基于物理信息神经网络的船舶螺旋桨尾流场重构

Reconstruction of Ship Propeller Wake Field Based on Physics-Informed Neural Networks

将物理信息神经网络(PINN)应用于船舶螺旋桨尾流场的重建.介绍了PINN的原理和基本框架;应用PINN求解Burgers方程,对PINN求解偏微分方程的可行性进行验证.利用计算流体力学(CFD)软件STAR CCM+对KVLCC2螺旋桨的敞水特性进行了数值模拟,得到了该桨在敞水中运动的流场信息.基于数值模拟得到的敞水桨流场特性信息,构造PINN训练样本集对PINN进行训练;训练后的PINN用于推断控制方程在任意时间和空间坐标的近似解.将PINN得到的速度和压力分布与STAR CCM+模拟的速度和压力分布进行了比较,对比结果验证了PINN在尾流场重建中的可靠性.研究结果表明,PINN可以应用于船舶螺旋桨尾流场的重建.

Physics-informed neural networks(PINN)are applied to the reconstruction of the ship propeller wake field.First,the principle and basic framework of PINN were introduced.Then,the Burgers equation was selected to verify the feasibility of PINN in solving partial differential equations.After that,the propeller of KVLCC2 in open water is simulated using computational fluid dynamics(CFD)software STAR CCM+,and the flow field information of the KVLCC2 propeller is obtained.Based on the simulated flow field information data,the training sample set was constructed to train PINN.The trained PINN was used to infer the approximate solution of the governing equation at any time and space.Finally,the velocity and pressure distribution obtained by PINN were compared with the velocity and pressure distribution simulated by STAR CCM+.The results validate the reliability of PINN in propeller wake field reconstruction,which can be concluded that PINN can be applied to the reconstruction of the ship propeller wake field.