[ihsb]

Hi,

I have observed a problem running CFX V15.0 parallel that I can't explain.

The problem I am trying to calculate is an axial turbine. The geometry is build up with one guide vane passage one runner passage and a full 360°-draft tube with an artificial extension. The simulation is performed steady state with an physical time scale which would correspond to 5° runner revolution. Stage interfaces are used between the rotor and the stator.

When running serial the simulation is running without any problems. It is also possible to run the simulation on 4 cores parallel without any problems.

The problem appears if one wants to increase the amount of parallel partitions the run. The run is aborting after a couple of iterations with an overflow.
In the out-file it can be seen that the solver has placed 100% wall at the outlet.

I tied the following ways to solve the problem:
1) Putting an opening boundary condition at the outflow
2) Increasing and decreasing the time step
3) Starting the run from an initial result file, computed on less cores
4) Using a full 360°-geometry of guide vane and runner (with stage interface)
5) Running a full 360° transient simulation (with transient interfaces between the rotor stator domains) with different time step sizes
6) Using CFX 14.5 for the steady state simulations.

All the mentioned approaches didn't solve the problem. To be honest I actually don't know what else I can try to solve the problem. Does anyone have an advice how to solve the problem?

Thanks!

[ghorrocks]

Try different partitioning methods.

Also, you have forgotten to try the most important thing when dealing with numerical difficulties - improve the mesh quality.
And double precision numerics are worth a try as well.