Gust load alleviation for a flexible wing with camber morphing via deep reinforcement learning

Gust load alleviation (GLA), aiming to suppress the gust response of the aircraft, is important for flight safety and quality. The aeroelastic system presents additional complexities due to the coupling effects between aerodynamics, structure, and control. The gust response of a flexible morphing wing exhibits nonlinear characteristics with different gust amplitudes and frequencies. This nonlinearity poses challenges in terms of enhancing the generalizability of GLA control strategies. Compared to traditional approaches, deep learning is suitable for aeroelastic systems due to its intricate nonlinear fitting and strong ability to capture the characteristics of data rather than model. In this work, a GLA control strategy based on deep reinforcement learning is proposed for a flexible wing with a seamless morphing trailing edge (TE). The aeroelastic model is established based on the doublet-lattice method considering wing morphing. Then the interaction between the aeroelastic model and the GLA system is framed as a Markov decision process (MDP). The gust responses at the current and past four time steps are taken as the state input. The incremental discrete TE morphing is adopted as the action output. The proximal policy optimization (PPO) strategy is employed for the optimization of the GLA policy. When the training of the GLA strategy is completed, its control performance is tested at various gust amplitudes and frequencies. The GLA strategy shows significant gust response suppression effects under different gusts. The peak value of the wingtip acceleration is alleviated by at least 50%.