This title appears in the Scientific Report : 2020 

High-Performance Flow Simulation and Scale-Adaptive Turbulence Modelling of Centrifugal Pumps
Hundshagen, Markus
Casimir, Nicolas / Pesch, Andreas / Skoda, Romuald (Corresponding author)
John von Neumann - Institut für Computing; NIC
NIC Symposium 2020
Jülich Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag 2020
367 - 378
NIC Symposium 2020, Jülich (Germany), 2020-02-27 - 2020-02-28
Contribution to a book
Contribution to a conference proceedings
Publication Series of the John von Neumann Institute for Computing (NIC) NIC Series 50
Please use the identifier: in citations.
While for the design point operation of centrifugal pumps, where an essentially steady flow field is present and statistical turbulence models yield an appropriate prediction of the characteristics, the flow field gets increasingly unsteady towards off-design operation. Special designs as e. g. sewage pumps are characterised by a single-blade impeller and show significantly unsteady characteristics even in the design point. For such highly-unsteady and turbulent flow fields, statistical models tend to fail. On the other hand, Large-Eddy Simulation models, where the large-vortex part of the turbulent spectrum is directly resolved, show a much better flow prediction. However, the spatial resolution and thus computational effort are too high for engineering real pump applications. Therefore, we provide an assessment of scale-adaptive turbulence simulation (SAS) models that recover a statistical flow solution in regions of low unsteadiness and – like Large-Eddy Simulation – resolve a part of the turbulent spectrum down to the available grid resolution for highly unsteady flow regions. After a thorough validation on standard turbulence test cases e. g. the periodic hill case, it is shown that with a moderately higher computational effort than statistical models, the SAS yields a considerable improvement of the prediction of the turbulence field in part load operation of a centrifugal pump while the mean flow field could be well predicted even with a well-established statistical model.