In conventional airfoil optimization design method, the aerodynamic performance of the optimal airfoil may deteriorate at the non-design point, so it is necessary to develop the airfoil robust optimization method. This paper develops an airfoil robustness design strategy based on convolutional neural network and polynomial chaos method, which aims to improve the aerodynamic characteristics of the airfoil while improving the aerodynamic stability of the airfoil. This paper first proposes an aerodynamic force prediction method based on a convolutional neural network model; secondly, introduces the polynomial chaos method into the uncertainty quantification link in robust optimization; and then constructs a new method for airfoil robust aerodynamic optimization design. Finally, an optimization design test was carried out on the aerodynamic optimization design problem of the RAE2822 airfoil. The optimization results show that the robustness design strategy of the airfoil proposed in this paper is feasible. After optimization, the aerodynamic characteristics of the airfoil are improved in a wide range of design, achieving the purpose of robust optimization.