【作者】 王畅； 叶舒然； 张珍； 王一伟；

【Author】 WANG Chang;YE Shu-ran;ZHANG Zhen;WANG Yi-wei;Key laboratory for Mechanics in Fluid Solid Coupling System,Institute of Mechanics,Chinese Academy of Sciences;

【机构】 中国科学院力学研究所流固耦合系统力学重点实验室；

【摘要】 随着神经网络等人工智能技术的快速发展,利用数据驱动的机器学习方法在探索复杂流体动力学问题中的流场识别、提取、降阶等方面取得了成功的应用。空化是发生在水力机械等领域常见的水动力现象,作为高速水动力学的核心问题,云状空化流具有强烈的非定常特性,因此,利用数据驱动建立非定常多相流场的识别方法对水力机械等领域具有重要的意义。本文旨在研究一种基于卷积神经网络的空化翼型表面压力预测方法,该数据驱动方法通过提取非定常云空化流场中水翼尾流速度、空泡体积分数等流动特征构建水翼表面压力分布的完整模型。首先,对二维NACA0015水翼云空化流场进行了数值分析,得到了4种不同空化数下翼型速度场、压力场等数据信息。然后,建立了卷积神经网络(CNN)预测模型,通过对不同空化数下水翼云空化尾流速度U、空泡体积分数α及空泡密度ρ等进行提取与分类预测了空化水翼表面的压力系数Cp。最后,将CNN预测结果与CFD计算结果进行对比分析,表明该卷积神经网络的预测方法对非定常多相流场识别具有较高的精度。更多还原

【Abstract】 With the rapid development of artificial intelligence technologies such as neural network,the data-driven machine learning methods have been successfully applied in exploring the flow field identification,extraction,and reduced-order in complex hydrodynamic problems.Cavitation is a common hydrodynamic phenomenon that occurs in the field of hydraulic machinery and other fields.As the core problem of high-speed hydrodynamics,cloud-like cavitation flow has strong unsteady characteristics.Therefore,the data-driven identification method of unsteady multiphase flow fields is of great significance to hydraulic machinery and other fields.This paper aims to study a method for predicting the surface pressure of cavitation airfoil based on convolutional neural network.This data-driven method constructs a complete model of the hydrofoil surface pressure distribution by extracting the flowing features such as hydrofoil wake velocity and vacuole volume fraction in unsteady cloud cavitation flow field.Firstly,the cloud cavitation flow field of two-dimensional NACA0015 hydrofoil is numerically analyzed,and the data information of airfoil velocity field and pressure field under four different cavitation numbers are obtained.Then,a convolution neural network(CNN) prediction model is established,and the surface pressure coefficient Cp of cavitation hydrofoil are predicted by extracting and classifying the hydrofoil wake velocity U,vacuole volume fraction α and vacuole density p under different cavitation numbers.Finally,the comparative analysis of the CNN prediction results and CFD calculation results shows that the prediction method of the convolutional neural network has high accuracy for the identification of unsteady multiphase flow fields.更多还原