A deep learning-based algorithm for rapid tracking and monitoring of gas–liquid two-phase bubbly flow bubbles

Abstract

Gas–liquid two-phase bubbly flow has significant applications across multiple fields, including reactor design and separation processes in chemical engineering, oil well extraction and pipeline transportation in the oil and gas industry, cooling systems in the nuclear industry, and wastewater treatment in environmental engineering. Bubble monitoring is crucial in these applications as it can enhance mass and heat transfer efficiency, improve flow stability, and ensure the safe operation of systems. This study developed an advanced algorithm aimed at precisely detecting and segmenting small bubbles at the gas–liquid interface using semantic segmentation techniques. This technology leverages deep learning models to analyze images, automatically identifying bubbles at the gas–liquid interface and accurately delineating their boundaries. The technique provides precise contours for each bubble, offering essential foundational data for further bubble dynamics analysis. Building on this, the deep learning detection algorithm was combined with the Deep Simple Online and Realtime Tracking with a Deep Association Metric (DeepSORT) algorithm, tracking algorithm, enabling the system to rapidly and accurately identify and track the movement of the same bubble across consecutive frames.