of course this is a, a massive wave at the moment that we're going through. and it's no different in the energy sector. It, it is transforming the, the whole grid network. And one of the key things that we use AI for is predictive forecasting or probabilistic forecasting using AI and machine learning to predict energy prices a day ahead, a few days ahead, even a week ahead. And, and also be able to predict that down to five to 30 minute intervals.

Many grid networks operate on either a five minute basis or a 30 minute basis. and yeah, we can predict all the way from a week out, down to real time. also then when that combines with our digital twin models, which, which basically take into account dynamic and static production parameters. So like for in a cement mill, you know, you'd be looking at things like your silo capacities, your silo levels you know, the conveyor rates the energy load profile when you match all these things together and combine it with the AI based price forecasting, which will give you, let's say a week ahead view of where the high energy prices are gonna be happening.

This allows you to be able to optimize your production schedule based upon production constraints, energy prices, and marginal costs. so, you know, we call this production price optimization. so rather than running your plant in, let's say a more static way you can be very dynamic with regards to load shifting your, your production to times when the energy prices are low, filling up the silos with with storage, and then entering those out during times when the prices go higher. So, and, and this is all being scheduled and, and monitored by our digital twin platform. and then, yeah, that creates maximum benefit really.

So you, you turn your, your cement mill or your, your concrete plant from a passive asset into a dynamic energy asset which can actually then reduce your energy costs by being more, like I said, by being more price responsive when it comes to your production. and you can also maximize your you can create a whole new revenue stream by now, increasing the amount of megawatts of power that you can put into these various grid programs at the right times, which are very lucrative and can add a whole new revenue stream where you get paid from the energy grid.

and you can also increase your resilience and, and decarbonization as well, because every time a plant is called upon in a demand response service to, to slightly turn down their production or modulate their production, you're basically eliminating the need to turn up a power station somewhere that's, that's probably producing emissions and that has a widespread effect on the grid in terms of reducing carbon emissions as well. So it's all around benefit. just a little bit more detail around how weak ahead production price scheduling works is to help you kind of visualize it. So volatile electricity prices obviously mean that it's hard to, you know, to predict your operating costs.

And we, and the market is more volatile than ever in many grids across the world because of the advent of renewables. So energy procurement and production planning are often siloed and separate. manual production scheduling frequently misses the cost reduction opportunities. So the way we do it is we basically configure your asset parameters and constraints into our digital twin. We either use data, we either use our own sensors to capture that data, or we just integrate with your SCADA system to capture that data. We then look at your production targets for the weak ahead and run this against our AI based price forecasting.

And then what we're, if you look on the right hand side here, what doing here is we are running the production harder in the cement mill in the raw mill during times when the prices are lower, proactively building up stock in the raw mill silo, the cement silo, clink storage. and then once the, when, when the price actually spikes, we are then emptying the silos to feed the rest of the process downstream from the storage buffer. And that's creating flexibility. So that means that you could effectively, you know, turn down your your raw mill, your cement mill, and as long as the storage silos are feeding the downstream processes, it's not gonna impact your production.

And actually, by turning down at those times when the grid requires you to do so you'll, you'll get paid a very lucrative amount of revenue for doing that, or all make a significant cost reduction if you're doing that during very high prices. So this is a, this is like a seven de view of a cement mill. So on days one to three, the, in the pink line here, this is the load profile, IE the energy consumption. So the load profile here is quite flat. You can see because the price, this is a price here in blue, the prices are quite low, but as the price rises up on day four, we basically empty the silo during that time and reduce the load to cement mill or the raw mill.

