2025 - AVEVA World - San Francisco - AVEVA PI System
Operations on-ramp to AI and advanced industrial analytics
Industrial companies face many challenges such as sustainability, production quality, reliability, and maintenance optimization – all while needing to maintain profitability. Finding an optimal solution for this multifaceted problem is best achieved with a unison of powerful AI-driven analytics and the contextualization provided by the AVEVA PI System. Join us in this session to learn how you can leverage AVEVA PI System with AVEVA Predictive Analytics to identify impending asset failures in advance, allowing for proper planning of maintenance activities while minimizing impact on production. We will also share how AVEVA Advanced Analytics helps in identifying the optimal operating conditions for increasing product quality and recommending impactful process adjustments.
Company
AVEVA
Speaker
Alex Jenkins
Alex Jenkins has a degree in Electrical Engineering, and has been working in the industrial analytics space for over 15 years. As a member of AVEVA's AI Center of Excellence, he focuses on real-world deployments of AI-driven analytics solutions to enhance the reliability, performance, and efficiency of industrial assets.
Company
AVEVA
Speaker
Ilaria Michelizzi
Ilaria Michelizzi is Presales market leader for Southern Europe working for AVEVA since 2019. Her team of expert consultants supports customers in streamlining engineering and capital project execution, adopting industrial digital twins to achieve the operational excellence, optimising the value chain, improving the maintenance practices, and driving sustainable use of world’s resource through AVEVA portfolio solutions. Her 20 years’ experience spans from large EPC projects to digital twin & AI solutions adoption for continuous improvement in the energy, process, and manufacturing industries.
Session Code
SESS-113
Transcript
Hello, everyone. Thank you for being here today in this amazing session session about how to really drive value from operations to our AI and analytics. So my name is Ilaria Michalizzi. I am presales manager for Southern Europe. I've been covering different roles in my career from performance engineer, rotating equipment engineer for EPC companies, and then excited to move to digitalization, digital transformation to predictive maintenance programs, and joined AVEVA as a APM presales consultant. So today, with my team, we drive the full scope of AVEVA portfolio, digitalization of customers. I'm here today with my colleague that I'd like to introduce.
Hi. Good morning, everybody. My name is Alex Jenkins. I'm with AVEVA's AI Center of Excellence, and my background is really in the application and deployment and and monitoring of of our analytics packages, which we'll talk about here today.
So first of all, let's start getting into our landscape. We know that we want to drive innovation, and currently, the industry is under reshaping. We no more have the typical industry, I would say, that we have traditionally seen. We need now to readapt in an environment that is continuous changing with electrification, with movement towards biofuels, and hydrocarbon industry needs now to reconvert towards an acceleration of decarbonization.
So renewables are now taking a lot of leap behind.
So for doing that, we know that data are very much important, but fewer than ten percent of the companies are really enabled today in taking decisions from their data, which is, I mean, a lot of potential for the future.
So how can we consider AI in this landscape? What AI can do for me? Okay? What what can do for you, for your companies? So independently on the industry, AI can support with predictive maintenance programs, with process optimization, with improvement of the energy efficiency of your facilities, enabling quality at the first time and optimize the supply chain. And for sure, if we get this, we can also reduce the waste. We better use our resources and also enable the new technologies, right, robotics, like also data driven decision making with digital prints, and improving this way the safety, the way we execute our task on a on a on the field as well as the environmental and compliance impact.
So in AVEVA, we've been driving data infrastructure enablement, I would say, for our customers in many years, And we know that the PI System is the foundation of this cutting edge intelligence. And as the base, I mean, once you have the data, a lot of potential is open to you. We can unlock additional value. And in AVEVA, we've been doing that for many years with our premise solution that we're going in enabling energy efficiency to PI Vision, for example, or supporting predictive analytics for getting reliability improvement, as well as advanced process control for stabilizing your processes.
