Three dimensional numerical prediction of icing related power and energy losses on a wind turbine

Regions of Canada experience harsh winter conditions that may persist for several months. Consequently, wind turbines located in these regions are exposed to ice accretion and its adverse effects, from loss of power to ceasing to function altogether. Since the weather-related annual energy production loss of a turbine may be as high as 16% of the nominal production for Canada, estimating these losses before the construction of a wind farm is essential for investors.;Our objective in this thesis is to develop a 3D numerical methodology to predict rime ice shape and the power loss of a wind turbine as a function of wind farm icing conditions. In addition, we compute the Annual Energy Production of a sample turbine under both clean and icing conditions. The sample turbine we have selected is the NREL Phase VI experimental wind turbine installed on a wind farm in Sweden, the icing events at which have been recorded and published. The proposed method is based on computing and validating the clean performance of the turbine, and then computing the ice shape and iced blade performance, under icing conditions. The first step is to compute the performance of the NREL Phase VI using the commercial ANSYSFLUENT computational fluid dynamics (CFD) tool. In order to reduce the computational cost, we use a rotating reference frame model which computes the flow properties as time-averaged quantities. A grid sensitivity study has been performed to eliminate the effect of mesh on the results. Of the existing models for characterizing turbulence, we have selected the two-equation SST k-pi model. In general, the computed pressure coefficients and bending moment have shown good agreement with the experimental data, particularly at pre-stall speeds. Although the torque deviates from the experimental data, the trend with respect to the wind speed is similar.;After the clean power curve has been computed, collection efficiency, which is directly proportional to the rate of icing of a surface, is analyzed. A multiphase analysis, for the air and water phases, is necessary to compute the rate of accumulation of the droplets on the blade surfaces. We study two different approaches that are found in the literature -- Eulerian and Lagrangian -- and determine the most suitable one for our study case. The former applies the governing equations to the liquid phase, while the latter computes the trajectory of each droplet present in the air. We eventually decided on the Eulerian model for our study, as it can be adapted to handle large and complex meshes better than the Lagrangian model. This step is validated on a NACA 0012 airfoil, as experimental data for 3D flows are not available in the literature.;The ice accretion on the sample wind turbine blades is computed using both a Quasi-3D and a Fully-3D method, which have a similar theoretical background, but a different order of modeling. In the former, all the steps are carried out in 2D and the overall power is computed using the Blade Element Momentum method, while the latter performs all the steps in the 3D domain. The Fully-3D method yields more accurate predictions for a clean blade. For icing conditions, a validation is not possible, owing to the lack of experimental data. However, the two methods produce quite different results for the performance of the ice shape and the iced blade. A critical analysis of the results shows that, although the computational cost of the Fully-3D method is much higher, icing analyses in 2D may lack accuracy, because the ice shape and the related power loss are compromised by not considering the 3D features of rotational flow.;A literature survey shows that most icing prediction methods and codes are developed for aircraft, and, as this information is mostly considered corporate intellectual property, it is not accessible to researchers in other domains. Moreover, aircraft icing is quite different from wind turbine icing. Wind turbines are exposed to icing conditions for much longer periods than aircraft, perhaps for several days in a harsh climate, whereas the maximum length of exposure of an aircraft is about 3-4 hours. In addition, wind turbine blades operate at subsonic speeds, at lower Reynolds numbers than aircraft, and their physical characteristics are different. A few icing codes have been developed for wind turbine icing nevertheless. However, they are either in 2D, which does not consider the 3D characteristics of the flow field, or they focus on simulating each rotation in a time-dependent manner, which is not practical for computing long hours of ice accretion.;While performing the CFD computations on the iced blade, the rough surface of the ice is smoothed to a degree, in order to prevent numerical instability and to keep the mesh size within a reasonable limit. However, roughness effects cannot be excluded altogether, as they contribute significantly to performance reduction. We consider roughness through a modification in the CFD code, and assess its effect on performance for the clean blade.