Abstract
Rotating stall is a non-axisymmetric disturbance in axial compressors arising at operating conditions beyond the stability limit of a stage. Although well-known, its driving mechanisms determining the number of stall cells and their rotational speed are still marginally understood. Numerical studies applying full-wheel 3D unsteady RANS calculations require weeks per operating point. This paper quantifies the capability of a more feasible quasi-2D approach to reproduce 3D rotating stall and related sensitivities.
The first part of the paper deals with the validation of a numerical baseline the simplified model is compared to in detail. Therefore, 3D computations of a state-of-the-art transonic compressor are conducted. At steady conditions the single-passage RANS CFD matches the experimental results within an error of 1% in total pressure ratio and mass flow rate. At stalled conditions, the full-wheel URANS computation shows the same spiketype disturbance as the experiment. However, the CFD underpredicts the stalling point by approximately 7% in mass flow rate. In deep stall, the computational model correctly forecasts a single-cell rotating stall. The stall cell differs by approximately 21% in rotational speed and 18% in circumferential size from the experimental findings. As the 3D model reflects the compressor behaviour sufficiently accurate, it is considered valid for physical investigations.
In the second part of the paper, the validated baseline is reduced in radial direction to a quasi-2D domain only resembling the compressor tip area. Four model variations regarding span-wise location and extent are numerically investigated. As the most promising model matches the 3D flow conditions in the rotor tip region, it correctly yields a single-cell rotating stall. The cell differs by only 7% in circumferential size from the 3D results. Due to the impeded radial migration in the quasi-2D slice, however, the cell exhibits an increased axial extent. It is assumed, that the axial expansion into the adjacent rows causes the difference in cell speed by approximately 24%. Further validation of the reduced model against experimental findings reveals, that it correctly reflects the sensitivity of circumferential cell size to flow coefficient and individual cell speed to compressor shaft speed. As the approach reduced the wall clock time by 92%, it can be used to increase the physical understanding of rotating stall at much lower costs.
