Design-Space Dimensionality Reduction Benchmark Dataset - Ship Propeller

This dataset provides a benchmark database for the evaluation and comparison of design-space dimensionality reduction methods in hydrodynamic shape optimization problems. The dataset consists of 16,385 configurations of a six-blade, right-handed marine propeller for a cruise ship, generated through Sobol sampling of a B-Spline-based parametric model describing radial and chordwise blade features. Each configuration is described by 74 propeller parametrization parameters. Among them, 38 parameters are active design variables, while the remaining parameters are fixed and retained to preserve the complete propeller parametrization structure. The skew distribution is kept unchanged with respect to the reference propeller. The propeller blade geometry is discretized using 1,326 grid nodes, stored as flattened coordinate vectors. Each configuration includes: discretized propeller blade geometry coordinates (1,326 × x, y, z points) propeller parametrization parameters defining the B-Spline model pressure coefficient distribution (Cp) at the nominal equivalent condition, evaluated at 1,250 panel centers pressure coefficient distribution (Cp) at the loaded equivalent condition, evaluated at 1,250 panel centers pressure coefficient distribution (Cp) at the unloaded equivalent condition, evaluated at 1,250 panel centers integrated hydrodynamic quantities at the nominal condition: thrust coefficient (KT) efficiency (eta) tip-vortex intensity (Gamma) Hydrodynamic quantities are computed using a low-fidelity Boundary Element Method (BEM) solver. Simulation conditions: nominal equivalent condition loaded equivalent condition, representative of the blade passing through the 90° position in the non-uniform wake unloaded equivalent condition, representative of the blade passing through the 270° position in the non-uniform wake The dataset matrix has size 7,805 × 16,385 and includes: geometry coordinates propeller parametrization parameters distributed pressure coefficients integrated hydrodynamic quantities Dataset statistics: total configurations: 16385 baseline configuration: 1 sampled configurations: 16384 valid simulations: 16306 invalid simulations: 79 Invalid simulations correspond to cases where the BEM simulation or the post-processing of hydrodynamic quantities did not provide valid physical outputs. NaN values are also used when non-physical negative values occur in scalar quantities, including efficiency, thrust coefficient, or tip-vortex intensity. Samples with invalid or partially invalid physical quantities may still be useful for geometry-based dimensionality reduction studies. This dataset is part of a benchmark collection aimed at providing standardized databases for evaluating dimensionality reduction techniques in engineering design spaces. The dataset follows a physics-informed representation of the design space, where geometric, parametric, and physical information are combined consistently within a unified framework. The parametrization and solver used are described in: Serani, A., Palma, G., Wackers, J., Quagliarella, D., Gaggero, S., & Diez, M. (2025). Extending parametric model embedding with physical information for design-space dimensionality reduction in shape optimization. Engineering with Computers. Gaggero, S., & Serani, A. (2026). Physics-informed dimensionality reduction for propeller shape optimization. Applied Ocean Research, 166, 104932.