Written by Barbara Rook
Wind developers can increase wind-farm operational productivity by quickly noting and addressing underperforming turbines. AI-based software with predictive models can help prevent component failures before they occur, saving unnecessary asset downtime and lost revenue. (Image: Greenbyte)
Increasingly, predictive analytic software takes much of the work and worry out of deciphering the massive amounts of data generated by wind turbines. The goal is to link real-time data to proactive decision-making without drowning in “an unmanageable ocean of data,” according to independent market research firm, New Energy Update. Ultimately, wind operators are trying to reduce their reliance on OEMs for operations analytics and take control of their data.
“Predictive analytics helps owners establish a single source of independent truth across the organization,” explains Ryan Blitstein, VP of Renewable Energy at Uptake, an artificial intelligence (AI) and internet of things (IoT) software provider.
Data delivered directly into the wind flow can predict a problem, sending alerts that allow proactive investigation and repairs. This “fix it before it breaks” strategy is preventing more extensive repairs down the road, thereby saving money and downtime.
Uptake’s asset performance management system, Uptake APM, delivers actionable insights directly into the wind flow, says Blitstein. “So, if the data science engines predict that something is going to happen, it gets to the right person at the right time before something goes wrong.”
Uptake recently released an updated version of Uptake APM. The software is based on the Asset Strategy Library, a comprehensive database of content on how industrial equipment fails. This data, with its extensive industrial machine data experience, helps wind companies predict more failures with greater accuracy, according to Uptake.
Last year, the AI company worked with Iowa-based MidAmerican Energy Company on a blade inspection and repair project.
“We heard over and over again that software companies just don’t ‘get’ wind. So, we put a lot of time configuring our asset analytic products for the particular way wind sites work,” says Blitstein. Within two days of deploying the software at a pilot site, the Uptake data science model predicted a main bearing failure, preventing gearbox failure and saving MidAmerican $250,000.
“We’re allowing companies to be more proactive than reactive,” says Blitstein, leading to a more self-directed approach to managing operations and maintenance.
Likewise, Predict, a new addition to Greenbyte‘s Energy Cloud, a renewable data management system, detects small deviations so operators can avoid secondary failures, which could lead to more downtime and costly repairs.
Greenbyte’s Predict software mimics how the brain works. An artificial neural networks mathematical model learns the behavior of the system, stores it as experience information and then uses it to perform the task it is assigned. (Image: Greenbyte)
The Swedish renewable energy company is piloting Predict with several customers and plans to launch availability early in 2019. In its year-long pilot, Greenbyte found that its system can predict faults as much as two to nine months in advance.
Predict sees alarms as they happen instead of waiting for alerts from the vendor, according to Dr. Pramod Bangalore, Greenbyte’s Head of Research. In one example, Predict was able to alert an operator to a fault in the rotating union, which will be replaced. The system also saw the same pattern within the same wind farm and alerted the customer to a second, identical alarm.
While Predict uses data from SCADA (supervisory control and data acquisition) systems, it sees faults that SCADA data might miss, says Bangalore. In one case study, a maintenance error caused temperatures to rise during peak performance times. Predict alerted the customer to the problem, which was then corrected.
“The customer told us that these kinds of faults are difficult to detect from SCADA statuses,” he said. SCADA typically takes a long time to recognize a problem during high-power events. In addition, Predict can see faults in electric and hydraulic components, an advantage over vibration-based condition-monitoring systems, he notes.
Predict’s monitor portfolio includes a rating that identifies how certain the system is about a problem in a component. High certainty means follow-up is needed and low certainty means an alarm may have been triggered by changing conditions, and that no real threat exists.
These types of predictive analytics save wind-farm operators the headache of managing data themselves.
“Operators don’t have time to analyze data themselves,” says Bangalore. Predict was a response to a customer’s request to more effectively manage critical operational data.
Predictive analytics solve current problems operators face dealing with multiple, often incompatible data sets, including SCADA data and vibration data. In addition, wind companies often manage turbines from multiple manufacturers, which may operate differently across a wind farm. Having a single view of all turbines eliminates the complexity of integrating point solutions.
“It’s about having the power and the knowledge to take greater ownership to drive changes on the ground,” adds Blitstein.
Tell Us What You Think!