基于数值模拟结果聚类的两相流气体形态识别  

Recognizing Gas Shape in Two-phase Flow by Clustering Numerical Simulation Results

作　　者：宋彦坡[1] 陶焰明 彭小奇[1,2] 陈卓[1] 高东波 SONG Yan-Po;TAO Yan-Ming;PENG Xiao-Qi;CHEN Zhuo;GAO Dong-Bo(School of Energy Science and Engineering, Central South University, Changsha 410083, China;Department of Information Science and Engineering, Hunan First Normal University, Changsha 410205, China)

机构地区：[1]中南大学能源科学与工程学院,长沙410083 [2]湖南第一师范学院信息科学与工程系,长沙410205

出　　处：《工程热物理学报》2019年第10期2345-2352,共8页Journal of Engineering Thermophysics

基　　金：国家自然科学基金项目(No.51634010,No.51676211);湖南省重点研发计划(No.2017SK2253)

摘　　要：气体分布形态对气液两相流中的传热、传质及化学反应均有重要影响,为从数值模拟结果中提取该信息,论文构造了一种阈值自适应的凝聚层次聚类算法,并提出了一种基于此聚类算法的气体形态识别方法。针对数值模拟结果数据量巨大的特点,论文还提出了一种"数据空间软分割-子空间分别处理-结果合并"的处理方法。利用二维、三维人造数据及实际数值模拟结果对本文方法的测试结果表明:即使在离散网格尺度不均一、部分网格单元有一定扭曲的条件下,论文方法仍可根据数值模拟结果高效地识别气液两相流中的气体的聚集形态(气泡/气泡群);数据空间软分割方法能有效地降低算法的空间复杂度,便于在大数据条件下的实现气体形态识别。Gas shape has significant influence on kinetic processes such as heat transfer,mass transfer and physiochemical reactions in two-phase flow.To extract this information from mumerical simulation results of two-phase flow,a new hierarchical clustering method using adaptive threshold and a clustering-based gas shape recognition method are proposed.In view of numerical simulation results for two-phase flow are often with large volume,a strategy of "partition softly data space,then processing subspace data respectively,and lastly fusing the process results in all subspaces, is proposed to reduce the storage complexity of gas shape recognition.As testing on 3 datasets(two of them are synthetic and the other is from simulation results) shown,methods proposed in this paper can recognize effectively the gas shape according to the two-phase flow simulation results,even when the discretization scales are not unified or some discretization cells are distorted to some extent,and the strategy of "partition softly data space" can reduce the storage complexity and make gas shape recognition realizable even when the volume of simulation results are huge.

关 键 词：两相流 气泡 聚类 数值模拟 气体形态识别 

分 类 号：O359.1[理学—流体力学] TP391.4[理学—力学]