Development of the AI-assisted thermal hydraulic analysis method for condensing bubbles in vertical subcooled flow boiling

Abstract

Subcooled flow boiling is critical in various industrial applications such as nuclear reactors and thermal management systems. The rapid and complex dynamics of condensing bubbles, from their inception to collapse, pose significant challenges for conventional bubble detection methods. In light of this, a state-of-the-art AI method is developed and validated for the detection and tracking of condensing bubbles in subcooled flow boiling, thereby enabling the effective execution of thermal hydraulic analyses. This study initially employs computer vision technology to efficiently construct a bubble dataset. A bubble detection model, utilizing YOLOv8 with an attention mechanism, is then trained on this dataset. Following successful bubble detection, a multi-object tracking algorithm tracks the bubbles across successive frames. The developed AI-based method has proven highly effective, detecting 95 % of condensation bubbles and streamlining the extraction of key thermal hydraulic parameters, including aspect ratio, Sauter mean diameter, void fraction, interfacial area concentration, departure diameter, growth time, bubble lifetime, Nusselt number, and nucleation site density. The model's accuracy and consistency are demonstrated compared to empirical correlations, affirming its reliability in analyzing the intricate dynamics of subcooled flow boiling. Additionally, it provides detailed fluctuation data on thermal hydraulic parameters. This AI-based method not only improves the reliability and efficiency of monitoring and analyzing subcooled flow boiling but also exemplifies the transformative potential of AI in refining complex industrial processes.