Fast boundary layer suction optimization with neural networks

Abstract

Abstract

Previous experiments on a flat plate with four suction panels in a wind tunnel have shown that it is possible to optimize the distribution of suction so as to minimize the suction energy required to maintain laminar-turbulent transition at a fixed position. The location of the transition can be detected by surface-mounted microphones. This can form the basis of a ‘smart suction’ system which can adaptively minimize energy consumption in flight. The gradient descent optimization scheme previously employed finite difference estimates of the gradient of transition position as a function of the suction velocity at each of the panels. This is slow if the number of suction panels is large. It is shown that the optimization can be made considerably faster if the suction-transition function is pre-identified using a radial basis function neural network. Because of the robustness of the optimization algorithm to the gradient estimate, the number of measurements needed to pre-identify this function is surprisingly small, in the present case 3 ^N , where N is the number of suction panels.