Expanding the Continuous Generation of a Family of High-Mach Number Centrifugal Compressor Stages

In conventional centrifugal compressor design, families of stages are typically defined within a domain spanned by swallowing capacity (flow coefficient, Phi) and peripheral Mach number (Mu2), which represents the non-dimensional impeller speed. However, this parameterization loses effectiveness when the flow regime transitions from subsonic to transonic. Under these conditions, the relative Mach number at the impeller eye (M1r) emerges as the most representative non-dimensional parameter for defining the operating point within the design space.

In a previous work by the authors, a continuous generation of impeller geometries was achieved using feature importance analysis. In that study, the interpolation was performed in one dimension only. The variation was carried out along the flow coefficient Phi, while M1r was kept constant.

The present study extends this methodology. The interpolation is now carried out in two dimensions, simultaneously along Phi and M1r. This allows the design space to be explored in a broader and more representative way. The example presented in this work switches between high-subsonic and transonic regimes, while maintaining flow similarity between machines at the same M1r.

Furthermore, the same data-driven framework is applied to automatically generate not only the impellers, but also the static components of the whole stage (i.e. the diffuser and the return channel vane geometries).

The proposed approach enables a consistent and continuous design methodology for complete centrifugal compressor stages, maintaining physical coherence across different flow regimes and offering a robust tool for tailoring machine performance to specific operating requirements.