Experimental study on bubble dynamics in rod bundle sub-channels using enhanced deep learning

Abstract: This study constructed a narrow-spaced rod bundle experimental platform and employed an advanced SF- MR-DST method, which integrates improved Mask R-CNN and DeepSORT algorithms, to systematically in- vestigate the bubble dynamic behavior and void fraction. The experiment focuses on the influence of parameters such as nozzle diameter, flow rate, and shooting height. The results indicate that an increase in flow rate en- hances bubble quantity and morphological complexity, with the maximum nozzle diameter being 0.5 mm. The bubble diameter (1.5–4 mm) shows a positive correlation with flow rate, nozzle size, and height, exhibiting a centralized distribution pattern, with approximately 10-20% of the bubbles displaying irregular shapes. Vertical velocity (0.25-0.37 m/s) increases with higher flow rates while exhibiting an initial deceleration followed by acceleration under the influence of nozzle diameters and height, whereas horizontal velocity remains relatively stable at around 0.2 m/s, compared to 0.4 m/s in unconstrained conditions. The void fraction increases nearly linearly with flow rate, with consistent trends across the three methods despite minor discrepancies. This study provides fundamental data and theoretical insights in bubble dynamics for the initialization and operational optimization of experimental reactors, offering significant guidance for enhancing the operational safety and thermal-hydraulic performance of experimental reactor systems.