基于机器学习的二维流体VOF方法
A Machine Learning-based VOF Method for 2D Flowfield
孟巍 1梅桂林1
作者信息
- 1. 哈尔滨工程大学船舶工程学院,哈尔滨 150001
- 折叠
摘要
以网格顶点和网格中心的流体体积分数为输入,利用深度学习神经网络(DNN)进行两相界面的快速几何重构,改进了一种几何VOF方法——isoAdvector算法.在原始isoAdvector算法生成训练集并进行预处理的基础上,通过神经网络隐藏层中神经元个数和激活函数对预测结果的影响来确定适当的神经网络;并根据质量守恒修正神经网络预测结果,完成两相界面的几何重构.在OpenFOAM框架下,嵌入神经网络修正改进后的isoAdvector算法.二维溃坝数值模拟结果表明:改进算法与原始算法计算精度相当,计算效率有所提升.
Abstract
The Deep Neural Network(DNN)is used to improve the Volume-of-Fluid(VOF)method in isoAdvector algorithm.Based on the Volume fraction in cell and at nodes,the two-phase interface can be reconstructed with high efficiency.The training set is generated and preprocessed by the original isoAdvector algorithm,which will reduce the difficulty of neural network training with high degree of accuracy.According to the volume conservation,the predicted data of the neural network are analytically corrected to complete the geometric reconstruction of the two-phase interface.The improved isoAdvector algorithm is used to simulate the 2D dam break by using OpenFOAM.The accuracy of the improved method is equivalent to the original one,and the computational efficiency is improved.
关键词
自由液面捕捉/几何VOF方法/深度神经网络/isoAdvector算法/OpenFOAMKey words
free surface capture/geometric VOF/deep neural network/isoAdvector/OpenFOAM引用本文复制引用
出版年
2024
NETL
NSTL
万方数据