Time-history performance optimization of flapping wing motion using a deep learning based prediction model

Abstract

Flapping Wing Micro Aerial Vehicles (FWMAVs) have caused great concern in various fields because of their high efficiency and maneuverability. Flapping wing motion is a very important factor that affects the performance of the aircraft, and previous works have always focused on the time-averaged performance optimization. However, the time-history performance is equally important in the design of motion mechanism and flight control system. In this paper, a time-history performance optimization framework based on deep learning and multi-island genetic algorithm is presented, which is designed in order to obtain the optimal two-dimensional flapping wing motion. Firstly, the training dataset for deep learning neural network is constructed based on a validated computational fluid dynamics method. The aerodynamic surrogate model for flapping wing is obtained after the convergence of training. The surrogate model is tested and proved to be able to accurately and quickly predict the time-history curves of lift, thrust and moment. Secondly, the optimization framework is used to optimize the flapping wing motion in two specific cases, in which the optimized propulsive efficiencies have been improved by over 40% compared with the baselines. Thirdly, a dimensionless parameter Cvariation is proposed to describe the variation of the time-history characteristics, and it is found that Cvariation of lift varies significantly even under close time-averaged performances. Considering the importance of time-history performance in practical applications, the optimization that integrates the propulsion efficiency as well as Cvariation is carried out. The final optimal flapping wing motion balances good time-averaged and time-history performance.