Propulsion prediction of morphing flapping wings using reduced order model from deep-learning

Morphing the wing shape properly could improve the flapping performance of micro air vehicles (MAVs) incredibly. To design the proper morphing wings, the unsteady flapping performance needs to be estimated and predicted accurately. Compared with traditional computational fluid dynamics (CFD) simulation, reduced order model (ROM) from machine learning presents a more rapid and accurate way for flapping performance prediction, which is the key to real-time prediction in flight. In this study, a morphing flapping wing is studied at low Reynolds number (Re = 1000). The ROM of propulsive performance in forward flight is established from deep learning. The source datasets are generated from CFD simulation, with input about flapping/morphing motion and output about propulsion performances. The relationship between flapping/morphing motions and propulsion is extracted based on bidirectional long short-term memory networks (Bi-LSTM). The results from ROM are compared with CFD simulation, which indicate that the ROM predicts the flapping propulsion performance accurately in a wide range of flapping and morphing parameters. The overall errors of the ROM prediction from CFD data for thrust coefficient CT and consumed the power coefficient CP are 6.14% and 2.72%, respectively. The ROM has strong generalization performance too, which can be applied to rigid wings or morphing wings, either in the time domain or in the frequency domain. At the same time, the computational time of the ROM prediction is only 8.65% of CFD simulation time, which enormously improves computational efficiency. The ROM can be applied for optimization design and real-time control of high-performance flapping wings with morphing capabilities.