All these was, I mean, adapted to the past economy because nowadays, we need to interact in a a very dynamic environment. So we need this data to be made available for creating communities and share the data that are only necessary for your partners, for your data source, I would say, of material to get the right decision to provide on time what you are necessity for your production. And this through an environment that can enable with connect, also not only data sharing in a secure manner, but also self-service capabilities of creating dashboards and leverage the digital twin with industrial AI assistant that we've been seeing in the last keynotes that can enable an investigation and understanding through natural language processing of all what outcome you are looking for and drive AI through a self-service, I would say, fit for purpose analytics.
That's what we are enabling so we can have perfect batch use cases. We can have energy efficiency, predictive quality, all at your fingertips.
This is what is really going to enable a connected industrial economy, Been sharing those data with your partners, with your data source, and with the different actors that are working with you in the industry transformation for enabling value.
But let's have a look. With PI System, as we say, we have very open, I would I would say, use cases and possibility of creating value. We've been doing this already with PI Vision, as we said. So you can now standardize to asset framework, different use cases, and deploy very quickly at scale.
We can leverage predictive analytics, thanks to an integration that now is native directly with both archive and asset framework, and the automated model building for creating models, AI models very quickly. We can enable also connect advanced analytics that can also provide self-service optimization of processes. We still see that. And also use this data for third party applications.
This one is Apollo, but we saw also Databricks. We have the possibility to leverage partners' application, sharing the data that already are necessary for those use cases, and accelerate the value through the AI assistant.
But let's see in practical how we start the journey.
That's right. Thanks, Hilaria. So how do you take the data that you you might have already in your your PIE system coming in from the various components at your facility?
And you start with that and the PI System can organize and contextualize that data, for example, with an asset framework hierarchy. And that's really key for deploying and developing AI models, right? We heard this morning AI models love good organized data, and that's very, very true. And it's not enough to just have a model sitting off in a corner somewhere. You've got to be able to visualize and understand and interpret the results of those models.
And so these tools can have their own native interfaces, but they also can write back to the PIE system so you can view the results of these models alongside the PIE Vision screens that your teams are already using. And so they can make quick and accurate decisions based off that data.
And AVEVA has a wide range of tools that can bring improvement to your business.
There's a couple we're going to talk about here today. For example, looking at the production process, you can think about improving your quality or your throughput or your energy efficiency. And on the maintenance side, you can look at having improved uptime and reliability of your assets.
And speaking of maintenance, maybe something like this has happened to you, hopefully not, but it does happen and it can be quite an unpleasant surprise when something like this happens and very, very costly, right? Just in terms of not only the cost to replace the equipment, but the lost production and associated downtime with that can be quite expensive. And these are often unplanned events clearly. So if you think about, if you could detect these kind of things early before they get to this stage, you could really avoid a lot of that unpleasantness. And this is where some AI driven tools can help.
Traditionally, you have a lot of data. In a historian, you might set single sensor univariate type rules. That's great. But in order to detect problems earlier in advance, this is where AVEVA predictive analytics comes into play and it works by looking at the ensemble of the sensors on your equipment and building this multivariate model to understand how they all relate to one another and be able to detect small deviations before they become too severe.
And this is done in a no code fashion. You don't need to be a data scientist or computer programmer or anything to use this. These models are very easy to use. They're template driven so you can deploy them very quickly across your entire fleet of assets.
And surrounding these artificial intelligence models, we've worked very hard to develop these workflows that can fit in with your existing business. So case management, alert management, and that really helps your teams get the most out of these models.
And of course, it's natively integrated to the PIE system. So predictive analytics reads from and can write back to the PIE system. If you have a suitably configured asset framework hierarchy, you can even use the bulk model building capability to very quickly take those elements you have in asset framework and mirror them almost, if you will, into predictive analytics models.
And because the data is written back, you can view those results directly in PI Vision. And we'll see an example of that here in just a second.
