Entropy Analysis of Two Phase Flow Pattern Identification in Vertical Tube

Abstract

This paper aims to analyze the time domain marginal spectral entropy of differential pressure signals collected from the gas-liquid two-phase flow pattern vertical tube bundle channel on the experimental platform. By comparing the visualized and measured differential pressure signals, four flow patterns were found, namely, bubble flow, bubble-mixed flow, agitated flow, and annular flow. The marginal spectral entropy values of the four flow patterns are calculated, and it is found that the spectral entropy distribution regions of the different flow patterns are different. Putting the aforementioned four flow patterns that is regarded as the feature quantity into artificial neural network support vector machine for flow pattern recognition. The experimental results show that the marginal spectral entropy can better reflect the flow difference between different flow patterns. The support vector machine has the advantages of fast calculation speed and high recognition rate. The overall recognition rate of the four flow patterns (the transition flow is included) is as high as 95.31% by combing the marginal spectral entropy with the support vector machine, which is suitable for online identification of flow patterns.