Best-practice guidelines for Wind Engineering applications

A number of collaborative studies have been made [Franke et al., 2007; Menter et al., 2002; Tominaga et al., 2008b; Yoshie et al., 2007b] to develop best-practice guidelines for CFD applied to wind engineering problems. The main findings are

summarized below.

2.4.3.1 Computational Domain

In case of wind tunnel comparative studies, the dimensions of the computational domain should reproduce the geometry of the boundary layer wind tunnel test- ing as recommended by Best Practice guidelines [Franke et al., 2007]. For field measurements comparative studies in urban areas the height of the computational domain should include at least the height of the boundary layer as determined by the upstream terrain classification [Tominaga et al., 2008b] and 5Hmax away from the tallest (Hmax) building [Franke et al., 2007]. Such large dimensions avoid an artificial acceleration of the flow, since most boundary conditions prevent the flow out of the top of the domain. The lateral boundaries should be placed around 5Hmax from the edges of the region of interest [Tominaga et al., 2008b], or even closer as recommended by [Franke et al., 2007]. In any case, testing two different configu- rations is preferable, since the impact of the lateral boundaries on the flow in the targeted buildings is highly case dependant. The inflow boundary should be located about 5Hmax from the built area and the outlet about 15Hmax behind it depend- ing on outflow boundary conditions. For outflow boundary conditions, where the derivatives of all flow variables are forced to vanish the boundary should be placed far enough away to allow the flow redevelopment. Regarding surrounding buildings of the region of interest in urban areas, they should be explicitly reproduced if they are located less than 6Hn away from the area of interest, where Hn is the height of the building. However, parametric simulations can be performed, with and without the distant features, in case of uncertainty regarding their influence on the flow in the area of interest. The boundary conditions applied always play a key role on the decisions for the geometry of the domain [Franke et al., 2007].

2.4.3.2 Computational mesh

In the urban environment the flow field around the buildings is characterised by separation near the walls and the roof. In order to capture these important phe- nomena a fine grid arrangement is required near the corners and a minimum of 10 cells per building side [Yoshie et al., 2007b]. In regions with high velocity gradients the grid should be almost equidistant or at least the stretching ratio remain less than 1.3 [Tominaga et al., 2008b]. As the grid resolution is highly case dependent, a sensitivity study with at least three systematically refined grids is recommended [Franke et al., 2007]. Although the use of wall functions is not recommended for the building surfaces, their influence is not important in the case of bluff bodies with sharp edges where the separating points do not depend on the Re numbers and are always formed at the leading edges. Regarding the shape of the cells, a hexahedral mesh is preferred to tetrahedra and the grid lines on the wall should be perpendicular to it [Menter et al., 2002].

2.4.3.3 Boundary conditions

The computational domain includes the region of interest and the surrounding fea- tures at some distance from it. The rest of the elements influencing the flow should be represented implicitly using the boundary conditions. Most often, the boundary conditions are not well known and the assumptions used introduce an uncertainty to the solution. Hence, the boundaries to the computational domain should be located far away from the targeted area to have a small effect on the results [Franke et al.,

Inflow boundary conditions

The vertical velocity profile is given either by a power law as suggested byTominaga et al.[2008b] or a log law according to the [COST] recommendations [Franke et al.,

2004] that use the formulas suggested by Richards and Hoxey [1993]. The power law is given by the equation:

U (z) = Us( z zs

)a (2.27)

where Us is the velocity at reference height zs and a is the power-law exponent determined by terrain category [Choi, 2009]. The log law is given by the equation:

U (z) = U ∗ ABL κ ln( z_{− d} z0 ) (2.28)

where κ is the Karman constant (= 0.4) and U*

ABL the atmospheric boundary layer friction velocity, which assumes a constant shear stress with height and hence the computational domain should be much lower than the atmospheric boundary layer [Tominaga et al., 2008b]. If wind tunnel data are available, they should be used.

Top and lateral boundary conditions

In a large computational domain the top and lateral boundaries slightly influence the solution in the area of interest [Mochida et al., 2002; Yoshie et al., 2007b]. When the top boundary is outside the boundary layer and the lateral boundaries far away from the built area the use of symmetry conditions which impose zero normal velocity, makes the computation more robust [Tominaga et al., 2008b]. If wind tunnel measurements are available and obtained within a closed test section, the top boundary should be classified as solid wall [Franke et al., 2007].

Outflow boundary conditions

Outflow boundary conditions are usually used at the boundary downstream and all the derivatives of the flow variables are set to zero. Therefore, this boundary should be placed far enough away from the built area to allow the flow to fully redevelop.

Ground surface and building wall

For the velocities at building and ground surfaces, the no-slip boundary condition is used. For the shear stress in urban areas the smooth wall condition is recommended byFranke et al. [2007]. The rough wall condition usually leads to a bad resolution close to the wall, since the first calculation node of the wall must be placed at least ks away from the wall [Tominaga et al., 2008b], where ks is the roughness height.

Blocken et al. [2012] have shown a method to use the wall roughness properties and circumvent this limitation using a commercial CFD code, but this leads to significant streamwise changes of the inflow profiles. In practice, the shear stress is estimated using the wall functions which apply a logarithmic velocity profile between the wall and the first computational node in the wall-normal direction. For the logarithmic profile to be valid, the first computational node should be placed at a non-dimensional wall distance of z+ between 30 and 500 for smooth walls [Franke et al.,2007].