Nalu-Wind simulations of the neutral inflow Scaled Wind Farm Technology (SWiFT) benchmark were used to analyze which quantities of interest within the wind turbine wake and surrounding control volume are important in performing a momentum deficit analysis of the wind turbine thrust force. The necessary quantities of interest to conduct a full Reynolds-Averaged Navier-Stokes (RANS) formulation analysis were extracted along the control volume surfaces within the Nalu simulation domain over a 10 minute period. The thrust force calculated within the wake from two to eight diameters downstream using the control volume surfaces and the full RANS approach matched the thrust force that the wind turbine applied to the flowfield. A simplified one-dimension momentum analysis was included to determine if the inflow and wake velocities typically acquired during field campaigns would be sufficient to perform a momentum deficit analysis within a wind turbine wake. The one-dimensional analysis resulted in a 70% difference relative to the coefficient of thrust (Ct ) determined by the full RANS method at 2D downstream and a 40% difference from 5D to 8D, where D is the diameter of the turbine. This suggests that the quantities typically captured during field campaigns are insufficient to perform an accurate momentum deficit analysis unless streamwise pressure distribution is acquired, which reduced the relative difference to less than 10% for this particular atmospheric inflow.
An uncertainty quantification technique for nacelle-mounted lidar is developed that extends conventional error analyses to precisely account for residual uncertainty due to observed non-ideal features in processed Doppler lidar spectra. The technique is applied after quality assurance/quality control (QAQC) processing to quantify residual error, both bias and random, from solid-body interference, shot noise, and any additional uncertainty introduced to the data from the QAQC process itself. The approach follows from the one-time construction of a high-dimensional parametric database of synthetic lidar spectra and subsequent processing with an existing QAQC technique. A model of the correspondence between the spectral shape and the associated residual errors due to non-ideal features is then developed for quantities of interest (QOIs) including the geometric median and spectral standard deviation of line-of-sight velocity. The model is preliminarily implemented within a neural network framework that is then applied in post-processing to sample returns from a DTU SpinnerLidar. The initial analysis uncovers the effects of specific sources of uncertainty in the context of both individual spectra and full-field maps of the measurement domain. The technique is described in terms of application to continuous wave (CW) lidar, though it is also relevant to pulsed lidar.
Work has begun towards model validation of wake dynamics for the large-eddy simulation (LES) code Nalu-Wind in the context of research-scale wind turbines in a neutral atmospheric boundary layer (ABL). Interest is particularly directed at the structures and spectra which are influential for wake recovery and downstream turbine loading. This initial work is to determine the feasibility of using nacelle-mounted, continuous-wave lidars to measure and validate wake physics via comparisons of full actuator line simulation results with those obtained from a virtual lidar embedded within the computational domain. Analyses are conducted on the dominant large-scale flow structures via proper orthogonal decomposition (POD) and on the various scales of wake-added turbulence through spectral comparisons. The virtual lidar adequately reproduces spatial structures and energies compared to the full simulation results. Correction of the higher-frequency turbulence spectra for volume-averaging attenuation was most successful at locations where mean gradients were not severe. The results of this work will aid the design of experiments for validation of high-fidelity wake models.
An uncertainty quantification technique for nacelle-mounted lidar is developed that extends conventional error analyses to precisely account for residual uncertainty due to observed non-ideal features in processed Doppler lidar spectra. The technique is applied after quality assurance/quality control (QAQC) processing to quantify residual error, both bias and random, from solid-body interference, shot noise, and any additional uncertainty introduced to the data from the QAQC process itself. The approach follows from the one-time construction of a high-dimensional parametric database of synthetic lidar spectra and subsequent processing with an existing QAQC technique. A model of the correspondence between the spectral shape and the associated residual errors due to non-ideal features is then developed for quantities of interest (QOIs) including the geometric median and spectral standard deviation of line-of-sight velocity. The model is preliminarily implemented within a neural network framework that is then applied in post-processing to sample returns from a DTU SpinnerLidar. The initial analysis uncovers the effects of specific sources of uncertainty in the context of both individual spectra and full-field maps of the measurement domain. The technique is described in terms of application to continuous wave (CW) lidar, though it is also relevant to pulsed lidar.
