It is well documented that various large scale quasi-periodic flow structures, such as a Precessing Vortex Core (PVC) and multiple vortex helical instabilities, are present in the swirling flows typical of air swirl fuel injectors. Prediction of these phenomena requires time-resolved computational methods. The focus of the present work was to compare the performance and cost implications of two CFD methodologies — Unsteady Reynolds Averaged Navier Stokes (URANS – using a k-ε model) and Large Eddy Simulation (LES) for such flows. The test case was a single stream radial swirler geometry, which has been the subject of extensive experimental investigation. Both approaches captured the gross (time-mean) features of strongly swirling confined flows in reasonable agreement with experiment. The temporal dynamics of the quadruple vortex pattern emanating from within the swirler and observed experimentally were successfully predicted by LES, but not by URANS. Spectral analysis of two flow configurations (with and without a central jet), revealed various coherent frequencies embedded within the broadband turbulent frequency range. LES reproduced these characteristics, in excellent agreement with experimental data, whereas URANS predicted the presence of coherent motions but at incorrect amplitudes and frequencies. For the no jet case, LES-predicted spectral data indicated the occurrence of a PVC, which was also observed experimentally for this flow condition; the URANS solution failed to reproduce this measured trend. On the evidence of this study, although k-ε based URANS offers considerable computational savings, its inability to capture the temporal characteristics of the flows studied here sufficiently accurately suggests that only LES-based CFD, which captures the stochastic nature of the turbulence much more faithfully, is to be recommended for fuel injector flows.

This content is only available via PDF.
You do not currently have access to this content.