Our research highlights serve as a collection of feature articles detailing recent scientific achievements on GCS HPC resources.
In recent years, Germany has gained the ability to generate almost a quarter of its electricity through wind power—a 15 percent improvement in the last decade. As the country continues to push toward 100 percent of its energy through renewable sources, wind turbines will play an increasingly important role.
Despite these gains, wind energy is still geographically limited, with the vast majority of wind turbines located in the flat northern half of the country. Germany’s southern states, Baden-Württemberg and Bavaria, have hilly or mountainous terrain that make wind turbine placement more complicated.
From modeling weather patterns and the impact of forest on turbine efficiency, to understanding turbulence’s effects on wind turbines, to charting wear and tear on wind turbines, there are myriad issues to measure and take into account to ensure the best return on investment for new wind farms.
Research institutions in southern Germany have been working on improving wind turbine efficiency in their back yards for 70 years. Recently, the Baden-Württemberg Ministry of Science, Research, and the Arts (MWK) sponsored the WindforS project, which brings together multi-disciplinary experts to improve wind turbine efficiency. One of the WindforS subprojects, WINSENT, which began in 2017, supports researchers using high-performance computing (HPC) in conjunction with experiment to help design a first-of-its-kind experimental wind energy test station in the Swabian Alb.
WINSENT is a collaboration between the University of Stuttgart, Karlsruhe Institute of Technology (KIT), University of Tübingen, Center for Solar Energy and Hydrogen Research Baden-Württemberg, Technical University of Munich, and Esslingen University of Applied Science (Hochschule Esslingen) and is funded by MWK and the German Federal Ministry for Economic Affairs and Energy (BMWi).
Researchers from Uni Stuttgart and KIT have been using Gauss Centre for Supercomputing (GCS) HPC resources at the High-Performance Computing Center Stuttgart (HLRS) and Leibniz Supercomputing Center (LRZ) in Garching near Munich to support an innovative, three-part modelling approach that integrates local weather and climate data with high-resolution simulations of turbulence around wind turbine rotor blades.
Three-phase simulations set the stage for more efficient wind energy
Finding the most efficient place for wind turbines in complex, hilly terrain requires much more than just being able to predict wind speed. Often, such terrain is also heavily wooded and has steep inclines and declines. The weather can also be more punishing for wind turbines, as wind gusts and heavy storms can be more severe at higher elevations.
Even modern-day supercomputers are unable to do meaningful simulations that take into account the complex collection of environmental and engineering considerations for siting wind turbines in these environments. As a result, the WINSENT collaborators have developed a three-part simulation approach that focuses on different aspects of the simulation at different resolutions. Each step in the process generates datasets that ultimately inform other steps in the process, making the overall simulation suite more accurate, and thus, more predictive.
Researchers need to ensure that wind turbine investments not only are located where wind will keep them operating at the highest efficiency possible, but also ensure that increased maintenance costs and smaller maintenance intervals do not raise costs too high. Further, by better understanding the turbulent dynamics around turbine rotor blades and the influence of the wake—the complex, altered airflow downstream of a turbine that can cause ripple effect on other turbines close by—researchers are able to better understand how to place groups of turbines to prevent some turbines from negatively affecting the efficiency of others.
Despite focusing on different calculations at different resolutions, the WINSENT collaborators all create a computational mesh for their simulations—that is, they divide their simulations into a grid and solve equations that govern the behavior of wind, turbulence, or other factors within each box of the grid over time. The smaller the mesh or the finer the focus on intricate details, the more computationally expensive the simulations become.
In the first step of the three-phase WINSENT simulation process, KIT researcher Dr. Daniel Leukauf uses SuperMUC-NG at LRZ to run simulations primarily focused on the meteorological aspects of turbine siting, creating meshes at a resolution of roughly 150 metres. These simulations serve as the basis for boundary conditions for increasingly higher-resolution models, and assist the researchers in identifying promising sites for a wind farm.
Leukauf uses the Weather Research and Forecasting (WRF) model, developed by the National Center for Atmospheric Research in the United States and widely used for high-resolution weather forecasting and climate modelling simulations. WRF offers the possibility to model weather conditions at local scales, but its resolution is insufficient for accurately modeling the influence of weather conditions on wind turbines’ rotor blades in complex terrain. Despite not being able to include these aspects in his simulation, Leukauf is able to simulate multiple days of detailed meteorological data that can then be passed on to collaborators at Hochschule Esslingen.
The Hochschule Esslingen team uses Leukauf’s data to inform the initial conditions for a higher-resolution simulation. The team uses OpenFOAM, a widely used computational fluid dynamics code, to simulate a wind turbine site at roughly 20-metre resolution, further refining complex terrain simulations by resolving air flow and turbulence happening on a level more relevant for the turbines themselves. The team then passes its data to a research group at Uni Stuttgart to inform its ultra-high-resolution simulations.
Patrick Letzgus of the Uni Stuttgart group runs extremely accurate detached eddy simulations (DES). The team’s DES simulations can model the subtle but essential turbulence that happens around wind turbine rotors, including the wake, while also taking into account the foliage and local topography near a wind turbine site at much greater detail. At a resolution of one metre, the DES simulations are extremely accurate, but also extremely computationally expensive. One simulation requires roughly 7,000 compute cores on HLRS’s HPC systems and need to run for weeks in order to calculate only a few minutes of simulation time. Early runs were done on the Cray XC40 Hazel Hen machine and more recently on the early access platform of the Hewlett Packard Enterprise Apollo system Hawk as well as SuperMUC-NG at LRZ.
Despite a relatively small amount of real time being modeled in these simulations, these calculations can be extrapolated to understand the turbulent dynamics for a wide variety of weather conditions at wind farm.
Standing up a test site and developing tools for industry
The WINSENT project collaborators have finally gone through the lengthy permitting process for building their experimental facility on the Swabian Alb near Göppingen, Germany, and anticipate that it will be constructed in the coming months. “A wind energy test field of this kind doesn’t exist in complex terrain,” Letzgus said. “It is great that it is coming so that many groups and disciplines can use it for research.”
The WINSENT collaboration has streamlined its simulation chain, and is looking forward to being able to start comparing simulation data directly to experimental data gathered at the test site. Leukauf indicated that doing so will ultimately make it possible to develop software that will help industry engineers without access to HPC to determine where wind turbines will operate most efficiently in complex terrain.
Despite the end goal being software that can run on a laptop, Letzgus indicated that without access to GCS HPC resources, the team would be unable to have created such an accurate, detailed modelling chain. “Without the access I have to HPC resources, these highly resolved simulations, and my work in general, wouldn’t be possible,” he said.