One of the most notable presentations from the 2016 Ovation Users Group Conference (Pittsburgh, July 24-28) showed that the industry truly is on the precipice of a new era with control systems, one in which analytics and machine intelligence are part and parcel of the control system itself.
Today, most powerplants pull data from the distributed control system (DCS) into a separate data historian, such as PI. Software applications, including advanced pattern recognition (APR), then take data from the historian, crunch the numbers, and provide indicators around machine health by comparing real-time, actual data trends to historical expected patterns (figure). When a deviation is detected and becomes significant, an operator and/or a specialist is alerted.
One Ovation™ user, a coal-fired plant owned and operated by one of the largest utilities in the country, is pilot-testing a different approach—having the analytics engine integrated into the automation system itself. While this has been a topic of discussion at previous meetings of the Ovation Users Group, the important ingredient this time is a solid rationale from a user for doing things differently.
As Emerson Process Management’s Azime Can-Cimino and Christine Anselmo noted, “speed is the luxury of Ovation,” referring to the fact that traditional APR techniques are based on data rates of one to five minutes, whereas in Ovation, data rates can be one second and less.
The utility is applying APR at faster data rates with coal mills and induced-draft fans and their associated motor drives. Mills and fans at this site experience an abnormal frequency of high motor-current amp readings, and, therefore, mill and fan trips. According to the utility’s representative at the meeting, these mill and fan motors are undersized, a hangover issue from original design, and so they are always running too near their trip points. In the case of the fans, this means too near their stall points.
“We need a detection and response capability beyond the commercial APR techniques,” he said.
In other words, if the motors were properly sized, there would a more comfortable period of time between a normal motor current reading and a gradual trend towards an abnormal reading. At this plant, however, the time interval between “normal” motor amps and a potential trip is greatly truncated. This is why the speed of detection afforded by much higher data rates has real value.
To date the owner/operator has been suitably impressed: “The Ovation analytics beta test actually predicted a fan failure one month before it occurred!” he reported, privately.
The prediction method or model used to crunch the data is beyond the scope of this article. However most are based on statistical analysis, such as linear and non-linear regression, clustering techniques, decision trees, Bayesian belief networks, and several others.
The benefits of integrating intelligence with control are far-reaching. Avoiding a separate software solution for APR and the digital apparatus that goes with it is an obvious one. Powerplants have been struggling to avoid “islands of automation” for years. In time, embedding many such analytics engines obviates the need for a separate historian and data network.
Beyond the obvious, traditional APR techniques have shown their best success for long-term trending at steady-state conditions, or detecting deviations from a well-established, reasonably constant baseline. With higher data rates, it could be possible to obtain value from APR in short-interval situations—such as plant startup, shutdown, and process upset periods.
For another, with faster detection of, and response to, deviations in key machine parameters, less margin may be needed when a plant is designed in the first place. In the coal-plant example, most large units have six pulverizers whereas full unit load can typically be achieved with four in operation.
Imagine if tighter integration between control systems and monitoring and diagnostic capabilities allowed you to avoid one pulverizer unit. That represents significant savings in capital cost for only a nominal increase in the cost of adding that capability to the automation system.
Most APR capability today is designed to alert operators and specialists of possible developing issues much faster than, say, the alarms built into the control system. Emerson called this a “shirt tug” type of warning. But taking the alert to its logical conclusion, the control strategy can be made “closed-loop,” meaning that the automation system itself makes the decision to trip the equipment, then notifies the operator that such action has been taken. It may be a while before closed-loop control schemes become standard, but when you think about pilots and commercial aircraft, that’s the direction powerplant automation is headed.
One of the challenges with data-driven analytics is that deviations are detected but usually the root cause is not identified. However, with experience, data patterns can also be correlated to root causes through look-up tables, troubleshooting references, or even communicating with a subject matter expert in a pop-up window.
