基于神经网络的减阻沟槽壁面形状优化

Sub⁃optimization of riblet shape based on neural networks

针对沟槽外形减阻问题,采用基于神经网络的方法对沟槽壁面形状进行外形优化。模型采用槽道流动模型,控制方程为黏性不可压缩Navier⁃Stokes(NS)方程,流动求解采用直接数值模拟(DNS)方法,对于对流项的离散采用紧致4阶中心格式,对黏性项的离散采用4阶中心格式,时间推进采用3阶Runge⁃Kutta格式。在神经网络优化过程中,约束方程为不可压NS方程,采用基于在线学习的自适应控制器,使用基于抑制展向切应力的控制律,控制量的产生由壁面变形提供。优化结果表明,壁面最大减阻效果可达17.41%。对于优化后的壁面,湍流强度降低了19.68%,同时壁面的涡量与雷诺切应力亦有所降低。由于湍流流动非定常,因此优化得到的壁面形状亦是时变的,但变化的过程中整体上仍呈现流向沟槽的形状。

For the riblet drag reduction,a neural network⁃based method was used to opti⁃mize the shape of the riblet surface.This work employed the channel flow model and the govern⁃ing equations were the viscous incompressible Navier⁃Stokes(NS)equations.The turbulent flow inside the channel was resolved by the direct numerical simulation(DNS)method.According to the numerical method,a compact fourth⁃order central scheme was used for the discretization of the convective term,a fourth⁃order central scheme was applied to the discretization of the viscous term,and a third⁃order Runge⁃Kutta scheme was employed for the time advancement.In the neu⁃ral network sub⁃optimization process,the constraint equation was the incompressible NS equa⁃tion,and an adaptive controller based on online learning was used.In the optimization procedure,the control law was based on weakening the spanwise shear stress,and the control quantities were provided by the wall deformation.The optimization results demonstrated that the maximum wall drag reduction reached 17.41%.For the wall optimization,the turbulence intensity was re⁃duced by 19.68%,besides,the vorticity and Reynolds shear stress at the wall also declined.Since the turbulent flow was unsteady,the shape of the optimized wall also varied with time,but the overall shape of the riblets still illustrated a streamwise riblet⁃like configuration.