Bubble Detection in Multiphase Flows Through Computer Vision and Deep Learning for Applied Modeling

Abstract

An innovative method for bubble detection and characterization in multiphase flows using advanced computer vision and neural network algorithms is introduced. Building on the research group’s previous findings, this study combines high-speed video capture with advanced deep learning techniques to enhance bubble detection accuracy and dynamic analysis. In order to further develop a robust framework for detecting and analyzing bubble properties in multiphase flows, enabling accurate estimation of essential mass transfer parameters, a YOLOv9-based neural network was implemented for bubble segmentation and trajectory analysis, achieving high accuracy. Key contributions include the development of an averaged mass transfer model integrating experimental data, neural network outputs, and scaling algorithms, as well as validation of the proposed methodology through experimental studies, including high-speed video imaging and comparisons with mass transfer coefficients obtained via the sulfite method. By precisely characterizing critical parameters, the algorithm enables accurate gas transfer rate calculations, ensuring optimal conditions in various industrial applications. The neural network-based algorithm serves as a non-invasive platform for detailed characterization of bubble media, demonstrating high accuracy in experimental validation and significantly outperforming traditional techniques. This approach provides a robust tool for real-time monitoring and modeling of bubble flows, laying the foundation for novel, non-invasive methods to measure multiphase media properties.