Gas-liquid flow in small channels: Artificial neural network classifiers for flow regime prediction

Abstract

To design and operate multiphase apparatus with mini- and microchannels, it is important to know how the fluids stream inside. Most literature reports the occurrence of flow regimes in dependence on gas and liquid superficial velocities in flow maps that are valid for fluids with similar properties and channels with similar geometry. Attempts to develop universally applicable flow maps show limitations in the number and variation of considered model parameters or in the number of considered flow regimes. This paper presents artificial neural network classifiers able to predict all relevant flow regimes: (a) Taylor flow, (b) bubbly flow, (c) Taylor-annular flow, (d) churn flow, (e) dispersed flow, (f) annular flow, (g) rivulet flow, and h) parallel flow in dependence on geometric and operational parameters as well as fluid properties with a high precision (R=0.92...0.95 and classification rates were generally above 80%). The classifiers were developed and validated by using more than 13,000 experimental data on gas-liquid flows extracted from 97 flow maps and are based on 7 significant dimensionless groups, namely, ReG, ReL, WeG, WeL,CaL, Θ*, and the channel form factor FC.