A study of online monitoring for two‐phase flow patterns in a microchannel based on deep learning

Abstract A low‐cost, low‐noise, high‐frequency microchannel photoelectric detection device based on a photodiode was developed. Based on this device, photoelectric waveforms of gas–liquid and liquid–liquid two‐phase flows in a microchannel were collected, and then online flow pattern detection was realized by using artificial intelligence algorithms. Light intensity distribution characteristics of slug flow were simulated using COMSOL software, and detailed analyses were conducted to examine the relationship between waveform variations and slug flow morphology. After denoising the collected data with wavelet analysis filtering, artificial intelligence algorithms were employed to identify flow patterns for gas–liquid and liquid–liquid two‐phase flows. It was found that the recognition accuracy of the Convolutional Neural Networks algorithm in a variety of fluid systems reaches 100%. Based on this work, a general automatic gas–liquid/liquid–liquid flow pattern recognition platform was developed, which can automatically and quickly identify flow patterns and generate flow pattern graphs from the collected data.