2024 - AVEVA World - Paris - Innovation (CONNECT, AI, etc.)
AVEVA's Industrial AI: Generate impact to achieve greater goals
For two decades AVEVA has delivered proven Industrial AI solutions at scale. Our broad, AI-infused software portfolio continues to evolve across operations and engineering with capabilities at the edge, in the cloud, and through advanced hybrid solutions. We don’t do AI for AI’s sake; we leverage it to make our industrial solutions better.This session will provide an overview of the breadth of AVEVA AI today, our vision and strategy, and specifically where we’re continuing to advance with AI technologies such as large language models and generative AI, including our partnership with Microsoft to take it to the next level.You can expect to hear about:Cutting-edge AI-infused products and solutions to help achieve your operational and sustainability goalsMultiple types of AI working together to create new capabilities Industrial AI Assistant on CONNECT leveraging GenAI to drive a new user experience for our cloud data and solutions
Company
AVEVA
Speaker
Lori Warda
Lori has been working in Industrial Software development for over 25 years. In her current role, she is a Product Manager in the AI & Advanced Analytics group at AVEVA and is working on several products including the new Industrial AI Assistant.
Company
AVEVA
Speaker
James Chappell
With over 30 years of experience in industrial software, Jim Chappell is currently global vice president and head of AI across AVEVA’s businesses and products. Prior to this, he led the Asset Performance Management (APM) products and related engineering/analytics services for Schneider Electric. He was also a founding partner and managing officer of InStep Software, a global leader in AI-driven Predictive Analytics, which was acquired by Schneider Electric in 2014.Jim holds a B.S. in Nuclear Engineering from RPI, a Masters in Nuclear Engineering from the Naval Nuclear Power School, and a M.B.A. from Chaminade University. Additionally, he graduated from the Civil Engineer Corps Officer's School. He also held a top-secret clearance while an officer in the U.S. Nuclear Navy.
Session Code
AW24-CIN-D2-SESS-290
Transcript
Good morning, everybody. Welcome to day two of AVEVA World. I'm Jim Chappell, and Laurie and I are going to talk about AVEVA AI this morning. We're gonna talk about how it generates impact and, to achieve greater goals but across all industries.
First, I wanna talk about what industrial AI is.
You know, there's really three areas of of industrial AI, you know, for pragmatic use. Because at AVEVA, we don't do AI for AI's sake. We do it as part of our industrial portfolio. We as part of our some of our industry leading software and match it with our domain expertise. But the three key areas that we use from a pragmatic perspective are unmatched speed. You've got all this data.
What if you can run AI against it to get new knowledge out of it, to get new information? All types of AI, and it can do it so quickly, much faster certainly than humans, but even other types of software. Then there's the complexity aspect of it, where you can, navigate complex processes, multivariate, and and so forth, and detect things that humans and other software cannot. And then there's the automation. It's constantly learning. It's constantly simulating and predicting, and and just on autopilot twenty four seven. And these are powerful capabilities that we use throughout all of our software, from design, to operate, to optimize, and data.
So we've been doing AI for over twenty years, you know, predictive analytics and, but now it's all types of AI across the board. And again, like I mentioned, we leverage our domain expertise and our industry leading software to combine that in there to get maximum value and push it to the limits. So today, we're up to eighteen commercially available software packages or AI infused software solutions.
And many of them are a big help in the sustainability world, and we'll talk about that as well.
One thing we get asked a lot is, what's your what's your AI strategy? What do you guys do in AI? And I really boiled it down to three simple things. One is we've been infusing AI for twenty years into different solutions to make them better. But then we've been integrating multiple types of AI together.
You know, there's many types of machine learning. There's deep learning, reinforcement learning. There's neural nets. There's genetic algorithms, of course, gen AI, like chat g p t with large language models. There's expert systems, all kind of stuff across the board.
And we integrate where it makes sense. We use the right type of AI for the right solution. And ultimately, we achieve this intellect. We call it knowledge linking, where it creates it becomes a dynamic knowledge graph.
And, it it's super powerful, and this is part of our industrial AI assistant that we're gonna talk more about a little bit later. But we do it across design, operate, and optimize. We we make our design, our engineering designs better, more performant, faster handover. We make our operations smarter, more responsive, and more reliable.
And when you do that, it's the help toward net zero. And so what do we do, you know, realistically, what do we do to achieve to help our customers achieve net zero? Let's talk specifics. Well, one is whether you're building a better factory plant facility, designing it, or whether you're improving your operations, optimizing things.
