Presentation: AVEVA MES - Advanced insights and guidance, optimized with AI

2024 - AVEVA World - Paris - HMI/SCADA

AVEVA MES - Advanced insights and guidance, optimized with AI

So my name is, Jean Pierre Caron or better known as JP. So, since we're gonna talk about artificial intelligence. So I am from the French Canadian side of the world, which means some people will say that I speak artificial French. But let's let's get going. So since we're I'm your last stop before missus Cloney's speech, so you don't wanna be late and collect your potential, gift if you're lucky to win. So that that presentation is is segmented into three, so we're gonna briefly, give you an introduction, go through the industrial assistance, and then finally, the advanced analytics. Statistics are statistics. So I'm not gonna read everything, but as a summary, industry had been investing a lot into into the digitization industry four o. The next wave is to digital twin, and the reason for that is because even though they have invested in, digitization or digital transformation, the value they get of it, they got some value, but it's not broadly it's not broadly, distributed within the organization. So there's still a lot of potential. And then, even though the asset performance up optimization is still a random thing, so they don't yet have full control on how their their asset behaves. So and and this is proven because the majority of our MES lighthouse customer that are onboarding the Connect platform is mainly for leveraging AI technology on top of Connect. So that's the general idea of those, of those. So if you haven't seen this slide or version of this, you must have been sleeping somewhere in a corner. So so the connect platform is a collection of five services. You know, there's the infrastructure with security, access, monitoring, and thing things things of that nature. The data service, which the service allows to collect data from the shop floor and then keep that into the connect platform as a data service. We have the visualization service that allows you to obviously visualize the data that is in data service. We're gonna focus a little bit more on the on the modeling and analytics, which is, basically the way we are gonna templatize, some of those, AI use cases so they can be, they can accelerate the adoption by by creating a digital twin and creating model to optimize some some specific use case and then application specific application development. Obviously, we can collect connect could collect data from our in our case, we're gonna focus on the MES, but collect data from Edge data store, AVEVA historian, AVEVA PI Server, or other, sources of of of data. Obviously, because we wanna templatize, the only way we can templatize it is if the data is in connect. Even though the technology allows to connect to other system, our template is going to be based by the fact that the data needs to be in the Connect data service. So again, this is kind of the idea behind an advanced analytics is to model your process, learn, monitor, alert, and notify or communicate, and and and and do performance or correction action to get the process back in in in in control. So that's the general idea about the, so we wanna templatize again on on specific use cases that all processed. Why? We wanna you wanna we wanna reduce the or we wanna reduce the time to value. And it may sound so you can get quick value or value very quickly, but you could also figure out that you can discover failure early also. So you don't invest a lot of time to figure out, is that gonna work for me or not? So so the the the the the the time reduction is could be positive or negative, but it's it's it's really to, to to accelerate the adoption. And, obviously, you wanna scale that across the enterprise, and we recommend to start with a small use case, deploy that small use case in multiple sites before trying to build too many use case and then deploy multiple use case across the organization. So that's the kind of the recommended approach, if you wanna be successful. Obviously, in order to templatize, we need to bring MES data into, the connect data service. So that's what our connected MES or hybrid cloud MES solution architecture is is is based on. So from our on prem MES systems, right now, we have, we have hybrid MES for both AVEVA MES or manufacturing execution system and production management, which is also known as EMPLAC. Next year, if you were to in the MES roadmap session before next year, we're going to add to this recipe management or and batch management as additional data, MES data pushed into Connect. So that's the general ID. And then our cloud MES, and I think my colleague here, Jeff, we also deliver a bunch of standard visualization, which allows you to build dashboard rapidly using standard widget for utilization performance. So we all the typical MES KPI. So that's the general idea about, our cloud MES or hybrid cloud MES. So now that you we have data in data service or in connect, Let's go into a little bit more detail how we apply artificial intelligence technology on top of MES. Obviously, the approach oops. Wrong side. The approach the approach is generic, so refer we refer this as the industrial AI assistant on connect. So that's kind of in partnership with, Microsoft so that some flavor of chat chat GPT is embedded. And we do we do the data or the knowledge linking between the different data that is in connect, time series data, asset data, event data, three d engineering data. And we take care for you of the data linking between those different source of data. And then the idea is to for the system, you provide typical question. You ask them specific question. And based on your question, the system will do its best to find the data from the right source. If it's more asset centric, it's gonna query the asset data. If it's more time series centric, it's gonna automatically create, or query the, the, time series data or if it's MES. Or if the question is more complex, you get get you can get data from more than one type of data. It could be a mix of asset, tree engineering, time series, and