Jet centrifugal pumps frequently suffer from low energy conversion efficiency due to severe fluid mixing and inter-stage interference. To address this, a structural optimization framework combining the continuous adjoint method and machine learning is proposed. Following numerical validation, continuous adjoint equations were employed to evaluate shape sensitivity, pinpointing the front ejector as the core region dictating energy conversion. Pearson correlation analysis subsequently extracted four critical variables: nozzle diameter, throat diameter, suction chamber length, and diffuser angle. Through a systematic evaluation of four predictive algorithms, Response Surface Methodology (RSM) demonstrated superior accuracy and was adopted as the surrogate model. The non-dominated sorting genetic algorithm (NSGA-II) was then applied to maximize pump head and efficiency. Results indicate that the optimized configuration achieves a 5.49% increase in head and a 6.17% efficiency improvement under design conditions, maintaining enhanced hydraulic performance across the entire flow range. Internal flow analyses reveal that the optimized geometry effectively mitigates aggressive momentum exchange and suppresses local flow separation and reverse vortices, thereby reducing hydraulic impact and shear friction dissipation. Ultimately, this study provides a robust methodological framework for optimizing fluid machinery equipped with complex pre-jetting components.
Mao et al. (2026) studied this question.