And what you've done there is you've basically shifted away your production from this peak. So you've basically saved significantly there by, by not operating during a very high energy price peak. And, you know, we work very closely with customers to determine the, the constraints, the the parameters, and also their marginal costs, the production, and and, you know, making sure that we work within their production targets. 'cause that, that is the most important thing. this is just giving you a UI kind of view of the, what digital, the digital twin platform looks like.

so you know, you'd have your assets on the top, you could click into each one, and you could see like the asset parameters that we are we're capturing here. and then this shows the optimized load profile of the cement mill. And this shows how, you know, we are monitoring the silo capacity in accordance with it. So when the load drops down, which is this blue line here, you can see we're emptying the silos basically to, to feed the process. And this pla this system that we have, it sits alongside your, it would sit alongside your current you know, monitoring systems and your current process optimization systems.

It's just adding on an additional layer with regards to energy optimization, IE within the wholesale energy markets. the, the tool can also be used for, for multi-site analysis as well. So I'll just go through the slide into this one here. So if you have multiple sites, obviously cement mills, concrete plants, et cetera we can look at the total energy cost savings on each site. We can look at the total production on each site. We can look at the average flexible megawatts that you, that you are putting into different grid, grid, grid services at any given time. And then we can benchmark using this as well.

So, you know, you can unlock greater flexibility by orchestrating across your whole portfolio and the by what this means is, you know, you can shift flexible production between sites to take advantage of localized lower energy prices. You can also do intelligent distribution of production tasks. you can maximize your energy savings while, while meeting overall output goals. and that, that is the most important for us to make sure that your, your production is not impacted, but you're making a significant saving on every unit of product that you're making. here's an example of a cement mill in the uk. So you can probably tell by my accent that I'm based in the uk.

so yeah, this is a 12 megawatt site with a eight megawatt eight megawatts of flexible load across the cement and raw mill, four megawatts inflexible load. and we used AI based price load forecasting digital twin production scheduling accessing multiple grid mar grid networks and even putting down an energy storage asset IEA battery on site. And by doing so, you could see that when we, when we first spoke with the customer, the existing value was around 650,000 mainly across capacity and a little bit across, they were doing, you know, most, most cement mill will be doing some level of load shifting or some, some level of being price responsive.

So the, this, this site was, but you can see that we managed to increase the value significantly to more like 1.35 million pounds. So, you know, double, double what they were on previously by accessing, by having, allowing them to dispatch more and, and putting more megawatts into the energy markets putting them into these grid balancing services, like I mentioned earlier, that that reward and, you know, reward revenue for doing so. And then you can see the, the benefit increased even further by installing a battery on site. So that's just giving you a, a kind of real, real world view of of yeah, the kind of benefits that we can, we can generate.

And this is a case study for for CRH in the uk. So we manage 25 megawatts in in grid grid markets for, for CRH across three, three or four cement mills and I think 15 quarries. we also provide a net zero consultancy service to them. so yeah what, what the way we work with them is the digital twin will model the customer's processes, and we use the customer's data around, you know, like the, their silos and storage and energy profile. And then it will we will de we've deployed our own sensors on their site that takes into account that data to then be able to put it in the dig digital twin and to be able to help 'em with a production price optimization.

And the actually results here are, the site went from being able to only turn down their load, let's say two or three times a year to n nearly every day, being able to find pockets of, of the day when they can shift their load and, and make significant savings in their production. And it, it offset their energy cost by 10%, which is quite significant. So that's everything I've got in terms of the presentation for today. Polly's was a bit rushed there, but I wanted to make sure that we keep in good time as well for people. So any questions? Manic? Thank you very much.

And I'm just thinking with electricity prices as they are in Europe and especially in the uk, this is a, a really, really important product. I can see how effective it's been. it's yeah, just, just generally getting set up with your, your product. Are you, are you also involved in the buying and selling process as well as the forecasting and the, and and the load shifting, like you've described? how kind of integrated do you, do you get into a, into the you know, into energy supply for a cement plant or any NE group of assets? So, good question. So we're, we're not an energy supplier, so we don't, we don't supply any power.

