TY - JOUR

T1 - Neural nerwork approach for plug flow analysis in microchannels

AU - Seredkin, Alexander V.

AU - Yagodnitsyna, Anna A.

N1 - Funding Information: This research was funded by the Russian Science Foundation (Grant No. 21-79-10307). Publisher Copyright: © 2022 by Begell House, Inc.

PY - 2022

Y1 - 2022

N2 - Liquid–liquid and gas–liquid flows in microchannels are widely utilized in various technological fields. The plug/droplet flow regime is preferable in many applications. The features of plugs and droplets, such as the length, vol-ume, and velocity, are critical parameters when developing new microchannel devices. The general approach employed to define plug features is based on image processing algorithms, in which the spatial filters are used in edge detec-tion. This approach’s main drawback consists of manually adjusting the parameters, such as the filter type, threshold, background removal procedure, etc. Here, we present a neural network approach for plug/droplet detection. A comprehensive data set for neural network training was compiled. The results of the neural network training are discussed, and a comparison with the image processing algorithm is provided. The proposed method has shown consistent numerical measurements. The average deviations of the measured plug size and velocity did not exceed 1.71% and 0.91%, respectively. New data on the plug size and velocity for an extremely low-viscosity ratio of the phases have been ob-tained.

AB - Liquid–liquid and gas–liquid flows in microchannels are widely utilized in various technological fields. The plug/droplet flow regime is preferable in many applications. The features of plugs and droplets, such as the length, vol-ume, and velocity, are critical parameters when developing new microchannel devices. The general approach employed to define plug features is based on image processing algorithms, in which the spatial filters are used in edge detec-tion. This approach’s main drawback consists of manually adjusting the parameters, such as the filter type, threshold, background removal procedure, etc. Here, we present a neural network approach for plug/droplet detection. A comprehensive data set for neural network training was compiled. The results of the neural network training are discussed, and a comparison with the image processing algorithm is provided. The proposed method has shown consistent numerical measurements. The average deviations of the measured plug size and velocity did not exceed 1.71% and 0.91%, respectively. New data on the plug size and velocity for an extremely low-viscosity ratio of the phases have been ob-tained.

KW - detection

KW - microchannel

KW - neural network

KW - plug flow

KW - two-phase flow

UR - http://www.scopus.com/inward/record.url?scp=85134852917&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/a86a8452-8d87-358e-b1b3-00ccaa73bf9f/

U2 - 10.1615/InterfacPhenomHeatTransfer.2022043493

DO - 10.1615/InterfacPhenomHeatTransfer.2022043493

M3 - Article

AN - SCOPUS:85134852917

VL - 10

SP - 15

EP - 24

JO - Interfacial Phenomena and Heat Transfer

JF - Interfacial Phenomena and Heat Transfer

SN - 2169-2785

IS - 1

ER -