But before we do that, I just want to dive into a little bit more detail about how predictive analytics works. We have the key output from these predictive analytics models, kind of what we call an anomaly index, anomaly score, asset health number, and that's telling you basically how far away from historically normal trends you are on this piece of equipment. And so not only does it tell you when something's starting to go wrong, it'll specifically indicate which sensors are driving that deviation so you know where to look.
And the fault diagnostics engine that's built on top of that looks at the pattern of of which sensors are starting to deviate, how they're deviating, and can tell you what the likely cause of the problem is as well as give you some prescriptive actions and next steps to take to remediate the issue.
And once you've found that issue, you can use the forecasting engine to then project out and understand how much time you have to act, to plan your maintenance, to get the spare parts, whatever you need to do so you don't surprised by these issues before they happen.
And of course, that's kind of the core modeling capabilities. But as I mentioned, all around that there's case management, there's transient analysis. If you have data scientists on your team, you can use custom algorithms and integrate that here as well. And of course, it's all template driven. So it goes very, very quickly to deploy these models.
So let's look now at an example of what this looks like. So we start here with a PI Vision screen here, for example, on midstream oil and gas pipeline. We can look we've got a lot of our numbers here.
And if we want to drill into one of the pumping stations, we see here that asset health index, as I was mentioning before, for all of our pumps, obviously, one's flashing in red, that's probably not good. Red is never good. So we've got, you know, an indication there, an early warning of a potential issue on this pump one sixteen a. So we can kinda drill into that, get all the details on this pie vision screen.
You see all the data, sensors, instrumentation on this pump. That's what's feeding into the predictive analytics model. In the upper right, we've got the asset health kind of area quadrant on here. These are the outputs, the main outputs from the predictive analytics model, including the anomaly score, some fault diagnostics information.
And in one click, we can go directly to the predictive analytics application and see all all the details about this model. Always at the top is that asset health score. We can see that that's starting to starting to go up. We're starting to have some deviations on this pump compared to the historically normal behavior.
And again, this is a multivariate model, so we can see which sensors are driving this deviation or causing this. We see there's actually many contributors here around a lot of the bearing temperatures starting to deviate. So we can look at those. We can drill in, look at the trends for one of them. We see the actual value compared to the model predictions and we can see that that's starting to separate away from each other, that deviation indicating abnormal behavior. If we wanna see the what the fault diagnostics engines thinks the issue is, it's saying we we likely have an issue with our motor bearings. So we can drill into that and get a lot more information about that, including which sensors are involved in it and what the what the prescriptive actions we should take to kind of start remediating this issue are.
And again, these are configured by subject matter experts. You can AVEVA has a library of fault diagnostics that we we offer and your teams can add to that and configure that for your own specific applications. Here we see all the possible faults on this pump, on this motor, and can understand what might be driving them.
And now that we've kind of seen we have an issue, we can go over to the forecasting tab and start to try to understand how much time we have left before we reach a critical limit like our trip limit, for example. Let's say it's a hundred degrees c. So if we wanna know how much time do we have to react, we can see it here and know how urgent this this issue might or might not be. Then what we can do is categorize this issue.
We can open a case so we can start communicating with the rest of the team about this so that they can take action. You know, we can do things obviously like adding a title, adding a description about what we think might be going on here. We can set some categorizations if we want to and assign it to the maintenance or operations team as needed so that they can go address it. We can leave some kind of discussions like a rolling log of what's going on with this.
And, you know, we've asked the site to go, you know, take a look at it. Maybe they can come back, leave the comments with what they found, what's causing this rise in bearing temperatures across the machine.
And, you know, perhaps it's a it's a plug just a simple plug strainer on the on the cooler. So not a not a big deal, but we found that early and they're able to go take action before it actually causes any any damage to the bearings itself.
And back here, we see that reflected on the PI Vision screen, the asset health score is back to normal, the pump's green.
It's not red anymore, so that's always good to see. And then finally, we can come back to this case and close it out. And we can leave some information about how the issue was resolved, what was found. And these cases are all searchable in the application so you can start to build up this library of knowledge about previous issues you've had. So the next time you have an issue on this pump or similar pump, you know exactly exactly what the problem was and how to address it.