It could be, you know, asset reliability. It could be optimizing processes.
It's making things more efficient. And when you do that, you're burning less fuel.
You bet you build a better factory, you're burning less fuel. You improve your asset performance, you're burning less fuel. You improve your operations, you increase efficiency, burn less fuel. And that's a step toward net zero.
But ultimately, we actually go to the source and we've been doing this for a number of years, working with, wind farms and solar fields and and things like that. And we help drive down the cost of it, but how do we do it? Well, if you look at a solar or a wind farm, there's about two and a half failures. Studies have shown about two and a half failures per wind turbine per year.
And so, if you have a hundred wind turbines in a wind farm, just as an example, that's two hundred and fifty failures on average per year. If you can find those failures early, days, weeks, months before any type of control system or anything else would alert you to them, you can dispatch crews very early on and very efficiently and fix them all before they actually fail. So you end up producing more power and you increase the reliability because you're fixing them before they fail.
And when you do that, when you drive down the cost of renewable energy and you increase the liability, reliability to where it's equal or greater than traditional forms of energy and forms of energy production, you're moving to net zero. That's gonna achieve net zero because then it's not only an environmental thing, it's a fiduciary thing. It's it's money. And it's an every company in the world is gonna just massively do it quickly, and that is the AI driven digital twin.
That's a big step toward achieving net zero. And this is how a good example of how AI can help. It can help on the front end to make you improve efficiency today and help on the back end on the production of the power itself. Because once it's all green, then the usage of energy from a carbon footprint standpoint is irrelevant.
So if you look at what we do, I mentioned we have eighteen solution. These are our, you know, some commercial AI capabilities that we have across design, operate, and optimize.
Here's our portfolio where we have engineering and execution and operations control and asset performance, production optimization, planning and scheduling, simulation learning.
And we do AI across every single area of that, end to end, design, operate, optimize.
We do things with our e three d, unified engineering, where we've infused AI into that, AI and simulation.
We've got, you know, things where we've infused different types of AI into our SCADA systems. And with the optimized, we've got that's the the biggest area, probably the the area where we've been doing it the longest, where we have automated analytics, no human intervention.
We have guided analytics, point and click. We have what's called advanced analytics, where on a process line, it can optimize things. It can look for anomalies. It can help with throughput and and energy optimization.
We also have a thing called predictive asset optimization, where we've taken predictive analytics a lot farther and integrated it with simulation, physics based simulation.
And what this does is it addresses the risk aspects. So it's great that you know you have a problem very early, and you can take action to fix it because time is money. But when is the optimal time to do your maintenance? When is the optimal time? Should I shut down now or should I wait? Can I make it to the next planned outage?
And is there something operationally I can do different to improve things? And we've done many we have many case studies where this has been super successful to that level of depth, so it's really taking it to the next level. And then we have schedule AI across value chain and and many other solutions.
But let's take a look at some specifics. Let's look at AI in the design space.
These are some of the new things that we're doing. You know, we've we have a lot of history, a lot of depth, but these are some things that are coming out. One is the intelligent point cloud, part of our point cloud manager solution.
What this does is you do a laser scan of a facility, and these days, it's pretty easy to do. You scan the facility. You have a three d, drawing that you can actually walk through. You can do a three d walk through, and it's super cool. It helps Brownfield facilities move into the digital twin space very quickly.
And that's super cool, but the problem is it's a dumb drawing. When you go around, you don't know. It's just a bunch of little dots.
And so, wouldn't it be cool if you could identify what everything is? So when you click on a tank or you click on a pipe, click on a valve, it says all these little tiny minuscule dots. You can see the granularity of the dots. They all fit together.
And this is a tank. This is a pipe. This is a valve. This is a pump.
This is a compressor. But not just any tank pipe valve, a specific tank pipe valve, and then you can tie it to your engineering information through tagging. And so it makes it intelligent, the intelligent point cloud by being able to relate all this, AI infused.
We're gonna be releasing that first commercial release, in first quarter of next year as part of our point cloud manager.
Another area is generative design, and I don't mean generative AI per se with large language models. I mean generative design where it creates something for you. You're a design engineer and you wanna accelerate or maybe you're not a design engineer and you wanna play around. Well, we started we're starting with three d pipe routing.