event in a single response. So it all depends on the complexity of your of your question. So basically, we take the industrial data, we apply a large language model, and then we give you a response. So so this is kind of the idea. So you through a natural language interface, you type in your question. It's gonna convert this into a, into a search, query the data. We actually provide you the citation. The citation is what data it actually retrieve in order to formulate the the response. And then we're gonna generate not only a textual answer, but also a graphical answer based on the data that it's returning. If we dive in a little bit more details so if you ask if you ask a question as an example, what is the longest downtime for this piece of equipment today? The chat interface or the natural language interface will take that question through the orchestration and the l l l m model. It's gonna convert that into a query. And, obviously, in our case, if it's MES data, it's gonna go to the MES event and percent potentially look at MES asset and then formulate an answer and return that answer back to to the user through the, through the, through the interface. If we start your recording So let me go to the next slide and I'll start this. So oops. So within connect visualization, you can build your own self-service dashboard, and there is an addition called, industrial assistant. So it it prompts you. You can ask a question, what are the asset associated? So that's a very basic question, just asking what are the asset available for me on this specific process. So it's gonna return all the asset related to that process. You can click on one. It's gonna bring you to the asset page. It's gonna it's gonna give you basic information, but it also gonna expose any dashboard that were prebuilt for that for that asset. So it's gonna bring that dashboard and it's gonna return it to you. You can go a little bit more fancy and and and ask a more complicated question that the one I specified. What is the what is the longest downtime for that piece of equipment, for today? So again, it's gonna query. Now it knows a little bit more that this question is more related to MES data. It's gonna return the textual question and give you a time time series chart of what when where that downtime is for that period. Then you can, go a little bit further and then ask what is the utilization performance for that piece of equipment for a certain certain duration. And as you can see, I intentionally even though there are some typos, the LLM model will adjust accordingly or do its best. So it's going to return for each hour what are the KPI for this piece of equipment and give you give you a chart. And if you click, you can get that open that chart in a bigger in a bigger format, and then you can go and modify it if you want to. So that's the that's the general idea about the industrial assistant applied to to MES. The next step for this is we're starting to work on this is to ask in the in the chat interface, build me a dashboard with this type of information. So instead of having predefined dashboard, the goal eventually is to create dashboard through that interface. That's that was it for the industrial assistant. Now let's dive into a little bit more into the advanced analytics. So the idea about advanced analytics is to take some typical use cases and develop templates for them to so so the, the adoption is is quicker. So we're gonna focus on the uptime optimization, but things like perfect quality, throughput optimization. And and some use case are more process centric, some use case are more asset centric, but the the the key here is there there's many here. I think the first one we're gonna templatize is the uptime optimization, which we're targeting early next year is my understanding. And but and that's the key here. Don't assume that any of those are templates you wanna deploy. So you need to find your use case that provides you business value, because it it's up to you to decide where and where it makes sense in which context so so you can get value out of it. So the way those templates are built are this is very generic. There's obviously there's a little bit more behind behind the scene, but the it's a wizard driven. Each of those boxes has their own wizard to actually come and and develop each of the of of those, of those configuration, but those are the main top one. So the first one is to define the data source. So what data you wanna bring into your digital twin? So is it process data, MES data, and thing things of that nature? The second is to build, a what we refer as a thread or it's kind of a calculation engine, where it allows you to manipulate the data that you're you're you're collecting. So you may wanna do adding two values together. You may wanna bring averages. So the the the digital twin those those thread allows you to manipulate the data. One of the one of the idea of using them is also to standardize. So instead of trying to harmonize all your data from multiple sites, you bring them as there are, and you use that digital twin or those thread configuration to harmonize the data for each of those, for each of those template across your across your different sites. And then you build a model, which is basically that's what, essentially, what the template is is the logic that will learn. The model is the learn learning logic that will will detect, information and detect anomalies and and predict when the next downtime might might come. So if we deep dive a little bit into uptime optimization, sometime referred as predictive uptime, This is the flow of how the system works. So obviously, there's the detection. So based on that model, once the learning has been performed, it's gonna detect or it's gonna prompt you with two things, anomaly score and probability. You may have heard that term a few times today already. But just so it's clear, the anomaly score is a reflection, is the system in a current, in a normal situation now? I I compare this with temperature. Anomaly score is a temperature today. But normal temperature in Paris is different than normal temperature