Uncertainty Quantification of Reynolds-Averaged Navier-Stokes Turbulence Model in Predicting Centrifugal Compressor Performance

The prediction of the aerodynamic performance of a centrifugal compressor using the Reynolds-averaged Navier-Stokes (RANS) turbulence model often shows significant discrepancies with experimental results. To quantify the uncertainty of RANS models in predicting centrifugal compressor performance, the eigenvalues and eigenvectors of the realizable, physically constrained Reynolds stress tensor are perturbed. Specifically, a non-uniform perturbation method is applied within the barycentric triangle to disturb the base flow toward the three limiting states of turbulence, thereby obtaining the extrema of the production terms in the turbulence transport equations. This approach enables the quantification of the L2 uncertainty in the RANS model. At operating conditions on the left of the performance curve, the confidence interval fully encompasses the maximum deviations of 4.37% in total pressure ratio and 0.9% in polytropic efficiency observed between the unperturbed simulation and experimental results. Besides, the influence of different perturbations on predicting flow separation onset on the suction surface of a centrifugal impeller blade is investigated. The results indicate that eigenvalue perturbations of a single component of the Reynolds stress tensor yield more conservative predictions of flow separation, whereas eigenvector perturbations of a single component result in more aggressive predictions. The uncertainty in the RANS model’s predictions of performance boundaries can further be utilized for cross-validation with experimental measurement data.