Unstable internal flow structures significantly compromise the mechanical reliability of centrifugal pumps. This study elucidates the coupling mechanism between impeller geometry, vortex dynamics, and hydrodynamic forces. A hybrid optimization framework, integrating a neural-network surrogate model with both multi-objective genetic algorithm and non-dominated sorting genetic algorithm III (NSGA-III), was employed to refine the impeller. Comparative analysis reveals that the NSGA-III design offers superior stability, achieving a 31.1% reduction in peak radial force and eliminating axial-force directional reversals, while simultaneously increasing efficiency by 5.4% and reducing shaft power by 6.3%. Crucially, the stabilization mechanism is decoded using multiresolution dynamic mode decomposition (MRDMD) and entropy generation analysis. MRDMD results demonstrate that the optimized geometry attenuates low-frequency unstable modes driven by rotor–stator interaction, shifting spectral energy to stable high-frequency structures. Furthermore, thermodynamic analysis identifies a fundamental shift in dissipation pathways: the suppression of large-scale coherent vortex shedding reduces internal volumetric entropy generation by 63.6%, yielding a flow field stabilized by controlled wall shear layers. These findings provide a physics-based rationale for mitigating flow instabilities through targeted geometric refinement.
Zhou et al. (2026) studied this question.