in London. It's different than normal temperature in when do you think or what's the probability in the foreseeable future, and that's configurable. Is it in the next fifteen minute, thirty minute, hour, four hours? And it's it's similar if I use that same analogy of the weather forecast. So if it's sunny to if it's sunny now, based on the situation, is it going to get warmer? Is it going to get cloudier? Is it going to rain? So the probability oops. The probability is is essentially the chance of a downtime coming in the foreseeable future. Once an anomaly is detected, you can send an alert, and then the system will identify the top drivers that makes that probability go up. And then you can do, you can initiate a case management or an escalation and send that information to people that can fix or address that situation. And then use the recommended ideal run time condition to reset or to set your process parameter to get to reduce that probability. So you don't have a downtime that is not expected. When it comes to MES, and that's gonna be part of the template, MES stores in the connect data store a number of MES events. There are some that are listed here. Equipment utilization, job response, material produced. You can see that, you know, is the machine was down, not down, what state, production, what was produced, what product was produced, was it good, was it bad. So there's there's all the MES data that is accessible and then through a wizard and a simple user interface, you can bring that data will bring that data into those template and then apply that to the to the different models that you wanna you wanna use. Obviously, that list of MES event will grow over time, but that's the that's the general idea is to make that configuration as simple as possible. Now that you've detect or, obviously, the the system has a built in dashboard, but you can also build your own dashboard showing the same information into the connect visualization using the self-service and and build because all the data we use is in connect visualization or is in the connect data service, so you can view it with connect visualization. So the detect process, so it's gonna look at process parameter. MES data based on the learning is gonna learn from those values and predict or give you the current anomaly score and the probability of an event coming in the foreseeable future. So that's the detect phase, which is obvious. Then the top model driver. So based on the process parameter you provide the model, it's gonna give you, which parameter is contributing to that probability or that anomaly score to be to be to be not to what you expect. So there's different color coding. I don't have any one here, but maybe one of the the the what's affecting the current score is the a current value of a specific process parameter. It could be a value that is trending in a specific direction or a value that is, that has a a lot of variation. So that's the so that the the yellow one or orange one is the trending, the green is the variation, and the blue, there's none here, but the it's the maybe you have a specific value that is out out out of a standard condition. The system will also so he knows the anomaly. He knows the the, probability. It's gonna propose, and this can be automated. So it's a it's a it's really up to you to decide if you wanna push those value automatically to the control system or you wanna let somebody to interact with it. But it's gonna provide recommendation of process parameter set points you wanna change to get or to reduce that probably score. So that's that's the general ID. And then you can communicate and send all that then and information to somebody that's gonna take care of that information and address address the, or reduce that probability score. This all sounds good, but one of our lighthouse, while the system performed as expected, their biggest challenge is they were starting to get those notification. They were not ready from a change control management to have the process for their organization to deal with those prediction or those, future anomalies to be able to react in time. So, you know, what do you do if that communication comes in the middle of the night or you don't have the staff? So there is there is a thought process to be able to be able to consume that data, within your operational processes to, to make sure that you're going to be able to get value out of the system, unless you want to completely automate in getting those data right into the system. So there is a learning process. This is, one example. There's actually a session tomorrow on this. So the idea here is that's a and they actually did a lighthouse last year. They're actually going in production in the near future, but, you know, when they were packaging this quick, depending on the humidity of the powder, they were putting too much. If if the humidity is too low, they need to put more powder to to meet the one KG packaging limit. And if there's too much, humidity, yeah, you put less, but the quality of the product is less. So it's a it's a trade up between those things. So the payout was actually within a couple weeks. So and that's usually the case for advanced analytics. If you pick the right use case, you're gonna get the payout very rapidly. So that's that's kind of the the idea or the the trade off. So in this case, they save ten percent of, of of product in each, in each packaging too. So so that's a lot of money. These are all the sessions. So we have this in the expo area. There's there's some upcoming session that will talk about some of those, those, that advanced analytics or industrial assistant in the context of other other, other situation. But that's those are and, yes, I'm the last one of today. There was a couple, a few earlier this afternoon. So so you can you can listen to them or look at them after the conference if if you wanna learn more. Or you can still contact me if, if you're if you have question. I may not have the answer, but I'll find the right person to help you if you're interested in a lighthouse or something. So we have time for a couple of questions.

