Rapid airfoil design optimization via neural networks-based parameterization and surrogate modeling

Aerodynamic optimization based on computational fluid dynamics (CFD) is a powerful design approach because it significantly reduces the time compared with human manual design. However, CFD-based can still take hours to converge requires repeatedly running computationally expensive flow simulations. To further shorten time, we propose fast, interactive framework that allows us complete an airfoil aerodynamic within few seconds. This made efficient through B-spline-based generative adversarial network model for shape parameterization, which filters out unrealistic airfoils reduced space contains all relevant shapes. Moreover, use combination of multilayer perceptron, recurrent neural networks, and mixture experts surrogate modeling enable both scalar (drag lift) vector (pressure distribution) response predictions wide range Mach numbers (0.3 0.7) Reynolds (104 1010). verify our proposed framework, compare results ones computed by direct subsonic transonic conditions. The show optimal designs quantities (lift, drag, pressure obtained agree well optimizations evaluations. being integrated into web-based users predict lift, moment, distribution, shapes conditions