Recent advances in machine learning based optimization of porous propeller aerodynamics for unmanned aerial vehicle systems

This review investigates how the synergistic integration of porous material science and machine learning can be combined for designing and optimising next-generation UAV propellers. UAV propulsion must be disrupted with solutions that address critical constraints in endurance and noise. Porous propellers offer a promising direction, providing aerodynamic advantages of less tip-vortex drag, broadband noise damping, and better stall resistance. Nonetheless, porosity can be added to produce a huge, computationally infeasible design space. This review synthesises that ML is an enabler of transformation, explaining the underlying principles to specific applications. Examining optimisation for drastic cost reduction with surrogate assistance, deep learning in the case of inverse design, and instantaneous prediction of the flow-field and physics-informed neural networks in cases of data scarcity. The review highlights the performance improvements, major trade-offs, and significant limitations, including generalisation of the model, gaps in manufacturing feasibility, and a lack of experimental validation. The review concludes by proposing a research agenda for the future that serves as the combination of AI-driven digital twins, end-to-end generative design architecture, and the incorporation of manufacturing constraints in the optimisation loop. Reinforcement learning for in-flight adaptive control of porous flow can be investigated in future work. Thus, the convergence of porous aerodynamics and intelligent computational design is a paradigm shift, and will unlock a new category of high-efficiency, quiet and adaptive UAV propulsion systems.