Join this session to learn how AVEVA is infusing AI-powered advanced analytics and machine learning (ML) into its MES offerings to enable continuous production optimization. Learn how these capabilities are empowering manufacturing operations to accelerate and improve decision making and root cause analysis activities, as well as providing real-time insights and automated actions that facilitate remediation to directly improve operations performance.

Download Presentation

Company

AVEVA

Speaker

Jean-Pierre Caron

JP (Jean-Pierre) has been the product manager for 15 years with the last 7 for Model Driven MES and APS and MES connect Visualization components. He brings more than 35 years of experience in industrial application, solution development and technical Sales related activities all being around using Wonderware/AVEVA technology. Prior to his product management role in Wonderware, JP (Jean-Pierre) worked 10 years as a solutions architect for MES projects and 15 years as a Sales engineer. JP (Jean-Pierre) holds Bachelor degree in Electrical Engineering and located in Montreal.

Session Code

AW24-HMI-D1-SESS-362

Transcript

So my name is, Jean Pierre Caron or better known as JP.

So, since we're gonna talk about artificial intelligence. So I am from the French Canadian side of the world, which means some people will say that I speak artificial French.

But let's let's get going. So since we're I'm your last stop before missus Cloney's speech, so you don't wanna be late and collect your potential, gift if you're lucky to win.

So that that presentation is is segmented into three, so we're gonna briefly, give you an introduction, go through the industrial assistance, and then finally, the advanced analytics.

Statistics are statistics. So I'm not gonna read everything, but as a summary, industry had been investing a lot into into the digitization industry four o. The next wave is to digital twin, and the reason for that is because even though they have invested in, digitization or digital transformation, the value they get of it, they got some value, but it's not broadly it's not broadly, distributed within the organization. So there's still a lot of potential.

And then, even though the asset performance up optimization is still a random thing, so they don't yet have full control on how their their asset behaves. So and and this is proven because the majority of our MES lighthouse customer that are onboarding the Connect platform is mainly for leveraging AI technology on top of Connect.

So that's the general idea of those, of those.

So if you haven't seen this slide or version of this, you must have been sleeping somewhere in a corner.

So so the connect platform is a collection of five services.

You know, there's the infrastructure with security, access, monitoring, and thing things things of that nature.

