In order to increase the accuracy of the simulations, a new mesh was built on the same geometry: Nx=200 cells, Ny=50 cells, Nz=10 cells, that makes a total number of cells N=100 000.

We were now confronted to a new problem: each iteration was taking around 5 minutes !!! The lack of RAM on the SUN stations appeared to be the reason for this incredibly slow computation. Thus, we moved to the PC running windows NT. Equiped with PII 400 chip and 256 Mo of RAM, each iteration would now take around 13 seconds !!!

With this new mesh, each run would now converge in about 30 iterations with convergence criterium set at 1E-4.

Residuals after 50 iterations.

Vz after 50 iterations.

To test the quality of the simulations, we made very quick tests to estimate the numerical critical value of Rayleigh number.

Converged value of Vz at Ra=1600.

Converged value of Vz at Ra=1720.

The goal was now to to visualise 3D instabilities and to run unsteady simulations but the problems encountered with the first mesh, the SUN stations and the version of Fluent on the NT stations left very few time for this study. Moreover, the size of "*.cas" and "*.dat" files (nearly 10Mo each !) lead us to ask the administrators for special permissions.

We ran simulations with high values of Epsilon (from 10 to 40) and it appears that the rolls seem to be disorted.

However, we did not manage to make significant unstationnary runs to visualise the oscilations of the rolls.

We ran a few simulations with periodic conditions on the sides of the rolls. The results are quite encouraging but we did not have enough time to make further investigations.

VOF tests
back to summary
back to reports