This thesis describes the different steps needed to design a steady-state computational fluid dynamics (CFD) wind farm wake model. The ultimate goal of the project was to design a tool that could analyze and extrapolate systematically wind farm measurements to generate wind maps in order to calibrate faster and simpler engineering wind farm wake models. The most attractive solution was the actuator disc method with the steady state k-ε turbulence model.
The first step to design such a tool is the treatment of the forces. This thesis presents a computationally inexpensive method to apply discrete body forces into the finite-volume flow solver with collocated variable treatment (EllipSys), which avoids the pressure-velocity decoupling issue.
The second step is to distribute the body forces in the computational domain accordingly to rotor loading. This thesis presents a generic flexible method that associates any kind of shapes with the computational domain discretization. The special case of the actuator disc performs remarkably well in comparison with Conway's heavily loaded actuator disc analytical solution and a CFD full rotor computation, even with a coarse discretization.
The third step is to model the atmospheric turbulence. The standard k-ε model is found to be unable to model at the same time the atmospheric turbulence and the actuator disc wake and performs badly in comparison with single wind turbine wake measurements. A comparison with a Large Eddy Simulation (LES) shows that the problem mainly comes from the assumptions of the eddy-viscosity concept, which are deeply invalidated in the wind turbine wake region. Different models that intent to correct the k-ε model's issues are investigated, of which none of them is found to be adequate. The mixing of the wake in the atmosphere is a deeply non-local phenomenon that is not handled correctly by an eddy-viscosity model such as k-ε .