Computational Performance

All computations were carried out on a single processor PC at 1800 MHz, running under LINUX. Starting from a steady state solution the unsteady com­putation took about 18 times to pass the rotor leading edge behind the NGV trailing edge in order to achieve a satisfactory periodical behaviour. The un­steady mass fbw was taken as a convergence criteria (Fig.3). The total CPU time was in the order of 20 days, requiring about 1 GB of RAM. The overall level of convergence was slightly fhctuating around three orders of magnitude reduction in the total RMS residual.

The unsteady calculations were carried out using the domain scaling tech­nique. The rotor pitch was brought from 60 to 64 blades, allowing to mesh two rotor blades with the same periodicity as one NGV pitch. For convergence acceleration dual time stepping was used. The rotor turning was resolved by 32 discrete angular positions for one rotor pitch.

3. Comparison Full Discretization/Source Term Approach

Apart from the human effort of meshing 120 additional cooling holes, the source term approach requires considerably less computational resources. The larger RAM requirements are obvious, considering the higher number of grid cells and blocks. In addition, the CPU time increases over-proportionally since the coupling between the main fbw and the cooling jets is much stronger in case of the fully discretized approach. Here, convergence is slowed down due to the slow propagation from the main fbw through the holes into the plenum.

4. Results