Visual identification of oscillatory two-phase flow with complex flow patterns

• Published in: Measurement : journal of the International Measurement Confederation 2021-12, Vol.186, p.110148, Article 110148
• Main Authors: Huang, Yuqi, Li, Dominique H., Niu, Haoyi, Conte, Donatello
• Format: Article
• Language: English
• Subjects: Bayesian analysis; Cognitive science; Computational fluid dynamics; Computer science; Computer vision; Computer Vision and Pattern Recognition; Cooling; Cooling rate; Data Structures and Algorithms; Diesel engines; Distributed, Parallel, and Cluster Computing; Feature extraction; Flow distribution; Fluid dynamics; Fluid flow; Heat transfer; Identification methods; Image Processing; Information Retrieval; Low speed; Machine learning; Multiphase flow; Offset vector and sequence; Optical flow; Optical flow (image analysis); Pattern identification; Reynolds number; Reynolds number classification; Signal and Image Processing; Simulation; Two phase flow; Visualization
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2021.110148

We present an approach based on computer vision and machine learning methods to identify two-phase flow with complex flow patterns in oscillatory conditions. A visualization experiment bench was designed, constructed, and used to simulate the actual reciprocating motion of the cooling gallery inside the piston of low-speed diesel engines. The results of our proposed approach show that the feature vectors extracted from the optical flow images provides a valuable reference for the velocity vectors in two-phase flow. We show that it is possible to identify oscillatory two-phase flow videos with respect to Reynolds numbers from 10568 to 31704 using a Bayesian Network classifier, with the best accuracy of 94%. The approach purposed in this paper can not only be used to present the validating sources for numerical simulation results, but also be widely applied in the visualization of multiphase flow, which is a key area to be developed on the basic research of heat transfer systems. •First approach to identify visualized two-phase flow by machine learning methods.•Use high-speed camera to capture complex oscillatory two-phase flow patterns.•Use optical flow-based feature selection to characterize two-phase flow patterns.•The accuracy of video-based Reynolds number classification reaches 94%.•Verify CFD simulation results with the vector extracted by our method.