Presentation is loading. Please wait.

Presentation is loading. Please wait.

Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation,

Similar presentations


Presentation on theme: "Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation,"— Presentation transcript:

1 Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL Simulating the Entire Life of an Offshore Wind Turbine Matthew Barone, Josh Paquette, and Brian Resor Wind Energy Technologies Department Sandia National Laboratories Lance Manuel and Hieu Nguyen Department of Civil, Architectural, and Environmental Engineering University of Texas at Austin

2 High Performance Computing and Wind Energy Vestas “Firestorm” Computer 180 Tflop peak performance #3 ‘fastest’ industry supercomputer in the world From: Calaf, Parlange, & Meneveau, Phys. Fluids 23, From: Larsen et al, European TOPFARM project final report. Wind farm optimization LES of wind turbine arrays Example Applications of HPC to Wind Energy Growing Industry Resources

3 50-year recurrence ? ? ? ? Uncertainty in Wind Turbine Extreme Load Extrapolation For DLC 1.1 the characteristic value of load shall be determined by a statistical load extrapolation and correspond to an exceedance probability, for the largest value in any 10-min period, of less than or equal to 3.8 x 10–7, (i.e. a 50-year recurrence period) for normal design situations. From: IEC Ed. 3 – Wind Turbine Design Standards 6 weeks of simulation 2 different fits 128 hours of simulation (many different realizations) Fits to 2 different realizations

4 Research Questions for the Computer What are the probability distributions for various one-hour extreme turbine loads for an offshore wind turbine in shallow water? Compute these down to fifty-year recurrence probabilities. What turbulent wind inflow and wave conditions lead to the largest turbine loads? Save the input parameters for each simulation so that select simulations can be reproduced later. What are the uncertainties for a given load extrapolation procedure?

5 Sandia High-Performance Computing Resources Sandia continues to extend a distinguished record in high performance computing. These resources are available for solving problems in wind power. ASCI Red World’s First Teraflop Computer 1.3 Teraflops* World Rank (1997): #1 Thunderbird Cluster 53 Teraflops World Rank (2006): #6 450 Teraflops World Rank (2010): #10 Red Mesa Partition: Dedicated to energy-related work NREL & Sandia users 180 Teraflops *1 Teraflop = 1 Trillion floating point operations per second

6 Turbine Aero-hydro-elastic Model NREL 5 MW offshore reference turbine 3-bladed HAWT with upwind rotor Monopile foundation, water depth of 20 m Rotor Diameter = 126 meters Hub Height = 90 meters Variable speed, collective variable pitch controller, no active yaw control Cut-in, Cut-out, and Rated Wind Speed = 3 m/s, 25 m/s, 11.4 m/s Aero-hydro-elastic Simulation Code NREL FAST code Equilibrium BEM ‘inflow’, or ‘wake’, model  Chosen to avoid instabilities associated with dynamic wake models NREL Turbsim code used to generate inflow turbulence (Kaimal spectrum) Incident wave field computed using JONSWAP spectrum in FAST

7 Site Definition Forschung in Nord-Ostsee (FINO) research platform 45 km north of the Island of Borkum in the North Sea Measurement Period: November 2003 – May 2005 Wind: 10-minute mean values of the wind speed at 100-m height Waves: 1-hour significant wave height from wave buoy No data for turbulence intensity: we assumed uniform 10% value for all wind speeds

8 Aero-hydro-elastic Load Simulations DAKOTA Simulation framework developed at Sandia National Laboratories Enables large-scale parameter studies, sensitivity analysis, optimization, and UQ dakota.sandia.gov Simulation Procedure DAKOTA samples two random wind seeds, two random wave seeds, and mean wind speed for each sim using a Latin Hypercube sampling method Significant wave height and period are taken as expected values conditional on mean wind speed DAKOTA asynchronously schedules a simulation on each available core TurbSim and FAST are run in sequence for each simulation Random seeds, mean wind speed, and 1-hour extreme values are saved by DAKOTA Stats 552,809 simulations performed (~63 years) in four separate batches 1028 cores used on Red Sky 5 days of total wall-clock time

9 Extreme Blade Tip Deflections

10 Extreme Blade Root Bending Moments

11 Extreme Tower Base Moments

12 Extreme Tower Torsional Moment

13 Extreme Tower Base Fore-Aft Moment vs. Mean Wind Speed Max. load at U = m/s

14 Extreme Tower Torsional Moment vs. Mean Wind Speed Max. load at U = m/s

15 Maximum Tower Base Fore-Aft Moment Case Simulation No. 524,988 Hub Height Wind Speed (m/s) Blade Pitch (deg) Sea Surface Level (m) Tower Fore-Aft Moment (kN-m)

16 Evaluation of Uncertainty in Load Extrapolation: How much simulation is needed? 128 Simulations 512 Simulations 2048 Simulations 1.The 63 years’ of simulation was used to generate subsets of N simulations 2.Each subset was used to estimate the 1- and 50 year loads using linear least-squares regression below a probability level of Mean estimates and confidence intervals were generated for the 1- and 50-year load

17 Evaluating Load Extrapolation Uncertainty – Blade Root Flapwise Moment Fifty-year Return Load

18 Evaluating Load Extrapolation Uncertainty – Tower Base Torsional Moment Fifty-year Return Load

19 Simulation Challenges Large-scale loads simulations can be an “I/O bound” supercomputing application rather than “CPU bound” Many small files are written simultaneously to disk Caused a problem on Red Sky’s parallel file system Memory efficiency of sampling algorithm important for large numbers of simulations Dakota’s Latin hypercube sampling algorithm limited the number of samples in a single simulation batch

20 Future Directions Address dynamic wake robustness issue Treat turbulence intensity, significant wave height, wave spectral peak period, wind shear probabilistically Examine fatigue load spectra Investigate concurrent extreme loads Example: what is the probability distribution of edge-wise blade root moment when flap-wise moment exceeds a given value? Explore potential impact on wind turbine design standards

21 Acknowledgements Thanks to the Sandia Red Sky team: Steve Monk, Sophia Corwell, Karen Haskell, Anthony Agelastos, Jeffrey Ogden, Joel Stevenson Thanks to the DAKOTA team, Brian Adams and Mike Eldred Thanks to Jason Jonkman for assistance in modifying the FAST code

22 Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL Thank You


Download ppt "Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation,"

Similar presentations


Ads by Google