Dimensionality Reduction Optimization of High Subsonic Unmanned Aerial Vehicles Intake

To address the problem of poor aerodynamic performance in the S-shaped intake duct of a subsonic unmanned aerial vehicle (UAV), the study investigates how to optimize its design to achieve the highest possible total pressure recovery coefficient and minimize outlet total pressure distortion. Firstly, an S-shaped intake duct suitable for subsonic UAVs was designed using an isentropic relation and a superelliptical cross-section. Control points were placed on the duct to allow grid deformation, thereby altering the shapes of centerline and cross-section. To address the issue of excessive optimization dimensions caused by simultaneously altering both shapes, sensitivity analysis and principal component analysis (PCA) were applied, followed by dimensionality reduction using the Hicks-Henne function. A Kriging model was then used for fitting, and a genetic algorithm was employed to search for optimal solutions with total pressure recovery coefficient and steady-state outlet distortion index as the objectives. The results show that after dimensionality reduction through sensitivity analysis and PCA, the average error and coefficient of determination of the surrogate model were improved. Specifically, for outlet total pressure distortion, the Kriging model trained with sensitivity analysis data had an average error reduced by 0.060 47 compared to the training results without dimensionality reduction, while the Kriging model trained with PCA data had an average error reduced by 0.051 05. The optimized intake duct achieved outlet total pressure distortions of 0.129 9 and 0.132 8, with outlet total pressure recovery coefficients of 0.988 5 and 0.990 9. The data indicates that the dimensionality reduction method employed successfully reduced computational power consumption and improved the fitting accuracy of the surrogate model. Among them, sensitivity analysis is suitable for dimensionality reduction with a clear optimization objective, while PCA is more suitable for dimensionality reduction in multi-objective optimization.