1 March 2023 in Physics of Fluids
The traditional optimization of airfoils relies on, and is limited by, the knowledge and experience of the designer. As a method of intelligent decision-making, reinforcement learning can be used for such optimization through self-directed learning. In this paper, we use the lift–drag ratio as the objective of optimization to propose a method for the aerodynamic optimization of airfoils based on a combination of deep learning and reinforcement learning. A deep neural network (DNN) is first constructed as a surrogate model to quickly predict the lift–drag ratio of the airfoil, and a double deep Q-network (double DQN) algorithm is then designed based on deep reinforcement learning to train the optimization policy. During the training phase, the agent uses geometric parameters of the airfoil to represent its state, adopts a stochastic policy to generate optimization experience, and uses a deterministic policy to modify the geometry of the airfoil. The DNN calculates changes in the lift–drag ratio of the airfoil as a reward, and the environment constantly feeds the states, actions, and rewards back to the agent, which dynamically updates the policy to retain positive optimization experience. The results of simulations show that the double DQN can learn the general policy for optimizing the airfoil to improve its lift–drag ratio to 71.46%. The optimization policy can be generalized to a variety of computational conditions. Therefore, the proposed method can rapidly predict the aerodynamic parameters of the airfoil and autonomously learn the optimization policy to render the entire process intelligent.
1 May 2024 in Physics of Fluids
Gradient-based aerodynamic optimization methods involve complex gradient calculations with high computational costs. To address these problems, this paper proposes a rapid deep learning-based method for predicting the aerodynamic force gradients and establishes an aerodynamic optimization framework. A deep neural network is used to determine the latent mapping relationship between the near-wall flow field information and the aerodynamic force gradients, and the resulting gradient prediction model is integrated into an aerodynamic optimization platform. Combining grid deformation techniques with gradient-based optimization enables the aerodynamic optimization of airfoils. The deep neural network model can rapidly predict gradients that are basically consistent with those from the traditional discrete adjoint method, and decouples the gradient computation from the flow field solution process. The proposed method is validated through the aerodynamic optimization of the airfoil. Under various optimization objectives and initial conditions, the deep neural network model significantly improves the aerodynamic performance of the airfoil and substantially enhances the efficiency of optimization compared with traditional methods.