Deep learning-assisted characterization of bubble dynamics and heat transfer in subcooled flow boiling of seawater

Subcooled flow boiling of seawater is a complex process critical to marine-based thermal systems, yet the influence of dissolved salts on bubble dynamics remains ambiguous, with reported effects on heat transfer being contradictory. This study investigates the complex bubble dynamics and heat transfer characteristics of subcooled flow boiling in seawater, utilizing a deep learning-assisted computer vision approach. Subcooled boiling experiments were conducted on a zirconium alloy surface using artificial seawater with salt concentrations of 1% and 3%. To overcome the challenges of overlapping visual data in dense boiling regimes, a U-Net pipeline was deployed to accurately identify and segment bubble boundaries, enabling precise extraction of morphological parameters. The results indicate that increasing seawater salinity shifts the boiling curve upward, achieving higher heat transfer rates at lower Jakob numbers due to micro-cavities formed by salt deposition, which increase the active nucleation site density. Furthermore, evaporation at the bubble base induces localized concentration gradients, triggering Marangoni forces that pin bubbles to the heated surface, causing the Sauter Mean Diameter to increase alongside superheat and salinity. Statistical analyses reveal that bubble aspect ratios follow a log-normal distribution, while perimeters adhere to a Weibull distribution. Finally, new predictive correlations for bubble aspect ratio and Sauter Mean Diameter are proposed, significantly improving prediction accuracy for seawater boiling environments.