基于非监督机器学习方法的竖直环形流道流动沸腾流型研究

Identification of Flow Regime of Boiling Flow in a Vertical Annulus with Unsupervised Machine Learning

研究流动沸腾两相流动形态对封闭反应堆安全分析程序关键本构模型具有重要意义。本文基于非监督机器学习流型识别方法,提出将两相流物理知识融入数据驱动的机器学习模型,并构建输入特征的挑选原则:①机器学习在输入特征中捕捉到的应为流型相关信息;②机器学习的聚类准则应包络该流型下输入特征的所有可能性。依据挑选原则分析电导探针信号生成的汽泡分布特征,确定汽泡弦长累积分布函数数据可用于非监督机器学习流型判断。依据流型识别结果,获得了竖直环形流道内流动沸腾的二维局部流型特性,发现高位局部流型出现在流道中心位置并偏向内加热壁面;并判别了流道截面的全局流型,结果表明流动沸腾泡状流至弹状流的流型转变出现在空泡份额约为0.14位置。

Accurate flow regime identification of the boiling flow is of great significance for using closure correlations in thermal-hydraulic system codes.The flow regime identification is achieved by the unsupervised machine learning(ML)approach,and the current work integrates the data-driven ML model with the two-phase-flow domain knowledge.Two requirements are established to determine whether the feed-in data type is appropriate or not:①the information captured by the unsupervised ML should be regime-relevant features for distinguishing flow regimes;②the cluster criterion for a flow regime should cover all the possible feed-in features of the regime.Feed-in data types generated by the conductivity probe are examined;among them,only the bubble chord length Cumulative Distribution Function(CDF)fulfills the two requirements.With the feed-in data of chord length CDF,the two-dimensional local flow regimes are identified and analyzed for a boiling dataset conducted in an internally heated vertical annulus.The results show that the higher-rank regimes appear at the channel’s center leaning toward the inner-heated wall.The global flow regime map is obtained with the local regimes,and a new flow regime transition criterion between the bubbly and slug flow is developed as the void fraction equals 0.14.