Micro-Scale Flow Pattern Classification Based on the K-Means Clustering Algorithm

In the present work, an objective method to characterize two-phase flow pattern was developed and implemented. The method is based on the characteristics of the signals provided by transducers measuring local temperature and pressure plus the intensity of a laser beam crossing the two-phase flow. The statistical characteristics of these signals were used as input features for the k-means clustering method. In order to implement the method, experimental flow patterns were obtained during flow boiling of R245fa in a 2.32 mm ID tube. Experiments were performed for mass velocities from 100 to 700kg/m²s, saturation temperature of 31 °C and vapor qualities up to 0.99. The cluster classification was compared against flow patterns segregated based on high speed camera images (8000 images/s) and a reasonable agreement was obtained.