Computational Fluid Dynamics Models

where α0 is the zero lift angle of attack, and αSS is the static stall angle of attack. A

flow chart of the DMS model is shown in Figure 4.3, where the iteration process for each streamtube and the dynamic stall coefficient corrections are shown.

4.2.3 Computational Fluid Dynamics Models

The blade forces of the one and three-bladed turbines were simulated using two 3D CFD models that were meshed using ANSYS CFX 13.0 using unstructured tetrahedral elements (ANSYS 2010a). These models included all blades and were geometrically equivalent to the EFD turbines to allow validation of the numerical simulation approaches utilized against the published experimental results (Strickland et al. 1979, Webster 1978). The density of the mesh was varied according to expected flow curvature rates with increased density in regions near the wake and blade regions. Alternatively, mesh density was reduced away from the turbine surfaces to minimise computational effort, such as near the computational domain boundaries. The boundary layers near the turbine surfaces were fully resolved using 30 lay- ers, with the total boundary layer thickness estimated using turbulent boundary layer theory (Marsh et al. 2015a,b). This estimated thickness was doubled to ensure that the boundary layer was contained within the prescribed inflation layer region. Inflation layer mesh growth rates were limited to 1.2 as recommended (ANSYS 2010a). Due to the proximity of the tips

Turbine rotation was simulated by enclosing the turbine in an inner domain as shown in Figure 4.4. This domain was rotated at 0.746 rads−1 (λ = 5) corresponding to the EFD rotational rate (Strickland et al. 1979, Webster 1978) using the CFX transient rotor-stator model (ANSYS 2010a). This model uses a General Grid Interface (GGI) to interpolate flow values across the interface due to non-conformal mesh. To minimise any errors in the inter- section algorithm, the GGI was placed at 1.5 times the turbine diameter measured from the rotational axis (Marsh et al. 2013).

Figure 4.4: Domain boundary nomenclature and sizing. Dimensions in meters (m) and turbine diameters (D)

The CFD domain dimensions are shown in Figure 4.4, with the associated boundary con- ditions outlined in Table 4.3. The width and depth of the computational domain were set to that of the EFD towing tank to account for any blockage effects. However, the domain length was determined by doubling the domain length until variations inFnand Ftreduced

to less than 5% between successive length refinements, thus allowing full wake development. Simulations were performed to investigate the influence of domain size and boundary con- ditions on blade loading parameters. Results indicated that free slip walls could be used instead of no slip walls on the domain sidewalls, reducing the wall mesh density and thus increasing computational efficiency. To reduce computational effort the waters surface was modeled using a free slip wall, which simulated the surface pressure effects whilst simplifying computational requirements when compared to a full multiphase approach. Examination of simulation results against EFD from literature was performed to ensure the validity of this free surface modeling assumption. The bottom tank wall modeled using a no slip boundary condition to capture any boundary layer effects due to the proximity of the blades to the bottom wall.

Table 4.3: Domain Boundary Conditions for the CFD model (Strickland et al. 1979, Webster 1978)

Boundary Condition

Inlet Uniform flow: 0.091 ms−1 Outlet Relative pressure: 0 Pa

Walls Free slip walls

Blade No slip walls

Bottom No slip wall

Top Free slip wall

separation and adverse pressure gradients by the inclusion of transport effects into the for- mulation of the eddy-viscosity (ANSYS 2010a). This model has also demonstrated high simulation accuracy for turbine power output simulations when compared against EFD re- sults (Dai & Lam 2009, Lain & Osorio 2010, Marsh et al. 2012, 2013, 2015a,b, Nobile et al. 2011). The fluid was modeled as incompressible as all flow velocities were significantly less than Mach 0.3. Fluid flow was assumed to be fully turbulent, with no laminar-to-turbulent transitional effects considered to simplify computational effort. High order advection and second order backward Euler transient terms were used to ensure numerical accuracy, as previous simulations of vertical axis turbines found that low order schemes were unable to accurately resolve turbine performance parameters (Marsh et al. 2015a). An inlet turbu- lence setting of 5% was applied as no measurements of turbulence intensity were performed during EFD testing (Strickland et al. 1979, Webster 1978). Convergence was achieved when solution residuals reduced to below 10−4 and reduced by more than three orders of mag- nitude, similar to previous works (Marsh et al. 2015a,b). To minimise convergence times and hence solution time, all simulations were started from previous solutions, thus reducing overall computational requirements.

Mesh Independence Studies

Systematic independence studies were performed for the one and three-bladed turbine simu- lation models to ensure spatial, temporal, domain length, and boundary layer mesh indepen- dence, similar to that performed previously (Marsh et al. 2012, 2013, 2015a,b). Independence was achieved when doubling of these factors resulted in variations in Fn and Ft trending to

less than 5% between successive refinements, with an example of the trends shown in Figure 4.5. Mesh element count independence was determined forFn and Ft at 7.4 x 106 and 14.2

x 106 mesh elements for the one bladed and three-bladed turbines respectively. The values forFtwere highly influenced by mesh element count, as drag predictions are highly depen-

Figure 4.5: Mesh element count independence for three-bladed turbine at a rotational rate of 0.746 rads−1 and an inflow velocity of 0.091 ms−1

Time step studies were performed for the one and three-bladed turbine models to ensure temporal independence, critical due to the highly transient nature of the flow. Temporal independence was demonstrated at 0.9◦ of rotation per iteration step forFnandFtfor both

turbine models, with an example shown in Figure 4.6 for the three-bladed turbine. Again the values forFt were highly influenced by mesh element count, as drag predictions are highly

dependent upon the mesh density in the chordwise direction and on the leading edges. The time step size was low due to the high mesh density of the fully-resolved boundary layers.

Additional independence studies were performed for domain length and boundary layer mesh density. Domain length independence was determined at 24.4m for both turbine models, which allowed for full wake development. Given that the EFD tank length was limited, EFD force data was obtained on the fourth revolution of the turbine to minimise any startup tran- sient effects. Comparisons of CFD results with EFD were performed to determine whether modeling the full wake replicated the blade forces with sufficient accuracy when compared to EFD results. For both turbine models boundary layer mesh density independence was evaluated by examining the influence onFnandFtof the average height of the first cell from

the turbine walls, known as the non-dimensional variabley+ (Paraschivoiu 2002). Indepen- dence was demonstrated at an average y+=0.1 using methods similar to those outlined in (Marsh et al. 2015a,b).