The device management dashboard at Elmer’s plant lay neglected, and no one from operations questioned his team when a faulty, valve-position, feedback device slowed a plant restart. The consequences were befuddling. Why did they find out about the device only after restarting following a week-long outage? Probably because no one expected such defects should be detected thanks to digitization.
Elmer read some consultant commentary, lamenting that digitization was being adopted at a slow pace. It was necessary for applying computerized, data-analytics tools, machine learning and OpenAI. As observed in a Forbes magazine article published pre-pandemic, much of this slower-than-hoped-for pace was attributed to more “mature” leadership, who had their own concepts about how to use data to guide decisions and safeguard against malfunctions. That is, a generation steeped in old-school practices needed to retire, in the hope that a fresh generation of leadership acclimated to data analysis tools could pick up the pace.
Similar to many plants, where motivated technicians/technologists were cut to the bare minimum, Elmer only acted on the priorities and at the direction of the operations team, and in keeping with the old aphorism, “the squeaky wheel gets the grease.” Past endeavors to address a need that had been solved, such as the exchanger fouling too quickly until the cooling water quality was fixed, were cluttering his technician’s “device management” alert dashboard. The WirelessHART device’s batteries were low (or dead), their communication reliability was low, or they had a plethora of configuration-changed alerts that no one could understand or explain. After a while, no one bothered to look at it.
There was another category of smart diagnostics embedded in his control system, that were likewise largely neglected until operations complained about an issue. A journal of system events was constantly recording every diagnostic the system designers felt was worthy of attention. Few looked at them routinely or understood the meaning of two-thirds of them. There were diagnostics for device networks like Modbus or fieldbus that could be viewed, but freeing this information from their respective data silos was never explored. On another platform, the shutdown system’s PLC monitored every solenoid or physical relay coil powered by its output cards, so one could imagine where a faulty coil might reveal itself before a spurious failure—if someone was looking. Such information was only examined for forensics after the “patient” had died.
Machine learning has evolved some capabilities that could be useful for Elmer. A simplistic approach might consist of the correlation between a valve’s position and its related process variable (PV). When it becomes unusual or flatlined, send an alert to the end user. But like his other device management tools, the propensity to generate spam (i.e. meaningless or redundant messages) would threaten to overwhelm potential insights with noise. Noise or cycling is a phenomenon in the wheelhouse of Aperio Systems, for example, among a repertoire of data validation algorithms. Even in the hands of transient managers, such tools might empower them to challenge Elmer.
Get your subscription to Control's tri-weekly newsletter.
The real potential of AI might only be realized when diverse, siloed diagnostics are available to it. Perhaps data diodes—conceptually, a gateway that only allows data to flow one way from source to receiver—will address the concerns surrounding security. Creating pathways between industrial control assets, shutdown systems, field networks and business applications requires increased focus on systems hardening, patching and monitoring. One wonders whether Elmer’s team and their ilk have the time or expertise to learn and fulfill these roles.
Somewhat ironically, the dearth of embedded and motivated humans in plants’ operate-and-maintain roles might be the biggest drag on AI adoption—from management that’s focused elsewhere to techs who only act when operators complain. If Elmer’s reports caught the valve defect before it impacted the startup, there would have been no fanfare or parade. Only when such successes are recognized and celebrated are we likely to attract investment to increase their effectiveness.