So, again, that workflow is so key to using these AI models. And, Olari, if you could help explain what that workflow looks like.
Yeah. I was I think that what you saw, I mean, is incredible way of being driven in the decision making process. So what we saw is that we've been able to identify an anomaly, to inform the monitoring center, to start investigating and diagnosing the problem. In this way, very quickly, they would enable also to start engaging with the maintenance team and decide which type of risk and the time to failure was to be considered for preparing the intervention, inspection, and afterward, notify them for taking action in the right moment with the right spare parts, with the right crew involvement, maybe also using the same team for doing different activities without impacting the production constraints.
And what is important is that after the maintenance is done, this this knowledge remains within your team, within the library, and you can also follow over time the return on investment of this type of of predictive maintenance program, keeping track of the avoided failure cost and also the avoided production losses within the library. So at the end, the big advantage of this is that we've been following very long, I would say, deviation over time. Now let's have a look at a completely different case.
If we have a process, okay, in which whenever there is a deviation of behavior of an asset over a production line, and what does will it happen? Now we know that if there is a deviation with respect to the performance or reliability issue that makes a small, maybe, pumps to stop, the full production will be discarded. So in these cases, we don't wanna keep track of the issue for months. We want to act now.
Okay? We want to be informed immediately of the deviation that is going to start because maybe we're gonna stop our line in few hours. And that's why for this type of approach, we go with cloud and SaaS solution because in this way, we can interact very quickly and get value immediately. So what we are looking for is improving productivity.
Okay. Get the right throughput of our of our production.
We want to improve the compliance and the quality of your production and get in this way an impact on the environment that is minimized. So use the energy, the right quantity of energy, for the quality and the productivity that we're looking for, reducing in this way the waste from the source, making the right product at the right time. So how do we do that? Let's see a practical example.
Yeah. That's right. And it really all starts with developing a baseline understanding of how your process is operating today, right now. How have you been doing with creating this digital twin so that the model can understand what is normal, what are you doing right now.
And then the next step to achieving those outcomes Elaria was talking about is really to make sure your process is running stably, to achieve this stability. And so you can apply some simple AI models with advanced analytics to do things like anomaly detection and to understand the kind of ideal conditions of your process. So where where is a good center line for you to be running based off of how you have been. And from there, you can add more and more complexity and add and increase the number of models, the capabilities of the models that you have, including top driver identification.
So to take an example with if you're trying to look at improving the quality of your process, If you want to understand the parameters that are really driving and most influencing the quality of that process, you can use advanced analytics to do that. It will then give you some recommendations if you start if those parameters start deviating from what's ideal, it will give you recommendations about how to get back on the center line. And there's some dashboarding capabilities that allow operators to see that recommendation in real time and so they can take action.
And eventually, once you build trust in this model, you choose to, you can even enable autonomous control or right back to the control system if you need to.
And so let's look at an example here. This is on that beverage line that Alaria showed earlier. We've gotten an email here notifying us of some abnormal behavior on our beverage line and it looks like our syrup flow is is out of spec. So we can see we've got a couple simple parameters here showing that they're all running well, but our syrup flow is much, much higher than it should be.
So let's figure out why and what's going on. So in this dashboard, we can understand right away what product we're or what recipe we're making right now. We can see how much time we're spending in the various production phases. And this is important context to bring into the model because depending on the product or recipe that you're making, the stage of production that you're in, the ideal conditions might change.
And we can see that reflected here during in this graph, we can see as the product changes, our target sugar content in the drink is changing to reflect that.
And so as we start to investigate this case of high syrup flow, we start to bring together various pieces of evidence, if you will, and put it into this case. So, we can show that that analysis. Here, we see the some of the top drivers or main influence on the Brix or sugar content, and it happens to be syrup flow. That makes sense, right?
