Sensitivity analysis and optimization of a CO2 centrifugal compressor impeller with a vaneless diffuser

Abstract

Fluid flow in centrifugal compressors is complex and turbulent, making it difficult to achieve a robust equipment design. In this work, a computational fluid dynamics (CFD) analysis is conducted on the periodical domain of a CO 2 centrifugal compressor impeller and vaneless diffuser, with a 2.85:1 pressure ratio, using ANSYS CFX. The fluid flow is assumed to be steady state, turbulent, and three dimensional. Since polar angles are not considered in the traditional one-dimensional analysis, eight polar angles on hub and shroud were considered for modeling, sensitivity analysis, and optimization. After numerical verification and validation, a sequential sensitivity analysis (SA) was performed to identify non-influential variables. From the same design of experiment (DoE) used by the Morris qualitative SA, a response surface (RS) was trained to perform a quantitative SA by the smoothing spline ANOVA (SS-ANOVA) method. The Morris method was found to be more conservative than SS-ANOVA, keeping more variables as influent for the analysis. Both methods agreed on influential variables ranking. Low computational effort was required to submit the RS to a constrained optimization procedure using the NSGA-II method. The polytropic efficiency of the optimal centrifugal compressor configuration increased 0.7%, keeping the pressure ratio above 2.85, and the required power and outlet temperature below the base compressor. The impact of the polar angles at trailing edge on output variables is higher than leading edge. The optimal centrifugal compressor found is submitted to different mass flow rates and the overall performance for the optimal angles of the trailing edge and leading edge of the impeller was higher than the base compressor. The strategy adopted herein related to qualitative and quantitative sensitivity analysis coupled with response surface and the constrained optimization was shown to be robust, which can be applied to high-dimensional CFD models to reduce the computational cost with suitable results regarding fluid flow phenomena.