AI Integration for Danone’s Factory Automation Data Management Tool?

Jon Mills, Rockwell: AssetCentre also has huge potential in autonomous systems that are AI-enabled, and AssetCentre can act as a source of context and data for these AI agents to be able to consume. When you think about these language models that are becoming popular, and the types of applications that they start to enable, they’re built and trained with very large sets of data. This allows them to operate in a large number of domains, many of which we're taking advantage of in industrial automation. But in order for them to be effective, they need access to the information to be able to actually perform some sort of function on and so having solutions like AssetCentrethat are organizing and contextualizing the data, those are going to be primary inputs for the future of AI within these networks and systems.

Williams: I could imagine asking AI integrated within AssetCentre, ‘where’s my highest risk controller?’

Or, ‘what’s my oldest stuff--where's the obsolescence?’ The potential’s there since we have all information in place, as long as it uses the data effectively.

Mills: Bringing AI into a security posture is definitely interesting. One of the things that it enables is to become more reactive to any type of security threats or challenges that are happening within your network. That’s because these models are able to consume many different types of inputs, which means that you can feed it essentially logs from a FactoryTalk Security server or out of AssetCentre indicating that these are the types of operations that are being performed. And you can actually have the model grade or evaluate those logs, and then issue warnings if it sees things that are out of specification, or sees things that should not be occurring. And those processes can be done autonomously, with AI integration.

Today, these things occur, but they often weren't known until months later, when the damage has already been done, or they're not known until they're they start to have some sort of actual impact on your process and your ability to get work done.

Moderator: Which brings up security, right?

Williams: Cybersecurity is a big thing, and AssetCentre has been instrumental, because one of the first components in cybersecurity is to ‘identify.’  Everything that's on the network, so we have all the users and all the files. So it helps people organize. The second part of that is computing all that data. What versions are things? Where is any high-risk firmware, and minor revisions that we need to patch up? Where is something being thrown into the log that we don't notice because we’re looking at everything else. And it could give us information about where there's an event or the problem that is occurring, and we don't even recognize it yet.

Lua: Another piece is limiting access. Given that AssetCentre is such a central point for our organization and systems, we need to be able to limit who has access to which files. So you want to have Vendor A having access to Vendors B’s PLCs or their electrical schematic. So with AssetCentre and FactoryTalk security, we're leveraging those two in order to limit certain vendors to only have access to their equipment, versus giving them broad access to all our equipment. We're leveraging FactoryTalk security at the PLC level to say, ‘you are vendor A, so you only have access to these machines in these facilities.’ Or if I have a technician, and they're an apprentice, I want them to be able to go into a PLC and troubleshoot, but you don't necessarily want them making changes. There are ways to facilitate that. Those are the things that, from a cybersecurity risk standpoint, we're working to narrow how much people have access to.

Williams: It also blocks changes. So if a technician comes in, and they try to plug in to change something [that we don’t want them to change], it blocks enough from that. So everything's maintained at the AssetCentre level and that’s good for our system. So we can see all the revision history.

Moderator: So with all the changes that are coming [in coming years, such as potential AI integration with AssetCentre], would Danone make use of them in AssetCentre?

Williams: I’d say so. Some of the [other Rockwell] technologies like the CoPilot in Logix Designer, if you're troubleshooting an issues somebody else was working on, you can ask it, ‘where did the last person leave off? Where are the most changes, what program is this in? Just that time savings alone and getting back to production

Moderator: With version 15 of AssetCentre, they started with life cycle projections. Right now it's limited, but it's only going to improve more, with more details about each life cycle of each product that's on the floor. How do you use AssetCentre with life cycle management and obsolescence?

Lua: For life cycles, AssetCentre does have some features where it tells you what devices that are active and mature that you've configured for end of life. But the process is very manual to update. We have a database—with every component at every facility—that we update every couple of years to ensure we know where our obsolescence lies and our biggest risks.

If we can leverage AssetCentre a little bit further, since everything already resides in there, all the programs are in there, and it contains all the data that we need, in order to simplify these obsolescence sites, that would make it much better.

Moderator: From Danone’s point of view, what should Rockwell be focusing on with improving AssetCentre? What problems that you're currently facing, and how can we solve those problems?

Williams: One is to do more reporting, or be able to answer those questions that, when targeted correctly, give us a lot of value add, specifically around program changes. Who made what changes where?  That's a big chunk of it. The other part is reporting on who has you know what PLC check out, what files are blocked, and if it was a public point of access. We try to keep availability so [equipment] can be serviced. So having some of that reporting built in, having things looked at in different ways that aren’t are just analyzing data readily available. The reporting features that we do use now, we use them pretty extensively, but with auto logs and that querying data just so many more features that [could get those] questions answered.

Lua: Right now, we have so much data, we have a lot of reporting, so we can go in and analyze our systems, find out who's making changes and what programs are checked out. But it's the process to do is manual. Someone has to actively go and run these reports, or put them on the schedule and look at them in email.

If there were some type of dashboard where we can quickly see the main things going on with our system, telling us, ‘we have so many PLCs checked out,’ or ‘these are the PLCs that have had the most changes recently.’  That would help us quickly change our plan and say, ‘Okay, let's focus on these.’ Or, ‘these are our biggest vulnerabilities.’ [A dashboard feature like that would] give the tool more features that allow us to quickly narrow down what we need to focus versus having to do some of these additional searches.

Moderator: [summarized] From the Customer perspective, what are your next steps with AssetCentre and growing with the evolving platform?

Williams: We use AssetCentre quite extensively, and a lot of these [soon-to-come] features that we've discussed are ones that are making using AssetCentre even simpler and quicker. Right now, our network architecture drawings are manually created. We want to have these up to date, and have the drawing for our vendors or for troubleshooting, or just for reference. If we can tie this into IAB [Integrated Architecture Builder, a Rockwell tool] and have them auto update for us, that would save time, as well as audits for security and things like that. If there was some way to use AI to say, ‘what PLCs have been unlocked and why?’ Something quickly to get response so we can quickly tackle those.

Lua: Troubleshooting, too. Thinking about anomalies, and being able to query them as symptoms of a problem. AI can make that connection quickly. I'm sure not everybody's reading through notes on anomalies every da. So as issues come up that are unknown, AI could give you that solution faster.