Parametric modeling and optimization of rim-driven propulsor blades: insights into computational fluid dynamics and metamodeling

• Demonstrates the strong potential of rim-driven propulsors (RDPs) for sustainable and electrified propulsion, with applicability extending beyond aviation to marine and automotive systems. • Establishes a comprehensive CFD-based parametric optimization framework, identifying key geometric and aerodynamic parameters governing thrust, torque, and propulsive efficiency. • Achieves high predictive fidelity (Coefficient of Prognosis > 96%) using the metamodel of optimal prognosis (MOP) for thrust, torque, and efficiency across a wide design space. • Integrates multi-objective optimization via the SPEA2 evolutionary algorithm with TOPSIS-based decision making to robustly rank and select Pareto-optimal RDP designs. • Introduces a neural network surrogate model that complements the metamodel, offering improved robustness in extrapolation regions and accelerating design-space exploration with high accuracy. Rim-driven propulsors (RDPs) are promising candidates for next-generation electrified propulsion systems because of their high efficiency, compact architecture, and reduced mechanical complexity. This study presents an integrated framework for the aerodynamic analysis and multiobjective optimization of RDP blade geometries by coupling high-fidelity computational fluid dynamics (CFD) with surrogate modeling and neural network-based optimization. A parametric design space is explored, including the chord distribution, number of blades, hub ratio, taper ratio, pitch angle, and key airfoil parameters such as thickness and camber characteristics. Kriging-based metamodels were first developed to accelerate design space exploration, achieving coefficients of prognosis (CoP) exceeding 96% for thrust, torque, and propulsive efficiency. Multiobjective optimization is conducted using the strength Pareto evolutionary algorithm 2 (SPEA2), with Pareto-optimal solutions ranked via the technique for order preference by similarity to the ideal solution method under equal objective weighting. To improve the predictive accuracy and robustness—particularly near design space boundaries—a complementary neural network surrogate is introduced. The neural network demonstrates low prediction error and strong agreement with the CFD results, outperforming the metamodel in terms of the regression evaluation metrics and Pareto-front similarity metrics. The optimized RDP designs achieve significant improvements in thrust and propulsive efficiency relative to the baseline configuration, with only moderate increases in torque demand. Flow-field analyses confirm that these gains stem from enhanced pressure differentials and more uniform blade loading. The proposed CFD-surrogate-neural network framework provides a robust and efficient methodology for RDP optimization and is readily applicable to electrified aviation, marine propulsion, and automotive applications.