A data-driven flight optimization of two-segment flapping-wing aircraft based on deep learning

• A CFD-based unsteady flow simulation framework is developed for flapping-wing aircraft, which successfully captures complex vortex dynamics and coupled aerodynamic effects under various flapping parameters. • A deep neural network model is established to rapidly predict aerodynamic performance from flight parameters, achieving relatively high prediction accuracy while significantly reducing computational time compared with conventional CFD simulations. • Comparative analysis of optimization algorithms demonstrates that the differential evolution method outperforms the genetic algorithm in identifying optimal flapping parameters, resulting a faster convergence in solving the multiobjective problem of balancing lift-weight equilibrium while maximizing thrust. Flapping-wing aircraft, as a typical example of bionic aviation vehicles, hold significant importance in fields such as environmental monitoring, reconnaissance, and aerobiology research. However, the design of flapping-wing aircrafts usually relies on the imitation of birds or insects in nature and requires extensive iterative optimization. In this study, a computational fluid dynamics (CFD) data-driven flight optimization approach is developed for flapping-wing aircraft design. First, we conduct a bionic aerodynamic analysis of the unsteady flow field during the flapping motions and multiple cases with different combinations of flight parameters are simulated via CFD to investigate the factors influencing the flight performance. The simulation data are subsequently used to establish a deep neural network (DNN) model to explore the relationship between the flapping motions and the flight performance. Finally, optimization algorithms are applied to search for the optimal flight parameter combinations under different weights, which can provide a valuable reference for the overall design and flight optimization of flapping-wing aircraft.