Effects of Daytime Atmospheric Boundary Layer Turbulence on the Generation of Nonsteady Wind turbine Loadings and Predictive Accuracy of Lower Order Models

Open Access
Lavely, Adam Wesley
Graduate Program:
Aerospace Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
November 07, 2016
Committee Members:
  • James Brasseur, Dissertation Advisor
  • James Brasseur, Committee Chair
  • Sven Schmitz, Committee Member
  • Philip Morris, Committee Member
  • Eric Paterson, Outside Member
  • Dan Haworth, Committee Member
  • Wind Turbine
  • Atmospheric Boundary Layer
  • Engineering-Design Models
Modern utility-scale wind turbines operate in the the lower atmospheric boundary layer (ABL), which is characterized by large gradients in mean velocity and temperature and the existence of strong coherent turbulence eddies that reflect the interaction between strong mean shear and vertical buoyancy driven by solar heating. The spatio-temporal velocity variations drive nonsteady loadings on wind turbines that contribute to premature wind turbine component fatigue failure, decreasing the levelized cost of (wind) energy (LCOE). The aims of the current comprehensive research program center on the quantification of the characteristics of the nonsteady loads resulting from the interactions between the coherent energy contain gin atmospheric turbulence eddies within the lower ABL as the eddies advect through the rotor plane and the rotating wind turbine blade encounter the internal turbulence structure of the atmospheric eddies. We focus on the daytime atmospheric boundary layer, where buoyancy due to surface heating interacts with shear to create coherent turbulence structures. Pseudo-spectral large eddy simulation (LES) is used to generate an equilibrium atmospheric boundary layer over flat terrain with uniform surface roughness characteristic of the Midwest on a typical sunny windy afternoon when the ABL can be approximated as quasi-steady. The energy-containing eddies are found to create advective time-responses of order 30-90 seconds with lateral spatial scales of order the wind turbine rotor diameter. Different wind turbine simulation methods of a representative utility scale turbine were applied using the atmospheric turbulence as inflow. We apply three different fidelity wind turbine simulation methods to quantify the extent to which lower order models are able to accurately predict the nonsteady loading due to atmospheric turbulence eddies advecting through the rotor plane and interacting with the wind turbine. The methods vary both the coupling to the atmospheric boundary layer and the way in which the blade geometry is resolved and sectional blade forces are calculated. The highest fidelity simulation resolves the blade geometry to capture unsteady boundary layer response and separation dynamics within a simulation of the atmospheric boundary layer coupling the effect of the turbine to the atmospheric inflow. The lower order models both use empirical look-up tables to predict the time changes in blade sectional forces as a function of time changes in local velocity vector. The actuator line method (ALM) is two-way coupled and feeds these blade forces back into a simulation of the atmospheric boundary layer. The blade element momentum theory (BEMT) is one-way coupled and models the effect of the turbine on the incoming velocity field. The coupling method and method of blade resolution are both found to have an effect on the ability to accurately predict sectional blade load response to nonsteady atmospheric turbulence. The BEMT cannot accurately predict the timing of the response changes as these are modulated by the wind turbine within the ABL simulations. The lower order models have increased blade sectional load range and temporal gradients due to their inability to accurately capture the temporal response of the blade geometry to inflow changes. Taking advantage of horizontal homogeneity to collect statistics, we investigate the time period required to create well converged statistics in the equilibrium atmospheric boundary layer and find whereas the 10-minute industry standard for `averages' retains variability of order 10%, the 10-minute average is an optimal choice. We compare the industry standard 10-minute averaging period. The residual variability within the 10-minute period to the National Renewable Energy Laboratory (NREL) Gearbox Reliability Collaborative (GRC) field test database to find that whereas the 10-minute window still contains large variability, it is, in some sense, optimal because averaging times much longer would be required to significantly reduce variability. Turbulence fluctuations in streamwise velocity are found to be the primary driver of temporal variations in local angles of attack and sectional blade loads. Based on this new understanding, we develop analyses to show that whereas rotor torque and thrust correlate well with upstream horizontal velocity averaged over the rotor disk, out-of-plane bending moment magnitude correlates with the asymmetry in the horizontal fluctuating velocity over the rotor disk. Consequentially, off-design motions of the drivetrain and gearbox shown with the GRC field test data are well predicted using an asymmetry index designed to capture the response of a three-bladed turbine to asymmetry in the rotor plane. The predictors for torque, thrust and out-of-plane bending moment are shown to correlate well to upstream rotor planes indicating that they may be applied to advanced feed-forward control methods such as forward-facing LIDAR used to detect velocity changes in front of a wind turbine. This has the potential to increase wind turbine reliability by using controls to reduce potentially detrimental load responses to incoming atmospheric turbulence and decrease the LCOE.