Session: 34-13 High-fidelity CFD for Compressors II

Submission Number: 177519

Advances in CFD Prediction Using DES/LES Turbulence Models in a High-Speed Centrifugal Compressor 

The time-averaged results of a Detached-Eddy Simulation (DES) and an SST-RANS turbulence model are compared in their ability to simulate the aerodynamic performance of a rear-stage centrifugal compressor with an impeller and a fish-tail diffuser. The turbulence model results are compared to rig test data and Laser Doppler Velocimetry (LDV) data. The addition of surface roughness to both the SST turbulence model and the DES turbulence model showed an improved accuracy of pressure ratio and inlet corrected mass flow compared to the smooth model, with an improvement in efficiency solely with the DES model. While initially DES predicted a larger decrease than SST, increasing the roughness value from the inspection’s measured roughness to the drawing maximum roughness, DES saw the pressure ratio and inlet corrected mass flow increasing by 0.041% and 0.196% respectively and efficiency dropped by only 0.3% compared to the initial decrease of 1.05%. This pattern was the opposite for the temperature ratio, showing a smaller change than SST initially and an overall decrease in accuracy. The total effect of adding surface roughness to the model corresponded in both SST and DES, demonstrating a dominance of the impact of surface roughness over the impact of the turbulence model.

 

Performance sensitivity to mesh resolution revealed that for the coarse mesh (7.7 million nodes), the DES model deviated only by 0.013% in total pressure ratio compared to experimental data, while the SST RANS model showed a 1.612% deviation. In contrast, with the fine mesh (15.9 million nodes), SST RANS accuracy improved to a deviation of just 0.842%, whereas DES recorded a slightly negative deviation of –0.268%, suggesting an underprediction of pressure rise under finer meshing. Similar subtleties emerged in stage efficiency: the coarse mesh result for DES lagged by –0.889%, compared to –0.069% for SST, while in the refined mesh scenario, the roles reversed, DES performed better (–0.769%) than SST (–1.149%), albeit with only minor margins separating them. The fine mesh showed an overall improved accuracy for all performance parameters whereas the coarse mesh had a loss of accuracy in efficiency. The DES blending function comparison at streamwise sections along the impeller showed that the fine mesh was using LES mode 7% more than the coarse mesh which showed a slower transition between LES and SST mode. Incorporating DES variants with enhanced shielding, SDES and DDES, further escalated LES-mode utilization and noticeably influenced the fidelity of the results. DES showed improvements in modelling velocity profile. The DES model pressure prediction deviated from the LDV circumferential velocity data by 0.440% while the SST RANS model showed a 1.498% deviation. The SDES model performed similarly to SST in predicting the radial velocity with a deviation of 0.292% compared to SST’s deviation of 0.282%. The absolute angle saw DES (0.184%) again perform better than SST (0.237%).

 

The γ-Reθ transitional model with DES showed two peak velocity locations in the diffuser exit axial velocity profile, distinct from the one peak area of the other models. The γ-Reθ DES model improved the overprediction of the large high velocity area shown in the circumferential diffuser exit velocity profiles. The computational costs and runtime required for DES models limit the viability as the default turbulence model for centrifugal compressors as the SST RANS model operates with markedly lower computational demands. The additional insight into flow patterns from DES modelling is best gained with a hybrid investigation using both models, reserving DES for detailed analysis.

 

Keywords: detached eddy simulation, centrifugal compressor

Presenting Author: Zara Menheere Pratt and Whitney Canada

Presenting Author Biography: Zara Menheere is finishing her bachelor's degree in Engineering Physics-Mechanical at Queen's University in Kingston, Ontario. She completed a 16-month professional internship at Pratt and Whitney Canada in the Compressor Aerodynamics department.

Authors:

Zara Menheere Pratt and Whitney Canada
Mehdi Ebrahimi Pratt and Whitney Canada
Hong Yu Pratt and Whitney Canada