Void fraction detection technology of gas-liquid two-phase bubbly flow based on convolutional neural network

• Published in: Experimental thermal and fluid science 2023-04, Vol.142, p.110804, Article 110804
• Main Authors: Han, Bangbang, Ge, Bin, Wang, Fan, Gao, Qixin, Li, Zhixuan, Fang, Lide
• Format: Article
• Language: English
• Subjects: Bubbly flow; Convolutional neural network; Void fraction
• ISSN: 0894-1777; 1879-2286
• DOI: 10.1016/j.expthermflusci.2022.110804

•A detection method based on Faster R-CNN for real-time bubble recognition and feature extraction, is developed.•Void fraction of gas–liquid two-phase bubbly flow is studied.•The uncertainty of the method is 10 % and average error of this method is 7.07 %, which has high accuracy. Gas-liquid two-phase bubbly flow widely exists in the field of natural gas extraction. The identification and feature extraction of bubbles are becoming more and more important to the pipe system. This study develops a detection method based on Faster R-CNN for real-time bubble recognition, feature extraction, and void fraction calculation. The method detects ellipsoidal large bubbles in gas–liquid two-phase bubbly flow in a vertical closed pipeline with a high void fraction. The uncertainty of the method is 10 %. Then compared to the gas volume holdup from standard flowmeter, average error of this method is 7.07 %, which has high accuracy. This new detection method finds a new way for feature extraction of two-phase flow.