However as you can see, you know, in terms of optimizing production versus price for, for, for sites, we do need to get very involved with the procurement of the energy because we can help to effectively determine the energy procurement strategy. so what a lot of customers do when it comes to energy procurement is they will buy in blocks, you know, in time. So one year out, six months out, three months out, one month down to, you know, week ahead and then even day ahead. And typically you know, a lot of customers like to fix out their prices into the future with the way the market is at the moment, and it's so volatile.

with the integration of renewables and things I mentioned earlier, we are finding that there is an increased opportunity that if you could leave a larger proportion of your load on what we call the day ahead of the spot market or floating on the spot market, you have a lot more opportunity to capture value from this volatility. so we're, you know, we have seen in markets like the UK for example, there is a massive benefit of moving from like a forward purchasing strategy coming down to more of a week ahead, day ahead type optimization, even in real time. I guess it depends also on how flexible a an asset can be, a plant can be with its, with its production and what utilization rate it is.

And you know, if it's maxed out already, then it's not gonna have much room. But if it's got a lower utilization level, maybe it can choose a bit more. how, how to arrange its production schedules and time it to, to, to benefit the the lower, lower cost energy. is it is it something that works in all markets or only in Europe or only in the uk? Where, where are you active as grid beyond? Yeah, good question. So we're, we're active in many of what we call the deregulated energy markets. So UK Ireland, US particularly in pe, Pennsylvania, Jersey, Maryland, Texas, California, and these are like the very kind of highly deregulated areas Japan and Australia as well.

Japan is actually very interesting market. there's a massive increase in the amount of distribution connected batteries in Japan at the moment. And the, and the value at the moment per megawatt is something like 1.1 million euros per megawatt of batteries. So it's booming market for so yeah, but this is moving into many grids globally because, you know, with more, more and more renewables coming on board, it, it's definitely needed in order to be able to integrate that into the grid. And it's, that, that's partly it's, it, it it's the not just when an asset takes the power, it's also about feeding power back into the grid and also enabling grids, grids to balance is That's right, isn't it?

So it's not just simply about taking energy. No, it is, it is not. so, you know, when it comes to production load or demand, yeah, that's more around reducing demand, but the, you know, like I mentioned, batteries can, can go both ways. They can take power from the grid or dispatch power back onto the grid. Same with generation assets like combined heat and power plants that can, you know, reverse power back onto the grid. So yeah, for the grid rewards for, for doing both, but the effect is the same, which is basically a balanced demand and supply. Mm.

in the uk we're having a lot of discussion around our grid and you know, the investment taking to, to increase the transmission and distribution around the country. are we, are we an outlier? how do we compare to, how does the situation compare in the UK to other parts of Europe or America that you were describing? The UK is one of the most innovative and forward markets in the world when it comes to decentralization, decarbonization, and digitalization.

Like we, we are in the uk we're a hotbed at the moment because it's not only like flexibility is going now down to a residential level in the UK whereby you know, you've got basically hundreds of thousands of houses that are deploying heat pumps, solar batteries, very small assets that can be aggregated together into one unit and put into the markets. Yeah. So opened up a whole nother level in, in the UK market. but the, the other grids are also catching up, you know, like Australia is, is a, is a, is a grid where there's a lot of innovation happening. yeah.

And, you know, we've tried to position ourselves actually to, in all the areas that I mentioned to you where, where there are lots of innovations happening and it's moving quite quickly. Yeah. Well hopefully we can keep getting updated by you. It's really important, really interesting especially as I said for the uk, but all of Europe and, and as you say, the deregulated markets. So thanks for the update manic brilliant presentation. No worries. Thanks for having me back. great, well that's all we've got time for today. Thanks for sticking around and listening to those presentations. as I said we'll be in Asia off to Bangkok in June.

it'll be a, a fascinating country to wa watch their moving very fast on alternative fuels actually also on battery storage for cement plants, which is a, a big innovation in the sector. And one that SEG is a pioneer of. So lots of interesting things to, to hear and see in Asia. But for now, thank you. That's all we've got time for. that's another Cemtech live webinar. hopefully you'll receive the slides probably later today by email. Keep a lookout for them and see you next month. Thank you very much. I.