Physics-informed dimensionality reduction for propeller shape optimization

The design of marine propellers is challenged by high-dimensional parameter spaces and the need to balance efficiency with cavitation avoidance. Dimensionality reduction techniques offer a cost-effective way to address the curse of dimensionality, but geometry-based approaches such as parametric model embedding (PME) may neglect local features with strong hydrodynamic relevance. This work introduces the application of physics-informed PME (PI-PME) to propeller shape optimization, where physical observables, including pressure distributions and performance indicators, are embedded into the reduced-order space. A multi-objective optimization framework, based on boundary element method analyses and validated with Reynolds-averaged Navier–Stokes simulations, is applied to a cruise-ship propeller. Comparisons between original, PME-reduced, and PI-PME-reduced design spaces demonstrate that PI-PME preserves critical sectional features and significantly improves optimization results. The results highlight the benefits of integrating physical information into dimensionality reduction, enabling reliable, efficient, and physics-aware design optimization of marine propellers.