GE's Predix analytics system for the industrial Internet and its platform for constructing a digital twin of an industrial plant asset are often discussed as two separate things.
At its Minds + Machines conference in San Francisco Nov. 16, its VP of software research, Colin Parris, gave a video demonstration of how Predix analytics could be combined with information from a digital twin operating in a virtual environment to come up with recommendations for a rapidly growing wear and tear problem.
A digital twin is a recreation in a virtual environment of a physical asset. GE is creating them for major industrial assets, such as a locomotive, jet engine or power plant turbine. The twin includes the engineering plans, operational history and algorithms that mimic wear under different operating conditions. Each component of the asset has its own twin along with a composite twin that mimics who they work together in the real world. Gartner has named digital twins as one of the top ten technologies of 2017.
The video demonstration was triggered toward the end of the second day keynote oby Harel Kodesh, GE's VP for Predix and CTO. The demonstration of a digital twin and Predix working on the same problem was striking in part because Parris could address a natural language query to the digital twin and after a short processing delay, get a natural language response back from Predix. It was a sort of Serious version of Siri for the Internet of Industrial assets. The subject of the demo was a rapidly wearing out generating plant turbine. A YouTube archive of the presentation is available here.
The interaction is initiated by "Twin 6321," a virtual replica of a turbine, which alerts the operator responsible for it that it's detected a growing problem with a rotor in the machine. The voice activated part of what turned out to be the OpFlex application that was demonstrated at Minds + Machines is still under development. The query and response from the digital twin is possible with a conventional, command-line interface today, GE spokesmen told InformationWeek after the presentation, an account of which follows.
"OK, twin, help me understand the problem," responded Parris, standing in for the plant operator in the video.
Twin 6321 thinks a moment, or perhaps we should say Predix processes the query for a moment, and responds that the number of turbine start/stop cycles has gone up dramatically. "Over the last six months, my number of cold starts is four; my number of warm starts is eight; my number of hot starts is 39," the twin states in somewhat metronomic voice. That represents a 27.5% increase in starts and stops, it concludes.
Parris next asked, "Tell me about your rotor damage."
The twin through Predix responded: "My damage rate has increased by 4.0 times over the last six months. If this were to continue, I would lose 69% of my useful life."
Parris then commanded: "Twin, give me options for mitigating that rotor damage."
The response: "Based on weather forecasts, historical data, fuel costs, electricity pricing and my present condition, there are two options. Option 1) is manually slow down my startup ramp rate to reduce wear on my rotor. Option 2 is download the Opflex App (from GE) and apply controls to minimize stress and reduce fuel consumption too."
Parris asked how does he know he can depend on the rules and assumptions in the Opflex App.
Twin responded: "I used my 15 years of past history. I used fleet learning from 125 other steam turbines, like me. I did 58,900 simulation runs to get this calculation. I am 95% confident of my assumptions."
Parris than asked about the costs of option 2: "Tell me about the numbers."
"The numbers look good," said Twin in response. "We can reduce stress by 25%, which will reduce damage down to the normal range. Start-up time will come down 40% and start fuel costs will come down by 50%." And most of all, the generating plant will avoid an unplanned outage that would have cost it $12 million, the Predix analysis concluded.
At the end of the video of this dialogue between Parris and a digital twin, Parris walked out on stage to say such an exchange would be possible with a D11 steam turbine digital twin for a physical turbine that is currently running in Southern California.
"What you saw was the human mind working with a machine," he told the keynote crowd. About 2,700 attended the conference, held on Pier 48 behind AT&T ballpark where the San Francisco bGiants play.
A production user of digital twins is Exelon, which GE unveiled Nov. 15 as its largest electricity generation partner and user of Predix. In a follow up to an interview with InformationWeek, Brian Hoff, director of corporate innovation, said the firm is using digital twins of steam generators powered by natural gas at Exelon's Colorado Bend Generating Plant in Wharton County, Texas.
The natural gas generating plant has implemented 1,000 tags or instrumented components on 14 physical assets in the plant and "built analytics models using several combinations of those tags," GE spokesmen for Hoff wrote. Exelon operations will use Predix to predict equipment wear and conditions, prevent maintenance problems and lower operational costs and risks. It's built predictive models to show optimum "heat balance" in operating conditions, optimum mechanical operating conditions and degree of lubrication, optimum fuel usage and conditions under which performance loss may be observed.
"As we deploy Predix across the rest of the fleet, we will extend the digital twin across all generation assets," said the email response.