Prediction of flow pattern of gas-liquid flow through circular microchannel using probabilistic neural network

The present study attempts to develop a flow pattern indicator for gas-liquid flow in microchannel with the help of artificial neural network (ANN). Out of many neural networks present in literature, probabilistic neural network (PNN) has been chosen for the present study due to its speed in operation and accuracy in pattern recognition. The inbuilt code in MATLAB R2008a has been used to develop the PNN. During training, superficial velocity of gas and liquid phase, channel diameter, angle of inclination and fluid properties such as density, viscosity and surface tension have been considered as the governing parameters of the flow pattern. Data has been collected from the literature for air-water and nitrogen-water flow through different circular microchannel diameters (0.53, 0.25, 0.100 and 0.050 mm for nitrogen-water and 0.53, 0.22 mm for air-water). For the convenience of the study, the flow patterns available in literature have been classified into six categories namely; bubbly, slug, annular, churn, liquid ring and liquid lump flow. Single PNN model is unable to predict the flow pattern for the whole range (0.53 mm-0.050 mm) of microchannel diameter. That is why two separate PNN models has been developed to predict the flow patterns of gas-liquid flow through different channel diameter, one for diameter ranging from 0.53 mm to 0.22 mm and another for 0.100 mm-0.05 mm. The predicted map and their transition boundaries have been compared with the corresponding experimental data and have been found to be in good agreement. Whereas accuracy in prediction of transition boundary obtained from available analytical models used for conventional channel is less for all diameter of channel as compared to the present work. The percentage accuracy of PNN (similar to 94% for 0.53 mm ID and similar to 73% for 0.100 mm ID channel) has also been found to be higher than the model based on Weber number (similar to 86% for 0.53 mm ID and similar to 36% for 0.05 mm ID channel). (C) 2013 Elsevier B. V. All rights reserved.