What is the importance of turbulence modelling in capturing accurate 3D secondary flow and mixing losses in turbomachinery? An investigation on the effect of return channel (RCH) dimensions of a centrifugal compressor stage on the aerodynamic performance was studied to answer this question by A. Hildebrandt and F. Schilling as an effort to push turbomachinery one step further.
W. Fister was among the first to investigate the return channel flow using 3D-CFD. At that time the capability of commercial software was not extended and any computational effort was limited by the CPU-capacity. Therefore, only simplified calculations that included constant density without a turbulence model (based on the Prandtl mixing length hypothesis) embedded in in-house code, were performed.
Although separated flow without a predominant flow direction could not have been calculated, the method indicated separated flow regions with relatively accurate precision, and it predicted the magnitude of loss coefficients to a higher degree than experimental data. The study was further
simplified using incompressible flow, and an axial U-turn inlet flow.
The biggest drawback of using inverse methods for return channel design refers to the question of appropriate flow distribution across the RCH surface. Furthermore, flow separation cannot be predicted with the help of singularity methods. In order to circumvent the problem of predicting flow separation, nowadays compressible viscous 3D-CFD applied with different highly complex turbulence modeling is the state of the art even at the conceptual stage of the design.
Hildebrandt and F. Schilling analyzed three different centrifugal stages regarding the return channel system performance. All three stages featured the same impeller type, two of them being applied with a 3D-RCH at different flow coefficient and one impeller being applied with a 2D-RCH system. The 3D-RCH stage featured both CFD calculated and measured superior aerodynamics over the 2D-RCH stage regarding the overall performance as well as regarding the outlet flow angle. The comparison between the measured and the CFD-predicted performance showed agreement both when it comes to overall performance (efficiency, pressure rise coefficient) and also regarding detailed flow field (outlet flow field). The 3D secondary flow and mixing losses of the entire domain downstream the vaneless diffuser were either underestimated or overestimated by the CFD-calculations, depending on the turbulence modeling and the impeller fillet radii-modeling which affects the RCH-inlet flow conditions. The effect of fillet radii-modeling on the RCH-exit flow angle spanwise distribution was found to be significant in order to better match the experimental results.
It is worth noting that the rather simple Spalart–Allmaras turbulence model provided better agreement with the measured RCH-exit flow angle distribution than the more sophisticated k-epsilon model, which on the other hand, outputted a closer fit with the measured surface vane pressure distribution. Regarding the RCH total pressure loss distribution, none of the models showed a perfect agreement with the measurement data.
Moreover, the incident losses of the 3D-RCH system seemed to play a minor role within the overall RCH-loss which is significantly dominated by the 3D-secondary losses.