Deep learning based adaptive deformation of aerodynamic shape for ducted propellers

Ducted propellers are widely used in eVTOL. The leading edge shape of the duct plays an important role in the aerodynamic performance of the ducted propeller. In this work, an optimization framework based on deep learning and multi-island genetic algorithm is proposed, which can quickly obtain the optimal leading edge shape according to the current working condition. Firstly, a modified shape parameterization method realizes the accurate description of the duct profile, especially for the control of leading edge shape. Secondly, a surrogate model based on deep learning and numerical simulated dataset is established to quickly predict the aerodynamic performance of ducted propellers, which is used in the optimization framework. Finally, optimization tasks for hovering state and forward flights at different advance ratios are carried out and analyzed. The results show that the deep learning based surrogate model has high precision and efficiency. Compared with the original design, the performance of the ducted propeller with optimized leading edge shape is increased by 17.6% in hovering state, and by 13.2%, 16.7%, 16.2% at three forward flight states respectively. The proposed optimization framework will pave the way for the application of adaptive deformation technology on ducted propellers.