Research on internal flow mechanism and energy loss characteristics of centrifugal pump based on data regression algorithm

• An adaptive SST-CC turbulence model with SVR-based curvature correction is developed and implemented via UDF. • The proposed model enhances prediction of vortex structures, entropy generation, and flow separation in centrifugal pumps. • Numerical predictions achieve high accuracy with minimum relative errors of 1.02% (head), 1.33% (power), and 1.82% (efficiency). • The SST-CC model reveals multi-scale entropy production gradients and redistributes energy losses among pump components more realistically. • This approach provides a reliable and computationally efficient tool for analyzing energy dissipation and optimizing pump design. Standard eddy-viscosity models often constrain the accuracy of Reynolds-averaged Navier-Stokes (RANS) simulations for centrifugal pumps by overlooking streamline curvature and rotation effects. In this study, we implement a Spalart-Shur rotation/curvature correction within the SST k-ω framework (SST-CC) and develop an adaptive curvature-correction strategy. The model’s curvature coefficient is dynamically optimized using a support-vector regression (SVR) algorithm and deployed as a user-defined function (UDF) in the solver. This approach is validated on a single-stage centrifugal pump under design conditions using a rigorously converged mesh and laboratory measurements. Compared to the SST and a constant-coefficient modification (SST-MS), SST-CC enhances the prediction of vortex core distribution, peak vorticity, and local entropy production, resulting in substantially better agreement with experimental data. Validation shows minimum relative errors of 1.02% in head, 1.33% in shaft power, and 1.82% in efficiency for the SST-CC solution. The model also reveals clearer multi-scale gradients of local entropy production and a more physically consistent redistribution of momentum and viscous losses among the impeller, volute, and chamber domains