What if you could just tell it what you want and it designs it for you?
Here, you tell it the start and the end points.
And you say, okay, start, end, but I wanna put constraints because I don't want it to clash with this other piping or these structural elements.
And then I wanna set the plane. Maybe I wanna put it down low or I want most of the piping up high because I wanna avoid foot traffic.
And then, it starts to create in three d these pipes and comes up with different options. And of course, you can you can modify it from there. But what it's trying to do is solve the very complicated equation of minimizing pipe length and minimizing the number of elbows so that you can keep you know, you don't you wanna keep head loss down and and optimize throughput of the pipes. So we're starting with three d, pipe routing.
That'll be released targeting, this fourth quarter before the end of the year that'll be out. And, super excited about it. And then we're gonna add HVAC and structural elements and other things to it as well as we continue to move forward. And we have a big plans for AI infused in the engineering space.
In the operate space, let's look at AI infusion there. We have what's called Vision AI Assistant that we released about three years ago, and we're continuing to make it smarter. And it's natively integrated into system platform and unified operations. And if you look down there, you have anomaly detection.
That's what we started with. This is good. This is not good. You just use a regular digital camera.
And it becomes like a smart sensor, and it can help with product quality. It could help with the space. You got smoke coming in a space or somebody's in a space. It's not what's normal.
It'll alert you. And we have discrete state, like this is open, this is shut, or it's a hopper, full, half full, empty, or or red or blue or And then we added more complicated things.
Like in a in an image, maybe you don't want the whole image to be monitored. You only want you know, the upper left might have foot traffic, you know, people walking by. You wanna ignore that. Let's just look at the lower right.
And then, from a product perspective, maybe I realize there's gonna be defects. There'll be a burr here or there. But some burr is, you know, some defect is acceptable, but then there's a point where it's not acceptable. You can set that.
And then the last thing is user pipeline where you can drag and drop all these different capabilities together to create your own relatively complicated vision analysis. Super simple drag and drop. And then when you put it into unified operations, you can look at it from an operator perspective or an engineer perspective or whoever, technician, you can look big picture, look at the red dots, drill down, see where the issues are, and then you're gonna have visual images because it captures it, it time stamps it, it shows you the visual image of the problem, of the defect, and you can dispatch crews to fix it.
And from an operator perspective, it gives you that visibility, that wide range. Because usually, you're thinking time series data and so forth, maybe AI driven time series data, but this becomes another type of smart sensor, twenty four by seven eyes.
And then in the optimized space, you know, you have we've been doing this for many years where you start with historical and real time data and apply predictive analytics to it, days, weeks, months before a problem occurs. Then we added prescriptive.
What sensors apply to the to the issue? Using AI to determine, is this an outlet pressure problem? Is this a bearing vibration problem? A temperature problem? What what is the root cause? And then what what is the actual physical root cause? Is it a rotor problem or or a, you know, a blade pitch problem or or, lube oil problem?
And then we added and then what do I need to do to fix it from a fault diagnostic perspective? And then we added prognostics. That's forecasting. So predictive analytics is here and now looking for deviations that are likely to get worse.
Prognostics are forecasting, using things like deep learning and statistical analysis, and you're forecasting out what's the remaining useful life of an asset. What are things gonna look like in three weeks? And so it it provides valuable tools. And then from there, we've added, like I mentioned with predictive asset optimization, physics based simulation and optimization to really look at the risk based, aspects.
And we've done more things with AI in simulation. We did we've done what's called gray box modeling where you take physics based simulation and replace some of the components with AI. Good example is carbon capture unit because from a physics perspective, it takes it a long time to converge, minutes or tens of minutes to converge, and that's slow. What if you could use AI and converge in a second?
And then it so the output of the physics based models feeds the AI model and the AI model output feeds back into the physics model. And if you could run this in near real time with carbon capture and other stuff, that makes it really, really powerful. And so that's that's an example of gray box modeling. Many, many components. You know, from a customer perspective, you could build your own and and we have, many models built as well that can plug and play with physics based models.
Then there's autonomous operations, and this is something that's very exciting because, we've partnered with NVIDIA, and there's gonna be a presentation after this. I encourage you to stay. It's in this room on this topic. But just to give an overview, this is taking dynamic simulation training, a rein NVIDIA's reinforcement learning engine to create a brain, to create this super intellect that can optimize set points.
