Yu Ding brings data science to wind energy
Dr. Yu Ding isn’t the type of engineer that people normally think of when they think of wind energy, but perhaps that’s why over a decade ago he became the perfect engineer to solve a problem in the renewable energy industry.
The problem: How do you know if something is really working?
Due to both their massive size and the number of variables involved, proving that small changes in a wind turbine are having a measurable effect is a challenging task. Even the most basic element of a wind turbine isn’t constant.
“You obviously can’t control the wind itself; it’s changing all the time,” Ding said. “But there’s also all kinds of other environmental variables -- temperature, turbulence, humidity, etc. How do you possibly, accurately, control those types of things so that you can testify to the effect of an operational change?”
One way would be to control for the variables using some kind of wind tunnel, but wind turbines stand at over 300 feet tall.
“If you want to do a controlled experiment, you’re going to have to use an enormous wind tunnel in front of a utility scale turbine, but nobody has been doing that,” Ding said. “That would be a huge undertaking. So you have to rely on their normal operation.”
A proposed change in the wind energy field was installing vortex generators on turbines to improve the power production efficiency. Vortex generators, simply put, are small devices that change the way air moves around the turbines to increase energy productivity. The industry simply wanted to know, “Do they work?”
Finding an answer is extremely complicated, in part because a field test is costly.
“Adding the generators, material wise, is not very expensive, but you need to put a bunch of them on turbine blades, so the labor is pretty expensive,” Ding said. “Someone has to climb up there with special equipment. They also have to stop the production, and not just for one turbine. You have to shut down other turbines in its neighborhood. So you’re looking at about $7,000 to $10,000 per turbine. If you’re on a big scale, of say 200 turbines, you’re looking at up to $2 million upfront for a whole farm.”
Another problem is, even if they do work, the expected improvement is small.
“Twenty-five years ago when someone studied vortex generators in a lab, they said it could get significant improvement, say 15 percent or 20 percent,” Ding said. “Obviously that would be really easy to detect. In the field, the general consensus is that it’s probably between 1 and 5 percent. And with all of the noise you have in this data, that 1 percent or 5 percent could be overwhelmed by the noise.”
One percent does not sound like much, but if true, it is enough to get the investment back in just a couple of years, and the benefit over the life of a wind farm really adds up. So, with all of the given variables, how can you possibly hope to reliably measure a change as small as 1 percent?
Searching for a weak signal in that much noise requires data--a lot of it--and that is Ding’s specialty. Ding, the Mike and Sugar Barnes Professor in the Department of Industrial and Systems Engineering at Texas A&M University, considers himself a data scientist, and his Ph.D. dissertation was about advanced manufacturing. His early work out of college primarily was about using data for quality assurance. He applied the same principles to these new problems starting in 2004. Over the next several years he worked on some wind energy problems, but it wasn’t until connecting with EDP Renewables in 2011 that his research in this area really took off.
“We created data science models to try and see, other than the wind speed, how you possibly can incorporate other kind of measurables,” Ding said. “The environmental conditions, wind direction, air density, turbulence, wind shear, humidity. Can we put enough data into a model to tell you how conditions are going to affect wind power production? So when you feed in new data, it’s like you create a mathematical surrogate, and you can compare your data’s theoretical power output to the actual power output.”
His work on data science for wind turbine performance and upgrade quantification drew the attention of the German company SmartBlade. Ding published a study about quantifying wind turbine upgrades, and SmartBlade asked him to conduct a joint study to see if his method could verify their internal assessments on their own data. The two published an academia-industry case study in Renewable Energy in 2017, “Quantifying the Effect of Vortex Generator Installation on Wind Power Production,” which presented “a strong case of cross validation, testifying to the respective method's capability and credibility.”
The result of that study was enough to convince EDP Renewables, the fourth-largest wind owner-operator in the world, to rethink how a turbine upgrade and retrofit effect should be quantified. They decided to fine tune Ding’s original approach and jointly developed with Ding’s team a standard procedure that evaluates decisions of installing retrofits on a large scale.
“Based on Ding’s innovative approach that re-affirmed gains with more certainty, EDP Renewables installed blade retrofits on up to 1,000 turbines across our fleet,” Brian Hayes, executive vice president of EDP Renewables North America said. “This was an investment of greater than $7 million and increased the amount of clean energy produced by more than 100,000 MWh -- a positive impact to say the least!”
Hayes went on to say, “The change in thinking he (Ding) is driving with our team is significant given the history and general risk aversion to changing methodologies.”
Ding compared the relationship with Hayes to the relationship between (Oakland Athletics general manager) Billy Beane and Bill James in the book “Moneyball.” Ding said that Hayes encouraged him to read the book when they first met. In the book, James is a data analyst who had some innovative ways of looking at the game of baseball, but it wasn’t until Beane and the money-strapped Oakland Athletics began to buy into them that they were ever applied and accepted.
“It only took one insider, in that case Billy Beane, to believe in the data science, and now data analytics is routine in major league baseball,” Ding said. “That’s a complete change in mentality.”
The impact of Ding’s research is why he was a recipient of a College of Engineering Research Impact Award earlier this year. Ding said that data science’s impact on engineering applications is everywhere. He is now working on developing innovative data science methods for advanced manufacturing, the subject he worked on in his Ph.D. dissertation and always something he wants to make an impact on.