AI, digital twins and IT/OT modernization: what a mid-size plant should do first

AI and digital twins get a lot of attention. Here's the less exciting work that has to come first, in the order we'd do it, and what those tools are realistically good for today.

A lot of noise

AI for automation, digital twins and IT/OT modernization get a lot of attention right now. Some of it is real and useful. But for a mid-size plant with a mix of old and new PLCs, the next step usually isn't an AI project. It's making the data and the network trustworthy enough that any of those projects could work.

This is an opinion piece, not a tutorial, so there's no code on this page.

The order we'd do things in

  1. Know what you have, and back it up. A list of every PLC, HMI and drive with its firmware and where the current program lives. If a controller died tonight, could you restore it? (Our free checklist starts here.)
  2. Make the data trustworthy. Consistent tag names, engineering units and timestamps; counters and totals that don't overflow or reset; alarms that mean something. Analytics built on bad numbers give confident wrong answers.
  3. Connect it cleanly. Standard, secure connections such as OPC UA from the PLCs to SCADA or a historian, instead of one-off links. (See our Siemens communication guide for one example of choosing a method.)
  4. Segment and protect the network. Separate control networks from business networks and control what crosses between them. NIST SP 800-82 and the ISA/IEC 62443 series are the usual references for this.12
  5. Then try the new tools, on one well-understood problem with a clear measure of success.

Digital twins, plainly

"Digital twin" covers everything from a full physics model of a machine to a simple simulation that lets you test PLC logic before it goes to the floor. For a small plant, a sensible starting point is the simple end: an emulated controller or simulated I/O so changes are tested before a shutdown window, not during it. A full model is worth it when the machine is expensive to stop or expensive to get wrong.

AI, plainly

AI tools can help with things like writing first drafts of documentation, searching manuals, or spotting unusual patterns in trends. They don't remove the need for someone who understands the process to check the answer. Any AI-suggested change to control logic should be reviewed and tested like any other change, and nothing in a safety function should depend on it.

Where to start

Pick one line. Check its backups, its data and its network. Fix what's wrong there first. That work pays off whether or not you ever start an AI or digital twin project, and it's what makes those projects possible.

Sources

  1. NIST, SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security (2023): csrc.nist.gov.
  2. ISA, ISA/IEC 62443 series of standards (overview): isa.org.

Not sure where your plant stands?

A free health check is a good start: onsite in California's Central Valley, or a free remote review if your plant is elsewhere in the U.S.