Flow state can be changed by multiple disturbances and uncertain factors in a complex flow environment, which calls for great interest to adjust the control law automatically to adapt to the changing flow environment. Model-based control can obtain predetermined control effects, but its adaptive ability is limited due to the modeling accuracy and unmodeled dynamics of the reduced-order model. To overcome these limitations, the data-driven adaptive control of transonic buffet flow based on the radial basis function neural network (RBF-NN) is carried out in this work. The actuator is the trailing edge flap, and the feedback signal is the lift coefficient. The historical input and output are used in the RBF-NN adaptive control to calculate the current control input from the neural network. When the flow state changes, the parameters of the neural network are adjusted by an adaptive mechanism to make the system work in an optimal or a near-optimal state automatically. Results show that buffet loads can be suppressed completely by RBF-NN control, even if the freestream Mach number and the angle of attack change continuously [from (M, α) = (0.7, 5.5°) to (M, α) = (0.8, 1.5°)]. The control strategy proposed in this work only needs the historical response data of the flow field, and it shows little dependence on the low-order linear model of the system. Therefore, it can be applied to the unstable flow control, in which the low-order model of the flow is difficult to construct and automatically adapt to the changing flow environment.

Transonic flight has high economic benefits, but the appearance of transonic buffet limits the flight envelope. The shock control bump currently used for transonic buffet suppression tends to degrade the aerodynamic performance of the non-buffeting state. In this study, a smart skin system is used to eliminate the fluctuating load of transonic buffet by measuring the airfoil lift coefficient as the feedback signal and adjusting the local skin height using data-driven, model-free adaptive control. Since the actuator height is dynamically adjusted only after the occurrence of transonic buffet, the smart skin can completely suppress the fluctuating load and does not affect the aerodynamic performance in the non-buffeting state. The suppression effect of the proposed smart skin on transonic buffet is verified by numerical simulation of the flow. The simulation results show that due to the introduction of closed-loop control, the fluctuating load of transonic buffet can be effectively suppressed for different positions and maximum heights of the actuator. Even when the flow state changes, the robust smart skin system can also achieve the control goal. Therefore, smart skins combining flexible materials and control technologies have the potential to effectively improve the aerodynamic performance of aircraft.

Transonic buffet is a phenomenon of large self-excited shock oscillations caused by shock wave-boundary layer interaction, which is one of the common flow instability problems in aeronautical engineering. This phenomenon involves unsteady flow, which makes optimal design more difficult. In this paper, aerodynamic shape optimization design is combined with reinforcement learning to address the problem of transonic buffet. Using the deep deterministic policy gradient (DDPG) algorithm, a reinforcement learning-based design framework for airfoil shape optimization was constructed to achieve effective suppression of transonic buffet. The aerodynamic characteristics of the airfoil were calculated by the computational fluid dynamics (CFD) method. After optimization, the buffet onset angles of attack of the airfoils NACA0012 and RAE2822 were improved by 2° and 1.2° respectively, and the lift-drag ratios improved by 83.5% and 30% respectively. Summarizing and verifying the optimization results, three general conclusions can be drawn to improve the buffet performance: (1) narrowing of the leading edge of the airfoil; (2) situating the maximum thickness position at approximately 0.4 times the chord length; (3) increasing the thickness of the trailing edge within a certain range. This paper established a reinforcement learning-based unsteady optimal design method that enables the optimization of unsteady problems, including buffet.