The data service, which the service allows to collect data from the shop floor and then keep that into the connect platform as a data service. We have the visualization service that allows you to obviously visualize the data that is in data service. We're gonna focus a little bit more on the on the modeling and analytics, which is, basically the way we are gonna templatize, some of those, AI use cases so they can be, they can accelerate the adoption by by creating a digital twin and creating model to optimize some some specific use case and then application specific application development.

Obviously, we can collect connect could collect data from our in our case, we're gonna focus on the MES, but collect data from Edge data store, AVEVA historian, AVEVA PI Server, or other, sources of of of data.

Obviously, because we wanna templatize, the only way we can templatize it is if the data is in connect. Even though the technology allows to connect to other system, our template is going to be based by the fact that the data needs to be in the Connect data service.

So again, this is kind of the idea behind an advanced analytics is to model your process, learn, monitor, alert, and notify or communicate, and and and and do performance or correction action to get the process back in in in in control.

So that's the general idea about the, so we wanna templatize again on on specific use cases that all processed. Why?

We wanna you wanna we wanna reduce the or we wanna reduce the time to value.

And it may sound so you can get quick value or value very quickly, but you could also figure out that you can discover failure early also. So you don't invest a lot of time to figure out, is that gonna work for me or not? So so the the the the the the time reduction is could be positive or negative, but it's it's it's really to, to to accelerate the adoption.

And, obviously, you wanna scale that across the enterprise, and we recommend to start with a small use case, deploy that small use case in multiple sites before trying to build too many use case and then deploy multiple use case across the organization. So that's the kind of the recommended approach, if you wanna be successful.

Obviously, in order to templatize, we need to bring MES data into, the connect data service. So that's what our connected MES or hybrid cloud MES solution architecture is is is based on. So from our on prem MES systems, right now, we have, we have hybrid MES for both AVEVA MES or manufacturing execution system and production management, which is also known as EMPLAC.

Next year, if you were to in the MES roadmap session before next year, we're going to add to this recipe management or and batch management as additional data, MES data pushed into Connect.

So that's the general ID.

And then our cloud MES, and I think my colleague here, Jeff, we also deliver a bunch of standard visualization, which allows you to build dashboard rapidly using standard widget for utilization performance. So we all the typical MES KPI.

So that's the general idea about, our cloud MES or hybrid cloud MES.

So now that you we have data in data service or in connect, Let's go into a little bit more detail how we apply artificial intelligence technology on top of MES.

Obviously, the approach oops. Wrong side.

The approach the approach is generic, so refer we refer this as the industrial AI assistant on connect.

So that's kind of in partnership with, Microsoft so that some flavor of chat chat GPT is embedded. And we do we do the data or the knowledge linking between the different data that is in connect, time series data, asset data, event data, three d engineering data. And we take care for you of the data linking between those different source of data.

And then the idea is to for the system, you provide typical question. You ask them specific question. And based on your question, the system will do its best to find the data from the right source. If it's more asset centric, it's gonna query the asset data.

If it's more time series centric, it's gonna automatically create, or query the, the, time series data or if it's MES. Or if the question is more complex, you get get you can get data from more than one type of data. It could be a mix of asset, tree engineering, time series, and event in a single response. So it all depends on the complexity of your of your question.

So basically, we take the industrial data, we apply a large language model, and then we give you a response.

So so this is kind of the idea. So you through a natural language interface, you type in your question. It's gonna convert this into a, into a search, query the data. We actually provide you the citation. The citation is what data it actually retrieve in order to formulate the the response. And then we're gonna generate not only a textual answer, but also a graphical answer based on the data that it's returning.

If we dive in a little bit more details so if you ask if you ask a question as an example, what is the longest downtime for this piece of equipment today?

The chat interface or the natural language interface will take that question through the orchestration and the l l l m model. It's gonna convert that into a query.

And, obviously, in our case, if it's MES data, it's gonna go to the MES event and percent potentially look at MES asset and then formulate an answer and return that answer back to to the user through the, through the, through the interface.

If we start your recording So let me go to the next slide and I'll start this. So oops.

