Lead Session Chair:
Stephan Barth, Managing Director, ForWind - Center for Wind Energy Research, Germany
James Baeder (1) F P Seung Yang (1)
(1) University of Maryland, College Park, United States
Printer friendly version: Print
Download poster(1.39 MB)
Presenter's biographyBiographies are supplied directly by presenters at OFFSHORE 2015 and are published here unedited
Dr. James D. Baeder is a faculty member in Aerospace Engineering at the University of Maryland. He received his Ph.D. from Stanford University. His research interests are in developing and applying Computational Fluid Dynamic methods to better understand and predict rotary and flapping wing aerodynamics, acoustics and dynamics. Currently he is pioneering the development of improved CFD algorithms, with a focus on GPGPU technology, to: capture the details of laminar/turbulent transition; dynamic stall; as well as tip vortex formation, convection and interaction with other surfaces including fuselages, towers or the ground. He has authored more than 40 archival journal articles.
Wavy trailing-edge flatback aerodynamics using a GPU-accelerated Navier-Stokes solver
Many ongoing research efforts are focused on maximizing the efficiency of wind power generation through innovative and experimental turbine concepts.. Robust and reliable design tools become particularly important in predicting the aerodynamic and structural loads on modern large composite rotors. With the recent advances in computing hardware such as the multi-core graphics processing units (GPUs) as well as hybrid RANS/LES Navier-Stokes solvers coupled with free-vortex wake (FVW) solvers, it is now feasible to calculate the detailed 3-D aerodynamics around large-scale wind turbines in hours on a single high-end GPU card, rather than days on a cluster.
A body-fitted 3-D Navier-Stokes solver with implicit time marching has been developed to take advantage of GPU computing capability and runs about 50-60x as fast as a single CPU core on a single high-end GPU card. With 6GB of video RAM it is able to solve for the flow on mesh sizes of 4-5 million points. This is sufficient for detailed resolution around a single large modern wind turbine blade. The GPU-based Navier-Stokes solver has been extended to interact with an FVM analysis to account for the bound vortices from the other blades as well the trailed vorticity.
Main body of abstract
Of particular importance is the inclusion of advanced turbulence models (Spalart-Allmaras with Delayed Detached Eddy Simulation (DDES) capabilities) as well as laminar/turbulent transition models [Medida et al, AIAA Aviation 2013] in the GPU-based solver. It been validated against wind turbine airfoil data in 2-D [Rinehart et al, AIAA Scitech 2014].
A CPU-based version of the solver uses an overset mesh system to enable wake capturing, and been validated for both the NREL Phase VI rotor (Baeder et al, Offshore EWEA 2014] as well as the Sandia 100m blade design (with and without leading-edge tubercles [Rinehart et al, AIAA Aviation 2014]. However, this needs to run on a cluster and solutions take several days to converge. Thus, it is not feasible to examine new innovative concepts in a timely manner.
The GPU-based hybrid RANS/LES Navier-Stokes solver coupled with FVW solver was developed to look at helicopter brownout [Thomas, PhD Thesis, 2013]. For that simulation the FVW solver represented the rotor system of the helicopter and the RANS/LES solver represented the stationary ground-plane. Calculations that previously took approximately one day per rotor revolution on 48 CPU cores of a cluster now only require a few hours on a single high-end GPU card.
The grid motion terms have been added to the GPU-based Navier-Stokes solver to represent the rotation of the detailed 3-D body fitted mesh around the wind turbine rotor blade. Validations will be shown against wake captured solutions for the 100m Sandia blade and alternative concepts (tubercles) will be explored.
A GPU-Accelerated Free-Vortex Wake/ Navier-Stokes solver has been modified to include grid motion, to enable the high-fidelity simulation of the aerodynamic flow in the vicinity of a large wind turbine rotor blade on a single high-end GPU card with accuracy similar to that obtained with wake capturing. It thus enables more routine examination of novel and innovative rotor blade concepts such as tubercles for alleviating gust response or slotted tips to diffuse tip vortices. These new GPU-based solvers need to be coupled to structural solvers for aeroelastic simulations and adapted for use on multi-GPU cards to enable wind farm simulations.
It is hoped that attendees will see the value in the continued development of advanced computational fluid dynamic (CFD) solvers for application to wind turbines. GPU-based computing enables a single researcher to perform simulations on a desktop with relatively high-fidelity. These robust and reliable analysis tools are part of the process to enable the exploration of novel and innovative concepts for reducing blade fatigue and improving wind farm efficiency.