Industrial AI Doesn’t Need Another Screen

Eight dashboards, eight different systems, and one process engineer trying to work out what his process is doing. He works at one of our customers, and he isn’t short on data. He’s short on time, and on a clean path from what he’s seeing to something he can change.

I think about him a lot. The plants we work with have invested in sensors, historians, a CMMS, and HMIs, and each of those does its job well. The trouble shows up in the handoffs: an alert in one tool, a work order typed into another, and nobody quite sure a month later whether the fix held.

So when we design an agent at Augury, one of the first questions we ask is where its output will land. If the honest answer is “a new screen,” we usually go back to the whiteboard. Four ideas keep coming up as we build, and at Rockwell Automation Fair in Boston, November 16-19, we’ll show each of them running on systems plants already use.

1. Put guidance on the operator’s screen and let the operator decide

Our Augury Operations Agent watches how the process is running. When something drifts, it looks for the cause and recommends which settings to adjust, and by how much. But the recommendation is most useful if it reaches the operator where they work. That’s why the agent runs as a workload on FactoryTalk Optix, using live process data to deliver actionable recommendations directly to the operator’s HMI.

What makes those recommendations worth trusting is what sits underneath. The Augury Operations Agent is built on the industry’s only human-readable industrial context graph: a map of how your assets, signals, procedures, and operator know-how connect, written out as a plain-English Process Wiki your engineers can read and correct. A complete trust layer for Agents that hasn’t existed before. When the agent suggests a change, it’s reasoning from context your own experts have checked.

The scenario we’ll show comes from a mineral flotation process. Recovery slips below target, and the agent traces the drop to a cause and suggests set-point changes. The operator pushes back because one of those levers can’t move today. The agent works around it and comes back with a different plan. It’s my favorite moment in the demo. Operators know things about their line that no dataset captures, and the agent adjusts to that knowledge.

See it at the Fiix Innovation booth (6).

2. Make agents show their work

Reliability engineers are skeptical by training, and they should be. Our Augury Reliability Agent can run a root cause analysis from a plain-English request. It asks what an experienced engineer would ask about what failed, what changed recently, and what the crew suspects. Then it works through the machine’s own data and the maintenance history in Fiix CMMS, so the analysis covers both what the sensors saw and what the crew did. An investigation that usually takes days takes only minutes, which matters because RCAs often get skipped except on the biggest failures.

Every claim in the analysis points back to its source. A cause gets marked and confirmed only when two independent pieces of evidence back it up, and anything the agent couldn’t verify is labeled that way. Engineers edit and sign off before anything is final. I’d argue the “couldn’t verify” label matters as much as the analysis itself. People trust a tool more when it’s clear about what it doesn’t know.

See it at the Fiix Innovation booth (6).

3. Get more out of hardware that’s already installed

Many facilities already have protection systems on their critical machines, and we wanted to see what our models could do with that data as is. We took recorded data from Rockwell Dynamix 1444 on a paper line and ran it through the same Industrial AI we use with our own sensors, with no reformatting.

The model picked up a recurring unbalance on a roller motor. Paper dust was getting pulled into its cooling fan and building up. By generic vibration limits, the motor looked fine. Against its own history, it was running about three times higher than usual. The maintenance team found the dust, cleaned the fan, and the readings came back down.

For sites with these sensors already installed, that could mean earlier warnings and a specific next step from hardware they’ve already paid for. 

See it at the Dynamix 1444 booth (153).

4. Turn insight into a work order, then check the repair

A diagnosis only helps if someone acts on it. When Augury flags a developing fault, it can open a Fiix CMMS work order with the fault, the priority, and recommended steps, once one of our reliability experts has reviewed it. The technician closes the job the usual way. Then we analyze the machine again to see whether the problem is actually gone.

That second look is easy to skip, and it’s the step we care most about. A closed work order tells you a job got done. A fresh analysis tells you the machine is healthy, and every confirmed repair makes the next recommendation a little sharper.

See it at the Fiix booth (10).

Come find us in Boston

You’ll find Augury at the Fiix booth (10), the Fiix Innovation booth (6), and the Dynamix 1444 booth (153). Some of what we’re showing is already in customers’ hands. Some of it we’re still shaping with early partners, and that’s the part I most want to talk about.

If you’re working out where AI fits in your operation, stop by and tell us which machine or process gives you the most grief. We’d much rather hear that than give you a pitch.

Not making it to Boston? Request a walkthrough, and we’ll take you through the demos.