The incident review team determined that all operators should review their findings: An errant reading on an O2analyzer caused the air/fuel ratio to be abnormally high, and the temperature control kept firing harder to maintain temperature. Not only was it wasting fuel, it began increasing NOX emissions until they were approaching the permit limits. Fortunately, a veteran operator arriving at shift exchange recognized the symptoms and recommended someone call out an analyzer technician to look at the O2 analyzer. Once repaired, the controls and process returned to normal.
Instilling the sensibility or awareness to “see” the way the veteran sees complex integrated processes and systems is a growing challenge, as all industries struggle to replace retiring experienced operators with relatively new recruits. An incident like the one described above can be reviewed in written form or using PowerPoint in refresher training. It can be dramatized by the team leader as the trainee runs the control board, challenging them to think about possible outcomes, observations and remedial actions. But increasingly, industry is seeking simulation tools to more directly immerse trainees and experienced operators in challenging scenarios.
Other industries, like aviation and marine piloting, have been employing simulation for years, but it’s a little more difficult for a process plant, as every process is more than a little different. For example, while every fluidized catalytic cracker (FCC) is effectively doing the same thing—combining heavy gas oil from a crude distillation unit with catalyst and distilling the end products into higher-value fuels—one would struggle to find two that were similar enough to use the same simulation configuration. Consequently, engineers and trainers are challenged to use simulation packages that don’t require extensive manpower and months of engineering to configure.
Another approach that has some appeal is the “wayback machine”—yet to be widely marketed by Mr. Peabody but adopted by Google for archiving vanished web pages. Instead of using Google to look at the website you created in high school, DCS graphics can be populated with historical data and played back to allow trainees to have a glimpse of what the board might have looked like during an incident or upset. It’s an approach that’s not without merit but limited by the capabilities of the historian (not all can play back history to a graphic), sampling frequency of the historian, and the lack of alarms, trends and operator actions as they may have occurred in real time, way back when. You might need to use some service providers to transform your DCS graphics to, say, OSIsoft PI graphics, so they look like what the DCS operator sees. But it’s still a smaller investment than a high-fidelity simulator. While you can’t easily use it to explore what-ifs, it does have the advantage of showing precisely what the measurements and indications displayed—there is almost zero model uncertainty or drift.
Not unlike the simulator is the concept of a digital twin, not an especially new idea but one that’s gaining traction as another service for Microsoft or your favorite control systems supplier to garner your purchase orders. The thought is, a virtual replica of your process plant, complete with all its critical machinery, vessels, weather, controls and so on, is created on a computer. Perhaps it will be in the cloud or the fog/smoke, whatever term you like for “someone else’s computer.” We are drawn to the cloud solution because first, we don’t have to purchase a huge powerful computer (we will rent space on Microsoft’s or Amazon’s); second, we don’t have to maintain it/patch it/update the OS; and third, someone else will worry about firewalls, security, etc.—all for a fee. But will it work as well as OneDrive or SharePoint? [hilarious laughter here]. Someone clever (Watson?) will imbue the digital twin to mirror the real process, updating measurements in the twin, and somehow including dynamics, too. How has your experience been, scaling up pilot plant data and dynamics to a production-class plant? Perhaps Watson will be most incredibly more clever than us.
The digital twin has power and promise for the well known and highly-tamed processes—think boiling water, spinning turbines—but might find the arena of modeling chemical and refining process dynamics as challenging as we humans do.
