This paper contributes to the field of centrifugal compressor design methodology by proposing a method of geometry parameterization, then applying CFD analysis and optimization algorithms utilizing such parameterization. Most state-of the art design methodologies rely significantly on reduced-order loss models to explore the design space. While this is a useful and validated approach, it constrains the design to a previously explored domain. In the approach presented, kinematic velocity triangle analysis is used to arrive at an initial design, which is then improved using automated optimization algorithms and fundamental flow physics using CFD simulation. By allowing the design to evolve, guided by CFD, new and untested optimum designs are possible. This work deals specifically with the design and optimization of the flow path. Thus, design for structural aspects such as vibration, rotor dynamics, and other mechanical aspects is outside the scope of this work.

A key component to achieving this method is CAD parameterization of the machine geometry. The parameterization developed in this work enables robust specification of the flow path geometry while maintaining a sufficiently small set of parameters to make the problem tractable in a practical design space exploration. A specific parameterization of a single stage machine involving an impeller with optional splitter blades, a vaneless diffuser and a volute of a centrifugal compressor is revealed and analyzed. Steady-state RANS-based CFD is employed in the analysis of both the rotationally periodic components and the full volute geometry. The analysis is performed separately for demonstration of the method but can be coupled as a single analysis in future work.

Use of parameterization and subsequent analysis in both direct optimization and response surface optimization is demonstrated in the case of the impeller components (the rotationally periodic components) to maximize the design-point efficiency of the flow path. In both cases, a multi-objective genetic algorithm (MOGA) using the NSGA-II method is employed. In the direct case, a 3.9 percentage point improvement in total-total isentropic efficiency is obtained. In the indirect case a 4.7 percentage point improvement in total-total isentropic efficiency is obtained. It is noted that the indirect optimization used a three-layer ANN as the surrogate model (with only one hidden layer); and, while the absolute accuracy of the results of this model were poor when comparing the direct CFD-verified candidate points to the model predictions, the use of the surrogate model effectively optimized the fluid flow to maximize isentropic efficiency.

Optimization of the flow path of the volute is likewise demonstrated. In the case of the volute, a Kriging response-surface model was used in an indirect optimization. A 1 percentage point improvement in the overall flow path is shown for the case of the volute, incorporating both the upstream impeller (which is fixed in geometry) and the volute. Therefore, it is concluded that a 5 percentage point overall improvement is readily achievable using the proposed parameterization and workflow in single stage centrifugal compressors. Further research in utilizing improved Artificial Intelligence (AI) machine learning techniques to generate more sophisticated surrogate models is proposed for improving this method.


A Numerical Approach to Centrifugal Compressor Stage Flow Path Design Synthesis and Optimization

Division / Sponsor:
International Gas Turbine Institute (IGTI)

Paper Proceedings: https://asmedigitalcollection.asme.org/GT/GT2020/volume/84058

Turbo Expo - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.