Now, you're probably thinking, okay, well, if you're familiar, yeah, we've been optimizing set points for a long time in steady state, but not in transients. So what about a start up? What about a massive change of feed level or a big disruption and upset in a plant, in a facility, usually take experienced operators hours and hours to get the set points to stabilize things again if they if if they are able to. And in some cases, they have to shut down and start back up.
With this, because we use dynamic simulation and blast it with every possible scenario, good and bad, it knows what the set point should be and it can tell you super fast and it optimizes it. And so when you're training it, here's an example of training it, you're seeing they converge. You're feeding it with every situation, good and bad, and it's gonna converge on the set points for all this. This is happening in the background.
You're just blasting it with data, and then what the result is is this intellect, this this brain that can control transients.
So we've had some good successes with there and we're looking, looking at at, some big engagements coming up. So super excited. I encourage you to stick around for that presentation.
And then last thing I wanted to touch on is unify and humanize. So what do I mean by that? Let's think outside the box with AI.
You know, you probably use dozens of software packages and some you're really good at and some you're not because you had to learn the menus, you got to learn the button clicks, you got to learn the drop downs and how to use them. Right? A lot of training. Why do you have to do that?
Why can't you just ask the system, you know, little Star Trek like, right? So, but why can't you just ask the system, build me a simulation model or build me a SCADA screen, build me a P and ID, build or build me a dashboard.
Why why can't you do that? Well, that we should be able to. We need to lower the barrier of entry for software using AI. And it's not just one type of AI, you use multiple types of AI together.
Well, when you use Gen AI with large language models, you're thinking chat g p t. Well, this is not chat g p t.
This is knowledge linking. This is something that we've it's patent pending for AVEVA, and it automatically looks for relationships among data, engineering data, PDF, you know, manuals. It's looking for events and time series data, all types of data, and it relates them together. And if you do have a data hierarchy or data model, it'll use that as well and and be even better.
And then, we have we have the ability to ask it objective driven questions, complicated questions. So, a simple question would be, what was the maximum output of of unit three last month? You could trend that. You can get that information, right, today using data.
But using AI, you can look at things like, why is my compressor running with lower efficiency now than it did previously? And you can ask it various questions and dig in, and it'll break some of these questions down into subqueries against a large language model. But the key with knowledge linking is we never blend your data with the large language model. We don't train it, and we don't blend your data with any other customer data.
So it keeps secure, but even equally important to security is that it grounds it. We built intelligent agents, intelligent interfaces, tools as we call them, into the various data types of data and it understands industrial data so that it minimizes hallucinations to give you the most accurate responses possible. And this is all knowledge linking. Now, take it to the next level, we partnered with Microsoft and we're working on the generative side of things to do things like automated dashboards.
Describe a dashboard. Give me an energy management dashboard with a pie chart and a stack bar and this and that and these types of tags. And Lori's got a really cool demo to show you on that.
So where are we headed with all this?
So I showed you a lot of stuff. Again, we wanna lower the barrier to entry because when you're talking neural nets and deep learning and genetic algorithms, it can be intimidating. It shouldn't be. You just tell it what you want and it'll give you the answer.
Right? So that'll be the front end to other types of AI, other types of analytics, other types of processes. But we're starting on the data and information level. That's what's commercially available today as our industrial AI assistant.
Ask it questions and if if you're familiar with industry four o, that's the decision twin. Industry five o is AI and and humans working alongside each other.
Then, we move to the functional level, like build me a dashboard or build me a screen.
And then, ultimately, it's gonna converge and be a unified user experience in our connect platform.
So that you it lowers the barrier. So that, you know, when you're coming in, you don't have to be an expert at all these different types of software, engineering software and MES software and and, time series and building screens and and so forth. You can just get in there and just get started. Experts are still gonna use the the menus and all the advanced features, but lower the barrier of entry so it's a unified experience. Let's unify the capabilities end to end.
Design, operate, optimize.
So now I'm gonna turn it over to Laurie to go into more depth on industrial AI assistant and show a really cool demo.
Thanks, Jim. Thanks, Jim.
So for those of you who were here last year, you probably saw we had a pretty ambitious demo that we shared of our industrial AI assistant. Rob and Grievey showed it on the main stage. We showed it in some, breakout presentations.
And then we got back to the office and we realized we had to actually deliver it now, because last year it was just a a demo. But I'm really happy to say that we did as of last week. It's out in production and it's available for all of our connect visualization customers to request.
