AERODYNAMIC OPTIMIZATION OF HIGH-ALTITUDE PROPELLER COMBINED WITH MACHINE LEARNING METHOD

This paper presents an optimization design for a high-altitude propeller. The machine learning method are coupled into the propeller analysis and optimization workflow to improve the calculation accuracy and efficiency. The selected E387-based propeller is from a HALE UAV, which designed to cruise at 25.9km altitude, 0.4 Mach. The propeller aerodynamic analysis is conducted by ARI_PROP, an in-house code based on Standard Strip Theory. To better solve the problem of prediction inaccuracy and complexity at high altitudes (high subsonic and low Reynolds number condition), the code employs Gaussian Progress Regression (GPR) to generate airfoil section aerodynamic performance. The GPR model can be trained by small amount of RANS CFD or experimental samples, and output accurate predictions at desired state instantly. Compared with literature and Moving Reference Frame (MRF) CFD results, ARI_PROP code shows good accuracy and automation capability. Next, the ARI_PROP code is coupled with our ARI_XunZhu optimization software, which is advantageous in high-efficiency and robustness. The software employs the Surrogate-based optimization scheme, which uses models like Kriging to establish the surrogate approximation model and guide the searching for better designs. By using the proposed overall workflow, the radial distribution of chord and pitch angle of referenced propeller that can offer maximum thrust at design point is found. The result shows that the optimized propeller’s thrust increases 75.2% with efficiency set as 81%, compared to the original design.