The more syrup you add to the drink, the sweeter it's going be. But this is what has come out of the model. And so if we send this as include this as another piece of evidence into that case, the operations team can know to go adjust that syrup flow. So we can just easily add it right back to that same case.
And we can come back here to this ideal conditions screen and we can add this to the case as well. And what this is really telling you is where you are right now versus where you need to be. And we can see that the last syrup flow from the last stage of production was pretty good. It was right on target.
But currently, our syrup flow is too high. So perhaps we have a valve that got stuck open or something and didn't close successfully as we move to the next stage of mixing here on the on the drink. So that's what we kind of want to end up highlighting to the operations team so that they can go figure out why the the syrup flow is still so high and why it's stuck there. So we've we've got these pieces of evidence in here.
We can go and just easily at mention someone from the operations team, ask them to go take a look at it. They'll get notified. They can come in and see a list of all the cases and they can come in and see live values right away in the same interface that we sent it over to them in and look at this information. We see now our even our agitator speed is a little off of normal.
But if they're able to go then and get that current information and address it, they can come back and save the batch before they have to throw the whole thing out because it's too sweet.
Yeah. So let's recap what we see we saw here. In this case, the reactivity of the operations team, as we say, should be immediate. So we receive a notification in any device because we are in a in a SaaS solution.
We can receive it on our mobile, and then immediately start understanding about the issue for the specific product that we are producing right now, and for the specific phase of operation in which we are. Maybe we are in a transfer phase of product, we are in a filling phase, we are producing a specific batch, and we can readapt our process for then getting the right recommendation to our operations team and readjust our production to avoid any discard of production. So I think that's very powerful. And what it it tell us is that whenever we have the right solution in front of us that is user friendly, that can interact with our team, we can immediately gain value on quality, on energy efficiency, and productivity.
So these are some of the values that we are seeing from our customer using these analytics solutions. We can get up to hundred percent quality improvement. Okay? So get right from the first start our quality targets.
We can get reduction of maintenance cost to thirty percent. And for sure, the downtime, so preparing our our downtime related only to the planned condition that we want to consider, reduce the energy cost up to four percent. And you can tell me, okay. These are numbers.
Where they come from? Well, they come from our clients.
Okay? These are just a few of the customers that are using our solution. I I've been selecting PIE customers using our analytics solution. A good one that I like is Axiona.
They are using advanced analytics for improving the the solar the solar desalination plant's conditions. What they say that our events in in Madrid, I I remember I was there, that finally, their production plant manager was able to sleep. Because they were enabling in some way an automatic, I would say, understanding of the set point for their plants. Another good example here is Azarco.
I mean, they are working on mining, and they've been enabling pie pie use cases first for getting enough historical data and then creating value through predictive analytics for the critical equipment. So you can start maybe on condition based maintenance with pie, but then you need to look forward. Need to start having historical data for the critical component that will gain much more ROI if you go forward. And then, I mean, I will share this slides with you so you will have cases on pharma, on power, renewables, on energies, and come and see me this afternoon.
We have a panel with SCG, so you will hear directly from them.
I think that what is important so at the end, when you start this journey, start small. Find the the the, I would say, the location, the outside there for you, the most critical to start gaining, I would say, in perspective. But always think big. You need to consider your data as the strategy for the future of the new cases that you want to deploy.
Then for sure, PI System will drive key ROI on cases that very easily can be understood in terms of behavior, creating rules, OEE rules, condition based maintenance rules, etcetera, for gaining visibility. And then the money that you gain, you will invest in predictive analytics or advanced analytics for improving sustainable goals and accelerate predictive maintenance programs with results that can be driven oriented. Okay? You need to close the loop of the maintenance strategy in order to get results that are actionable.
And that's what your data can provide. Thanks to the prescriptive capabilities that we are enabling, the templatization approach that can, in this way, also accelerate your best practices infused into the software. So what I would say is that the real difference between failure and success is the way you operationalize your analytics program, and AVEVA is there for helping you with it.
