A Two-Step Approach for Centrifugal Compressor Performance Mapping Based on a Reduced-Order Model and an Evolutionary Algorithm

TitleA Two-Step Approach for Centrifugal Compressor Performance Mapping Based on a Reduced-Order Model and an Evolutionary Algorithm
Publication TypeConference Paper
Year of Publication2023
AuthorsBicchi M, Biliotti D, Toni L, Marconcini M, Grimaldi A, Arnone A
Conference NameASME Turbo Expo 2023 Turbomachinery Technical Conference and Exposition

Given a good efficiency, robustness, and wide operating range, centrifugal compressors have had a relevant impact in the last decades, and still today, they are pivotal in several industrial and civil applications. Indeed, whether focusing on different energy vectors or sources, on capture technologies, or even on the increase of system efficiency, centrifugal compressors seem to be crucial for the current energy transition, playing a key role in achieving net-zero carbon growth by 2050.

Against this backdrop, predicting the performance of a centrifugal compressor stage is essential for its aerodynamic design and its selection in a customer order. To this end, the scientific literature has recently shown several methods, involving pure data-driven models, regression techniques, machine learning algorithms, or statistics-based approaches. Most of these focus on the preliminary design of a centrifugal compressor and aim to provide an approach able to generate performance maps with limited prior knowledge of the compressor stage. Instead, once the final geometry of the stage is achieved, experiments and computational fluid dynamics (CFD) simulations are reliable methods for performance map predictions. However, these last methods are more expensive and time-consuming, in contrast to typical limitations of the industrial scenario. Therefore, attention must be paid when managing data from these techniques in order to reduce their impact on cost and time. For this purpose, in the present research a special focus is laid on the performance map creation once the aerodynamic design is close to the end. Specifically, the present paper provides a reduced-order physique-based approach leveraging the use of CFD analyses and an evolutionary algorithm. The main goal of the approach is focused on reducing the computational time and effort of CFD-based performance mapping while maintaining the accuracy of high-fidelity calculations.

Specifically, the main contribution of this article is twofold. At a theoretical level, the proposed approach expands the current scientific literature regarding performance mapping. While on a practical level, this approach represents a business-friendly tool for creating detailed performance maps once the aerodynamic design is close to the end. As an example case study, the developed approach is applied on a group of three centrifugal compressor stages for high-head low-flow applications, composed by splitter-bladed impellers, vaned diffusers, and return channel systems. Moreover, a comparison with experimental measurements from two stages of the aforementioned group is provided. Results show a good agreement between maps obtained with the proposed approach and those generated with CFD and experiment data.

In conclusion, this research lays the foundations for deriving new future studies that aim to further reduce the computational cost and time of a step in aerodynamic design of centrifugal compressors whose impact on time-to-market is not negligible.



Refereed DesignationRefereed