Reinforcement learning for gust load control of an elastic wing via camber morphing at arbitrary sinusoidal gusts

This paper presents the process of training and validation of deep reinforcement learning (RL) for gust load alleviation (GLA) based on camber morphing. First, a simplified aeroelastic model of the morphing wing considering the gust input is established based on the doublet-lattice method. The simplified model is adopted for GLA control training, and the proximal policy optimization algorithm is employed. When training is completed, the RL-based GLA controller is evaluated in the high-fidelity fluid-structure interaction environment. Results show that the controller effectively suppresses both the structural load and the aerodynamic force. It alleviates wingtip acceleration by 77.4 % at A_g=1m/s, f_g=2 Hz. The flow field suggests that the wing morphing counteracts the pressure distribution change induced by gusts and suppresses the lift fluctuations, thus alleviating the wingtip acceleration. Then, the RL-based GLA controller is tested at sinusoidal gust frequencies 1.5–3 Hz and amplitudes 1–4m/s, demonstrating a respectable alleviation effect within the maximum morphing angle. Finally, the comparison with the traditional PI controller further proves its superior gust alleviation performance.