Some of the key areas I want to talk about with this is really starting with what Jib was talking about that humanized interface, natural language. Rather than figuring out how to build a query or asking for really specific details, you can talk to the AI assistant as though it's a co worker, as though it's one of your colleagues. So that ability to just ask for what you want, and the industrial AI assistant will figure out where to get that information and bring you a nice formatted answer.
The assistant is really good at searching and summarizing your content. So things like your real time tags, your streams, your asset information, documents.
It goes out, it finds it, and it brings it back in an easy to consume and easy to understand format.
We added this year citations as part of our, really we're really been focused on really responsible use of AI, making sure that we're being careful and avoiding any pitfalls because we really wanna build confidence. We really believe this is the key to, you know, our future with this. So traceability and transparency are really important. So we've added citations that give you the date and time that we're looking at, the specific tags, the specific asset information. We're not trying to take the human out of the loop. We want the human to review the information that was used to make your answer, and make sure that it is the correct information, and always validate. And if it's not the right thing, you can go back and adjust.
And last, we have inline generation of content. So if you ask a question, we'll give you a summary things like minimums, maximums, counts. But we also give you right in line a chart so you can visually scan and see the history right there in the screen.
I want to talk a little bit about how we've architected this. It starts on the left with our connect visualization interface. And you open up your assistant and you type in a question, such as, what was the average output of my wind farm for the last week? And this is where it moves over.
It gets passed to our AI orchestrator. And this is really the the whole AVEVA secret sauce here. The the AI orchestrator then passes that request using a secure, link using Microsoft Azure OpenAI, and then the large language model takes your request and breaks it down into pieces. Then we take those pieces and we figure out which tools to talk to.
So we have a different tool that we've developed for each type of data that you have in your connect system. So for example, if it needs to figure out what the wind farm turbines are, it would go to the assets tool, and then that would tell us which tags are measuring the output.
So we'd gather up all of that data, and then we would pass it back to the large language model to help us format it. It would give you the charts, the summaries, and break it down in a really easy to read format.
A couple of other things I want to point out on this architecture diagram.
We've got all of our data. Right now, it's connect data sources, and those will be expanding over time. On the right there, we have what we call our semantic indexing. And we've built a hybrid indexing technique that takes advantage of both semantic indexing, which means it can take into account your intent and your context.
So you don't need to know the specific name of a tag or the specific name of an asset or a keyword. You can type in a word. So for example, there were asking for the output of Hornsea Wind Farm, but maybe your tag is called production. It can figure out that that's what you meant.
So rather than needing to know specific keywords or specific wording, we can figure out what you're asking for and give you the correct information.
The indexer service on the bottom right keeps those indexes all up to date. So anytime there's an update, it keeps it up to date, so you don't have to do anything. It's all managed by AVEVA in our cloud.
We also have the yellow security tag here, which indicates connect user security is making sure that nobody has access to any information that you haven't explicitly given them permission to see.
And then on the bottom, this is new, but I want to point out that we have built this to be flexible and expandable.
So in the future, you will see custom tools that will allow you to get to other types of information, whether it's a partner or some other software. This is an expandable architecture that'll allow, more information in the future.
So as chat g p t rolled out last year, lot of headlines, lot of concerns, and so we worked really hard with both our legal team and our technical team to develop a set of, good practices to make sure that we avoid a lot of those pitfalls. So starting on left to right here, when we talk about guardrail ground guardrails, we set the tone. We introduced, the assistant as a helpful assistant and the focus is really on industrial questions. So if a question is asked that's not, in the industrial space, we're just going to tell you, as an AI assistant as an industrial AI assistant, I don't I can't answer that question.
We have a built in critique to keep the, cycle, efficient and make sure we're getting good answers. In the middle, when we talk about grounding, Jim talked about this, but it's really important. We're getting the answers from your data that's in connect. We're not going out to the Internet and searching for other information. We're using your data as the basis for these answers. So that really helps us get good answers and the right answers, and then we let you check those answers with our citations.
So take a look at what tags we used, what time range, what assets, make sure that we did it the way you would. And if we didn't, you can adjust your question and make it what you need.
And last but certainly not least, security and property.
The LLM Jim mentioned this as well is never trained on your data. It's stateless, it's fixed, it's sitting there, static.
And when your data is, sent to it, when we're evaluating your question or looking at things, we have an agreement that your data will not be used for training and it will not be saved. It will be immediately thrown away.
