Multi-modality fusion: Illuminating the complexity of two-phase flow

Bubble dynamic analysis is crucial for modeling and optimizing phase change heat transfer. Recent advances in computer vision models have been widely applied to automatically extract bubble behavior parameters related to their performances. However, current segmentation models heavily rely on large datasets to achieve high accuracy, imposing a significant human workload. To overcome this challenge, we propose a novel data-fusion approach that enriches input information with dynamic features of bubble motion. Our methodology incorporates multi-modal generation techniques to transform conventional high-speed images into complementary modalities, and then fuse them for enhanced representation. This framework enables precise segmentation prediction using small datasets and without any pre-trained models. The resulting bubble statistics provide a quantitative foundation for advancing boiling heat transfer analysis and elucidating interfacial transport mechanisms. Overall, our multi-modal segmentation framework offers an automated, data-efficient, accurate, and versatile method for bubble behavior quantification in diverse two-phase flow systems.