Abstract
The design of centrifugal pumps remains an iterative, expertise-intensive process in which hydraulic optimization, three-dimensional geometry definition, physical validation, and the release of manufacturing documentation are weakly coupled, producing long lead times and a high number of physical prototype cycles. This paper proposes an integrated design technology that fuses one-dimensional hydraulic synthesis, parametric 3D modeling, a hybrid digital twin, and additive 3D prototyping into a single closed-loop workflow that also auto-generates engineering documentation. The novelty lies in a hybrid twin that augments a physics-based hydraulic model with a Gaussian-process discrepancy term trained on high-fidelity Reynolds-Averaged Navier–Stokes (RANS) simulations, enabling accurate, sub-second performance prediction across the design space and direct multi-objective optimization of efficiency and cavitation margin. The methodology was validated on three pumps of distinct specific speed and application: municipal water supply, oil-and-gas process duty, and HVAC circulation. Across the three units the surrogate twin reproduced CFD efficiency with a root-mean-square error of 0.61% (R² = 0.984), while the calibrated twin matched rig-tested head and efficiency within 2.4% and 1.9%, respectively. Relative to the conventional workflow, design lead time was reduced by 53–58% and physical prototype iterations by 60–75%. The results indicate that coupling hybrid digital twins with parametric CAD and additive prototyping yields a transferable, application-agnostic pump-design technology.