And then, the user context part again, connect user security keeps everything, nice and safe.
So let's look at a little demo. This is a new demo for this year. And, we've integrated ETAP data, our electrical system, into Connect for this demo. So we start by saying, I received an alert from Hornsea.
Are there any new events that I should take a look at? So again, we send this off to the, LLM to figure out what data sources we need to access. It goes out. It finds a bunch of event data.
It's giving us a summary here of total durations. It's telling us, what the status was, you know, why the breakers were down. Nice and easy to see. You can see there's links there to click on the assets to get more details.
We noticed that there's, eight turbines that are tripping with open breakers, and so we can ask a follow-up question. What kind of faults, you know, would cause this? We get, some more specific information from our document system.
We learn what the common causes of this might be.
And next, we can follow-up and, we ask for an SLD. And for the electrical engineers in the crowd, we'll see that it knew that an SLD was a single line diagram, and we can open up this, diagram and we can look at the connections and how things are related. We could also have clicked on some of those links in there if we wanted to go deeper. But now that we realized that our, wind farm is at a lower output, we also know that this is going to impact we're gonna have to buy power because we're powering a, a plant with this. It's also gonna impact our, sustainability measures. So first, we're looking at our import export power here. And so we see that we're importing power, and so this is costing us money.
And next, we're gonna look at our sustainability impact here. So can we compare our c o two for today versus when we were not having this problem?
And it goes out, it grabs the information and it's telling us the impact and that, yes, our c o two is higher, and it gives us a nice analysis and a summary. We get some charts in line so we can see what the history of it was.
And now, we're gonna show some new capabilities. This is our new dashboard generation capability, so we've asked it to create a dashboard with a line chart and with some tags on it.
And it brings up, a link to our dashboard. So it tells us, here's your dashboard. You can click on the link, open the dashboard.
Next, we can create some modifications to it. So we want to change the, access set up here. So we're going to ask it to change it from a single access to a multiple access trend. We're going to change the title of the graph.
So we send the request, and we apply the update to the dashboard.
Now, we're gonna add a pie chart to the dashboard.
We tell it which tags we want. We give it some specific instructions. And these instructions are pretty specific right now. Like Jim said, we're going to go to more implicit instructions.
But the beginning part here is, you have to pretty much tell it exactly what you want to do. But again, in English, you don't have to figure out which buttons to push or how to do it. So, now we've got our dashboards. And now, one of the really nice things, with using this is now we've had this conversation and we can ask it to summarize the conversation.
You know, give me notes of what we've talked about so I can share these with my colleagues.
And so, it gives me a nice summary of all the information we looked at, the things that we did, in a very easy to consume format and we can copy and paste this and send this off to our colleagues. So I'd like to share the re the release plan. As I mentioned, we're in we were in Lighthouse in September with a set of customers. We had several customers using this, with the basic capabilities of connect data streams and events.
Right now, this month, we're releasing kind of the full capabilities that we planned from last year, which includes documents from AIM, being able to ask specific questions. In the demo Rob showed yesterday, he asked a specific question in a document, asking questions about the connect visualization help.
We've added chat history management, which allows you to go back to previously asked questions and answers, and also to favorite questions that you, want to maybe ask again another day, see how things are going. We're introducing guided examples. When you start using these tools, you know, the first time I use ChetGPT or Copilot, you're not sure where to even start. So we're going to provide some starting points, some easy click on, you know, show me this, show me that.
And then also, voice to text capability. So if your hands are dirty or you don't have access to a keyboard, you could, use your microphone to ask a question. And then from there on, we'll be on a continuous release cycle. Connect visualization releases about every three weeks, and we'll be introducing new capabilities right along with that release cycle.
We'll be working toward the generative dashboard capabilities, which should lead into many other generative capabilities.
The dashboards will be first, but will lead into creating other things as Jim mentioned. And then, we'll also be adding more connect data sources and integration capabilities and integration into search.
And, with that, that concludes the industrial AI assistant. Thanks so much for your time. If you'd like to learn more about what we're doing here at AVEVA with AI, we encourage you to stay for the next session, which is a co presentation between AVEVA and NVIDIA.
There's also a presentation this afternoon, with Microsoft with some sustainability, information. And then, another one, with predictive analytics and AVEVA and Drax presenting together. So, thank you very much. We appreciate your time.
