基于一种贝叶斯优化框架的高空螺旋桨气动优化设计

Aerodynamic design of high-altitude propellers within a Bayesian optimization framework

为了在可接受时长内获得高空螺旋桨气动方案的最优解,提出一种基于贝叶斯优化框架的高空螺旋桨气动外形优化设计方法。该方法以拉丁超立方抽样获取螺旋桨气动外形参数的初始样本点,建立以该参数为输入、数值模拟获取螺旋桨气动性能为输出的初代高斯过程模型;以遗传算法和三种并行加点准则构成的子优化获取新样本点,求取新样本点的气动性能,并更新样本数据和高斯过程模型。该方法可以使新样本点快速向最优解附近集中,从而提高最优解附近的模型近似精度。为验证模拟方法,加工了某高空螺旋桨进行地面试验。与模拟结果相比,试验推力平均误差2.34%、扭矩平均误差3.33%。以课题组自研的6个低雷诺数翼型为基础,使用该方法对某高空太阳能无人机螺旋桨进行优化设计,优化结果显示:螺旋桨在设计点推力提高9.24%、效率提高8.13%。研究结果表明,该方法对高空螺旋桨优化设计及相关工程应用具有较强的参考价值。

To obtain optimal aerodynamic configurations of high-altitude propellers efficiently,we propose a method for the aerodynamic shape design of high-altitude propellers within the Bayesian optimization framework.This method parameterizes propellers'shapes by eight variables using the quadratic functions.Initial samples come from the Latin hypercube sampling.The corresponding aerodynamic performance is obtained by Computational Fluid Dynamics(CFD)simulations,which have been validated by ground tests.A Gaussian process is established between the shape parameters and the aerodynamic performance.New samples are obtained by the sub-optimization composed of the genetic algorithm and the infill sampling criterion.The infill sampling criterion enables generating new samples near the locally and globally optimal solutions to improve the approximation accuracy near the optimal solution without considering the prediction accuracy in the whole design space;the parallel strategy can enhance its efficiency.Samples and the Gaussian process are adaptively updated.We apply this method to optimize the propeller of a high-altitude long-endurance solar-powered UAV based on six low-Reynolds-number airfoils developed by our group.This method shows great potential in optimizing high-altitude propellers and,hopefully,other applications since the thrust and efficiency of the optimized propeller increase by roughly 10%.