University of Stuttgart Researchers Use HPC to Improve Wind Turbine Design

Keyvisual image main
Using a combination of AI and HPC, IAG researchers are improving wind turbine designs to save roughly €195,000 per year for a 10-megawatt turbine. Image credit: J. Renewable & Sustainable Energy (2019)

Using a machine learning algorithm and supercomputing, scientists are modelling wind turbine designs to improve energy efficiency.

Over the past several decades, green energy technologies have played an increasing role in nations’ energy production. With the growing emphasis on sustainability and the need to fight climate change, green energy coming from solar panels, wind turbines, and geothermal sources will only become more important.

To increase clean energy production, researchers and companies in the green energy sector would like to be able to build larger, more efficient turbines that generate more power.

Until recently, engineers have designed relatively modest wind turbines. Typical turbines are anywhere from 50–150 metres tall, have roughly 120-metre blade diameters, and generate roughly 3 megawatts (MW) of power, or about enough power for 2,000 homes. New designs are getting bigger, though. Engineers are designing wind turbines that have 200-metre blade diameters and are capable of generating 10–20 MW. At such large scales, designers have to make sure these large investments are generating energy as efficiently as possible, including mitigating inefficiency introduced by environmental factors.

To that end, a group of researchers at the University of Stuttgart has been using high-performance computing (HPC) resources at the High-Performance Computing Center Stuttgart (HLRS) to help design more energy efficient wind turbines. “When we are talking about more than 10 megawatts of power, even a one-percent increase in efficiency means a lot of additional energy and a lot of money saved,” said Dr. Galih Bangga, post-doctoral researcher at the University of Stuttgart’s Institute for Aerodynamics and Gas Dynamics (IAG). The team’s recent research was published in the Journal of Renewable and Sustainable Energy.

Survival of the sleekest

As wind turbines get larger in order to generate more electricity, so too must their constituent parts. Specifically, wind turbine blades need to have thicker bases, or airfoils, that attach to the main body and ultimately ensure the turbine’s structural integrity. While these thicker airfoils result in a safe, stable wind turbine, they also reduce efficiency due to reduced aerodynamic performance.

The IAG team wanted to figure out how to make blades more aerodynamic without compromising a turbine’s structural integrity. Unfortunately, building prototypes of many different blade designs and then running experimental tests on all models would be prohibitively expensive and time consuming.

Computer simulation offers a much more efficient and cost-effective way to optimize blade designs. In this case, the researchers virtually created many different variations of airfoils, then ran them through a genetic algorithm—an algorithm based on the same genetic laws that, for example, an agricultural researcher might use to maximize crop yields and resilience by breeding plants with the best traits.

Much like Gregor Mendel detailed how cross pollinating pea plants with the best traits would lead to better pea plants, the team’s genetic algorithm takes dozens of turbine blade designs and runs rough turbulence simulations to compare how the models perform, continuing the process until the “most optimized” candidate emerges.

Once the algorithm helps the team identify the best candidate, they turn to HPC to run a higher resolution computational fluid dynamics (CFD) simulation to verify that the algorithm was correct. In the team’s paper, they were able to identify and improve the aerodynamic performance on very thick blade-root airfoil by anywhere from 2.5–7 percent. In practical terms, Bangga noted that the optimization would save roughly €195,000 per year in energy savings on each 10-megawatt wind turbine using this design.

While the team can run its genetic algorithm on personal computers, the high-resolution CFD simulations needed to verify the model would be impossible without HPC. “HPC is absolutely needed to verify the huge number of databases through the use of high-fidelity simulations,” Bangga said. “The access to HLRS resources is a huge benefit to us and our work.”

Engineering the future

In the coming months, Bangga will be presenting the team’s airfoil findings, as well as other recent simulation work they have done related to controlling airflow at the blade level, at several conferences and workshops. In addition to discussing the airfoil, the team is also discussing the role that active flow controls (AFC) can have in improving wind turbine efficiency.

While making subtle design changes to wind turbine blades’ airfoils bring modest, but noticeable changes in energy efficiency, AFC is a more involved and expensive process that can lead to even greater energy efficiency. Active flow controls are similar to the flaps influencing the airflow on an airplane wing—they are controllable parts that influence how air flows around a structure or machine at a local level.

“We want our work to help engineering communities at a variety of scales, not just those that can afford large research and development budgets,” Bangga said. “We want to be able to find ways that improve energy efficiency that smaller companies or local governments can afford, but we also want to find ways to maximize energy efficiency as much as possible.

In the next phase of its work, the team wants to expand the level of detail while running the genetic algorithm, allowing them to achieve a higher level of confidence in their optimization recommendations.

-Eric Gedenk


This article originally appeared on the Gauss Centre for Supercomputing website