So within connect visualization, you can build your own self-service dashboard, and there is an addition called, industrial assistant. So it it prompts you. You can ask a question, what are the asset associated? So that's a very basic question, just asking what are the asset available for me on this specific process.

So it's gonna return all the asset related to that process. You can click on one. It's gonna bring you to the asset page. It's gonna it's gonna give you basic information, but it also gonna expose any dashboard that were prebuilt for that for that asset. So it's gonna bring that dashboard and it's gonna return it to you. You can go a little bit more fancy and and and ask a more complicated question that the one I specified. What is the what is the longest downtime for that piece of equipment, for today?

So again, it's gonna query. Now it knows a little bit more that this question is more related to MES data. It's gonna return the textual question and give you a time time series chart of what when where that downtime is for that period.

Then you can, go a little bit further and then ask what is the utilization performance for that piece of equipment for a certain certain duration.

And as you can see, I intentionally even though there are some typos, the LLM model will adjust accordingly or do its best. So it's going to return for each hour what are the KPI for this piece of equipment and give you give you a chart. And if you click, you can get that open that chart in a bigger in a bigger format, and then you can go and modify it if you want to. So that's the that's the general idea about the industrial assistant applied to to MES. The next step for this is we're starting to work on this is to ask in the in the chat interface, build me a dashboard with this type of information. So instead of having predefined dashboard, the goal eventually is to create dashboard through that interface.

That's that was it for the industrial assistant. Now let's dive into a little bit more into the advanced analytics.

So the idea about advanced analytics is to take some typical use cases and develop templates for them to so so the, the adoption is is quicker.

So we're gonna focus on the uptime optimization, but things like perfect quality, throughput optimization.

And and some use case are more process centric, some use case are more asset centric, but the the the key here is there there's many here. I think the first one we're gonna templatize is the uptime optimization, which we're targeting early next year is my understanding.

And but and that's the key here. Don't assume that any of those are templates you wanna deploy. So you need to find your use case that provides you business value, because it it's up to you to decide where and where it makes sense in which context so so you can get value out of it.

So the way those templates are built are this is very generic. There's obviously there's a little bit more behind behind the scene, but the it's a wizard driven. Each of those boxes has their own wizard to actually come and and develop each of the of of those, of those configuration, but those are the main top one. So the first one is to define the data source. So what data you wanna bring into your digital twin?

So is it process data, MES data, and thing things of that nature?

The second is to build, a what we refer as a thread or it's kind of a calculation engine, where it allows you to manipulate the data that you're you're you're collecting.

So you may wanna do adding two values together. You may wanna bring averages. So the the the digital twin those those thread allows you to manipulate the data. One of the one of the idea of using them is also to standardize. So instead of trying to harmonize all your data from multiple sites, you bring them as there are, and you use that digital twin or those thread configuration to harmonize the data for each of those, for each of those template across your across your different sites.

And then you build a model, which is basically that's what, essentially, what the template is is the logic that will learn. The model is the learn learning logic that will will detect, information and detect anomalies and and predict when the next downtime might might come.

So if we deep dive a little bit into uptime optimization, sometime referred as predictive uptime, This is the flow of how the system works.

So obviously, there's the detection. So based on that model, once the learning has been performed, it's gonna detect or it's gonna prompt you with two things, anomaly score and probability.

You may have heard that term a few times today already.

But just so it's clear, the anomaly score is a reflection, is the system in a current, in a normal situation now?

I I compare this with temperature. Anomaly score is a temperature today.

But normal temperature in Paris is different than normal temperature in London. It's different than normal temperature in when do you think or what's the probability in the foreseeable future, and that's configurable. Is it in the next fifteen minute, thirty minute, hour, four hours? And it's it's similar if I use that same analogy of the weather forecast.

So if it's sunny to if it's sunny now, based on the situation, is it going to get warmer? Is it going to get cloudier? Is it going to rain? So the probability oops. The probability is is essentially the chance of a downtime coming in the foreseeable future. Once an anomaly is detected, you can send an alert, and then the system will identify the top drivers that makes that probability go up. And then you can do, you can initiate a case management or an escalation and send that information to people that can fix or address that situation.

