Answers are grounded in this paper and cite their sources.
Centrifugal pump development still moves through separate islands of information: a one-dimensional sizing calculation, a manually built solid model, a physical prototype, and a drawing package produced after the geometry is frozen.This paper proposes and tests a design technology in which parametric 3D geometry, multi-fidelity CFD, a reduced-order digital twin, additively manufactured prototypes and machine-readable documentation are all views of one versioned design record.Thirty-four continuous variables plus one discrete blade number Z (35 optimisation variables in total) control the meridional channel and the blade loading, with manufacturability enforced inside the generator; a co-Kriging surrogate coupling coarse steady and fine transient cavitating simulations drives an NSGA-III search; a POD-Galerkin model corrected by an unscented Kalman filter forms the operational twin; and the sensitivity matrix produced by the optimiser is reused to allocate manufacturing tolerances.The framework was applied to three pumps of different duty (Case A, nq = 37.0; Case B, nq = 12.7; Case C, nq = 32.3;recalculated from Eq. ( 1) with Q in m³ s⁻¹) and validated against ISO 9906 Grade 1B tests.Three results are central.First, the two-level surrogate reached the same Pareto front quality as single-fidelity Kriging with 4.2 times fewer high-fidelity simulations, and measured best-point efficiency rose by 2.9-4.7 percentage points.Second, an equivalent-sand-roughness correction brought as-built selective laser melted prototypes to within 0.9 percentage points of cast production efficiency, against an uncorrected deficit of about 5 points, which turns a printed impeller into a quantitative performance predictor rather than a form-and-fit check.Third, in a two-unit field study over 4300 h the twin tracked head to 1.2 % mean absolute error and estimated wear-ring clearance to within 0.03 mm of the post-overhaul measurement.Design-cycle and documentation metrics from 14 releases are reported as associations rather than causal effects, since the comparison was not randomised.
No takes yet. Share an insight, caveat, or question.
Tuleshov et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: