OPTIMIZATION OF UNSTEADY CAVITATION PERFORMANCE OF AIRFOIL BASED ON MACHINE LEARNING

In order to improve the anti-cavitation performance of the airfoil under unsteady conditions, a modified leading-edge type function transformation method is employed to characterize the airfoil geometry. The optimal Latin hypercube sampling method was used to sample in the optimized space, and the cavitation performance parameters of each airfoil under unsteady state were calculated by CFD. The mapping relationship between airfoil geometry and cavitation performance parameters was constructed by BP neural network, and the time-averaged non-dimensionless cavitation area was used as the optimization objective. The cavitation performance of two-dimensional NACA66 (MOD) under the condition of σ = 0.83 was optimized by using Genetic Algorithm (GA). The unsteady cavitation flow field of the original airfoil and the two airfoils optimized by different neural network structures are calculated and analyzed. Finally, the multi-condition adaptation study of the 3D optimized hydrofoil is carried out, and the optimization results show that: Double-layer BP neural network has deeper data mining ability than single-layer BP neural network. Under the condition of σ = 0.83, the time-averaged non-dimensionless cavitation area of the two optimized airfoils is reduced by 12.8% and 19.2%, the lift-drag ratio is increased by 1.4% and 5.0%, and the caving cycle is reduced by 3.7% and 7.4%, respectively. The anti-cavitation and energy performance are improved. The modification of airfoil geometry affects the distribution of high-pressure regions and pressure gradients within the flow field, which not only suppresses the growth and development of cavitation but also enhances the intensity of the leading-edge backflow, leading to an increased frequency of cavitation shedding. The utilization of a two-layer BP neural network combined with genetic algorithm optimization results in a more effective hydrofoil design. Under the conditions of σ = 1.29 and σ = 1.44, the non-dimensional cavitation volume is reduced by 14.7% and 55.0%, respectively, while the lift-to-drag ratio is increased by 16.5% and 34.2%, respectively. This optimized design can better adapt to various cavitation conditions.