The continuously rising global demand for energy together with simultaneously decreasing resources has made the topic of energy efficiency — and therefore optimization — one of the fundamental questions of our time. Turbomachinery is one of the most important parts of the process chain in nearly every case of energy conversion. This makes the turbomachine a promising approach point for optimizations. The special relevance of this topic in regard to the global challenge of climate change can be illustrated by a simple calculation: If the efficiency of a turbo compressor with a power consumption of 15MW is improved by one percent, approximately 2t CO2 per day or over 760t CO2 per year can be saved.

This work describes the optimization of the operation characteristic of a highly stressed centrifugal compressor impeller with regard to the size of the operation range and the efficiency in the operation point. The base impeller used for this optimization has already been pre-optimized by classical engineering methods utilizing analytical and empirical models. Due to the high mechanical stress in these kind of turbo impellers, each design has to be checked for compliance with the structural constraints in addition to the fluid dynamic computations. This context results in a highly complex, multicriterial, high dimensional optimization problem. The main subjects of the presented work are a robust geometry generation and grid generation, a highly automated workflow for the computation of the operation characteristic and the mechanical results and the representation of the operation characteristic by scalar parameters. Utilizing these tools a DOE is performed and based on its results a metamodel is created. The optimization is carried out on the metamodel using a Particle Swarm algorithm.

The workflow presented in this work utilizes in-house preprocessing tools as well as the tools of the ANSYS Workbench. The operation characteristics are computed using an in-house tool to control the ANSYS CFX-Solver. The statistical and stochastic pre- and post-processing as well as the metamodeling are carried out in optiSLang.

This content is only available via PDF.
You do not currently have access to this content.