Hybrid RANS/LES Simulation of Corner Stall in a Linear Compressor Cascade

Current design-cycle Reynolds-averaged Navier–Stokes (RANS) based computational fluid dynamics (CFD) methods have the tendency to over-predict corner-stall events for axial-flow compressors operating at off-design conditions. This shortcoming has been demonstrated even in simple single-row cascade configurations. Here we report on the application of hybrid RANS/large eddy simulation (LES), or detached eddy simulation (DES), for simulating the corner-stall data from the linear compressor cascade work conducted at Ecole Centrale de Lyon. This benchmark data set provides detailed loss information while also revealing a bimodal behavior of the separation which, not surprisingly, is also not well modeled by RANS. The hybrid RANS/LES results presented here predict bimodal behavior similar to the data only when special treatment is adopted to resolve the leading-edge endwall region where the horseshoe vortex (HV) forms. The (HV) is shown to be unstable, which produces the bimodal instability. The DES simulation without special treatment or refinement in the HV region fails to predict the bimodal instability, and thus the bimodal behavior of the separation. This, in turn, causes a gross over-prediction in the scale of the corner-stall. The HV region is found to be unstable with rolling of the tertiary vortex (TV) over the secondary vortex and merging with the primary HV. With these flow dynamics realized in the DES simulations, the corner stall characteristics are found to be in better agreement with the experimental data, as compared to RANS and standard DES approaches.