Doubrawa Moreira, Paula; Quon, Eliot; Martinez; Tossas, Luis (Tony) M.; Shaler, Kelsey; Debnath, Mithu; Hamilton, Nicholas; Herges, Thomas H.; Maniaci, David C.; Kelley, Christopher L.; Laros, James H.; Blaylock, Myra L.; Van Der Laan, Paul; Andersen, Soren J.; Krueger, Sonja; Cathelain, Marie; Schlez, Wolfgang; Jonkman, Jason; Branlard, Emmanuel; Steinfeld, Gerald; Schmidt, Sascha; Blondel, Frederic; Lukassen, Laura J.; Moriarty, Patrick
Previous research has revealed the need for a validation study that considers several wake quantities and code types so that decisions on the trade-off between accuracy and computational cost can be well informed and appropriate to the intended application. In addition to guiding code choice and setup, rigorous model validation exercises are needed to identify weaknesses and strengths of specific models and guide future improvements. Here, we consider 13 approaches to simulating wakes observed with a nacelle-mounted lidar at the Scaled Wind Technology Facility (SWiFT) under varying atmospheric conditions. We find that some of the main challenges in wind turbine wake modeling are related to simulating the inflow. In the neutral benchmark, model performance tracked as expected with model fidelity, with large-eddy simulations performing the best. In the more challenging stable case, steady-state Reynolds-averaged Navier–Stokes simulations were found to outperform other model alternatives because they provide the ability to more easily prescribe noncanonical inflows and their low cost allows for simulations to be repeated as needed. Dynamic measurements were only available for the unstable benchmark at a single downstream distance. These dynamic analyses revealed that differences in the performance of time-stepping models come largely from differences in wake meandering. This highlights the need for more validation exercises that take into account wake dynamics and are able to identify where these differences come from: mesh setup, inflow, turbulence models, or wake-meandering parameterizations. In addition to model validation findings, we summarize lessons learned and provide recommendations for future benchmark exercises.
Moriarty, Patrick; Hamilton, Nicholas; Debnath, Mithu; Herges, Thomas H.; Isom, Brad; Lundquist, Julie K.; Maniaci, David C.; Naughton, Brian T.; Pauly, Rebecca; Roadman, Jason; Shaw, Will; Van Dam, Jeroen; Wharton, Sonia
The American WAKE experimeNt (AWAKEN) is an international multi-institutional wind energy field campaign to better understand wake losses within operational wind farms. Wake interactions are among the least understood physical interactions in wind plants today, leading to unexpected power and profit losses. For example, Ørsted, the world’s largest offshore wind farm developer, recently announced a downward revision in energy estimates across their energy generation portfolio, primarily caused by underprediction of energy losses from wind farm blockage and wakes. In their announcement, they noted that the standard industry models used for their original energy estimates were inaccurate, and this was likely an industry-wide issue. To help further improve and validate wind plant models across scales from individual turbines as well as interfarm interactions between plants, new observations, such as those planned for AWAKEN, are critical. These model improvements will enable both improved layout and more optimal operation of wind farms with greater power production and improved reliability, ultimately leading to lower wind energy costs.
Uncertainty is present in all wind energy problems of interest, but quantifying its impact for wind energy research, design and analysis applications often requires the collection of large ensembles of numerical simulations. These predictions require a range of model fidelity as predictive models, that include the interaction of atmospheric and wind turbine wake physics, can require weeks or months to solve on institutional high-performance computing systems. The need for these extremely expensive numerical simulations extends the computational resource requirements usually associated with uncertainty quantification analysis. To alleviate the computational burden, we propose here to adopt several Multilevel-Multifidelity sampling strategies that we compare for a realistic test case. A demonstration study was completed using simulations of a V27 turbine at Sandia National Laboratories’ SWiFT facility in a neutral atmospheric boundary layer. The flow was simulated with three models of disparate fidelity. OpenFAST with TurbSim was used stand-alone as the most computationally-efficient, lower-fidelity model. The computational fluid dynamics code Nalu-Wind was used for large eddy simulations with both medium-fidelity actuator disk and high-fidelity actuator line models, with various mesh resolutions. In an uncertainty quantification study, we considered five different turbine properties as random parameters: yaw offset, generator torque constant, collective blade pitch, gearbox efficiency and blade mass. For all quantities of interest, the Multilevel-Multifidelity estimators demonstrated greater efficiency compared to standard and multilevel Monte Carlo estimators.
Power production of the turbines at the Department of Energy/Sandia National Laboratories Scaled Wind Farm Technology (SWiFT) facility located at the Texas Tech University’s National Wind Institute Research Center was measured experimentally and simulated for neutral atmospheric boundary layer operating conditions. Two V27 wind turbines were aligned in series with the dominant wind direction, and the upwind turbine was yawed to investigate the impact of wake steering on the downwind turbine. Two conditions were investigated, including that of the leading turbine operating alone and both turbines operating in series. The field measurements include meteorological evaluation tower (MET) data and light detection and ranging (lidar) data. Computations were performed by coupling large eddy simulations (LES) in the three-dimensional, transient code Nalu-Wind with engineering actuator line models of the turbines from OpenFAST. The simulations consist of a coarse precursor without the turbines to set up an atmospheric boundary layer inflow followed by a simulation with refinement near the turbines. Good agreement between simulations and field data are shown. These results demonstrate that Nalu-Wind holds the promise for the prediction of wind plant power and loads for a range of yaw conditions.