And then use the recommended ideal run time condition to reset or to set your process parameter to get to reduce that probability.

So you don't have a downtime that is not expected.

When it comes to MES, and that's gonna be part of the template, MES stores in the connect data store a number of MES events. There are some that are listed here. Equipment utilization, job response, material produced.

You can see that, you know, is the machine was down, not down, what state, production, what was produced, what product was produced, was it good, was it bad. So there's there's all the MES data that is accessible and then through a wizard and a simple user interface, you can bring that data will bring that data into those template and then apply that to the to the different models that you wanna you wanna use. Obviously, that list of MES event will grow over time, but that's the that's the general idea is to make that configuration as simple as possible.

Now that you've detect or, obviously, the the system has a built in dashboard, but you can also build your own dashboard showing the same information into the connect visualization using the self-service and and build because all the data we use is in connect visualization or is in the connect data service, so you can view it with connect visualization.

So the detect process, so it's gonna look at process parameter.

MES data based on the learning is gonna learn from those values and predict or give you the current anomaly score and the probability of an event coming in the foreseeable future. So that's the detect phase, which is obvious.

Then the top model driver. So based on the process parameter you provide the model, it's gonna give you, which parameter is contributing to that probability or that anomaly score to be to be to be not to what you expect. So there's different color coding. I don't have any one here, but maybe one of the the the what's affecting the current score is the a current value of a specific process parameter. It could be a value that is trending in a specific direction or a value that is, that has a a lot of variation.

So that's the so that the the yellow one or orange one is the trending, the green is the variation, and the blue, there's none here, but the it's the maybe you have a specific value that is out out out of a standard condition.

The system will also so he knows the anomaly. He knows the the, probability. It's gonna propose, and this can be automated. So it's a it's a it's really up to you to decide if you wanna push those value automatically to the control system or you wanna let somebody to interact with it. But it's gonna provide recommendation of process parameter set points you wanna change to get or to reduce that probably score.

So that's that's the general ID.

And then you can communicate and send all that then and information to somebody that's gonna take care of that information and address address the, or reduce that probability score.

This all sounds good, but one of our lighthouse, while the system performed as expected, their biggest challenge is they were starting to get those notification.

They were not ready from a change control management to have the process for their organization to deal with those prediction or those, future anomalies to be able to react in time. So, you know, what do you do if that communication comes in the middle of the night or you don't have the staff? So there is there is a thought process to be able to be able to consume that data, within your operational processes to, to make sure that you're going to be able to get value out of the system, unless you want to completely automate in getting those data right into the system. So there is a learning process.

This is, one example. There's actually a session tomorrow on this. So the idea here is that's a and they actually did a lighthouse last year. They're actually going in production in the near future, but, you know, when they were packaging this quick, depending on the humidity of the powder, they were putting too much. If if the humidity is too low, they need to put more powder to to meet the one KG packaging limit.

And if there's too much, humidity, yeah, you put less, but the quality of the product is less. So it's a it's a trade up between those things. So the payout was actually within a couple weeks. So and that's usually the case for advanced analytics. If you pick the right use case, you're gonna get the payout very rapidly. So that's that's kind of the the idea or the the trade off.

So in this case, they save ten percent of, of of product in each, in each packaging too. So so that's a lot of money.

These are all the sessions. So we have this in the expo area. There's there's some upcoming session that will talk about some of those, those, that advanced analytics or industrial assistant in the context of other other, other situation. But that's those are and, yes, I'm the last one of today. There was a couple, a few earlier this afternoon.

So so you can you can listen to them or look at them after the conference if if you wanna learn more.

Or you can still contact me if, if you're if you have question. I may not have the answer, but I'll find the right person to help you if you're interested in a lighthouse or something.

So we have time for a couple of questions.