Strategies for Multi-Fidelity Optimization of Multi-Stage Compressors with Throughflow and 3D CFD

Abstract

The aerodynamic design of turbomachinery is an exciting engineering problem that involves a vast design space and requires proficiency in multiple engineering disciplines. For multi-stage compressors, the large dimension of the design space arises from the geometric freedom to shape each blade row individually. Although the basis of most designs is formed by throughflow computations, the state-of-the art for aerodynamic evaluation is steady-state RANS. The computational expense for a single evaluation of a multi-stage machine with RANS is affordable, shape optimizations with hundreds of evaluations becomes expensive. This paper presents two strategies to tackle these problems: Firstly, a novel airfoil family is employed that helps to reduce the number of design parameters. It was generated based on a parametric study of optimal airfoil shape for varying design requirements. Secondly, an optimizer is used that supports simultaneous evaluations of the objective function on multiple fidelity levels. Here, cheap throughflow calculations are conducted on the low-fidelity side and RANS is used on the high-fidelity side. Inside the optimizer, multi-fidelity support is enabled by a Co-Kriging surrogate model. This approach is expected to speed-up the optimization by guiding expensive evaluations of the 3D CFD setup into promising regions of the design space. These strategies are assessed by testing different optimization setups for a four stage compressor. Both purely low and high-fidelity optimizations are conducted as well as multi-fidelity optimizations. In comparison to the baseline design, the isentropic efficiency predicted by RANS for the working line points on the 95% and the 100% speed line is increased by above 1.2%. The driving mechanism is a load redistribution between the stages that mainly reduces the shock strength of the second rotor.