For this final timing exercise, we wished to not only employ another of the many OpenFOAM codes but also to model a real world simulation of industrial-scale. To this end, the author and Paul Garlick  worked closely to generate and run a 3D turbulent jet Large Eddy Simulation (LES).
The 3D model is a co-axial atomization model, where a central liquid jet is broken into droplets by a high-speed surrounding air jet. Applications of this technology are spray coating equipment and fuel combustors (for jet engines and gas turbines). This 3D case was based on the primary atomization example case in OpenFOAM, which itself is a 2D LES.
Initially, a relatively simple grid was developed for a typical geometry of a co-axial atomizer, whilst considering how to specify realistic boundary conditions to give a jet break-up length and drop sizes that can be predicted and compared to past experiment and analysis.
Then the 3D simulation was run using a Reynolds-Averaged Navier-Stokes (RANS) model, on a relatively coarse grid. The output was a steady-state solution which yielded information of the turbulence length scales. The minimum length scale was then used to define the cell size for the more detailed LES model. Finally, the results from the RANS model was used to initialize the LES model.
The image of the geometry for the atomization model is given in figure 4.
Figure 4, as generated by Paul Garlick , shows the inlet locations and boundary meshes. The liquid enters the domain through the central (blue) patch; the atomizing air enters through the annular (red) patch. This figurer uses a 1 x 1 x 1cm cube, and a mesh with elementSize=0.2mm. This sets the size of the elements at the inner edge of the air inlet patch. The element size reduces towards the centre-line, where the atomization occurs, and increases towards the outer boundaries.
For this benchmark, we have 3 different cases, namely jet_coarse, jet_medium and jet_fine, and are summarised in table 4.
In table 4 we can see the value for the time step, namely deltaT, is smaller for the fine case than for both the coarse and medium cases. This is because the maximum Courant number increases for the finer mesh, and this lower value ensures the the maximum Courant number will be 0.1.
The 2 x 2 x N decomposition should work well for small values of N. The Scotch Method may be more efficient for large N.
The file controlDict contains the following lines.
application pisoFoam; startFrom latestTime; startTime 0; stopAt endTime; endTime 0.00201; deltaT 5.0e-08; writeControl timeStep; writeInterval 200000; purgeWrite 0; writeFormat ascii; writePrecision 6; writeCompression compressed; timeFormat general; timePrecision 6; runTimeModifiable yes;