BF-GAN: Development of an AI-driven bubbly flow image generation model using generative adversarial networks

Acquiring a large number of high-quality two-phase flow images remains a time-consuming and expensive task for developing advanced models in fluid dynamics. To address these issues, a generative AI architecture called bubbly flow generative adversarial networks (BF-GAN) is developed, designed to generate realistic and high-quality bubbly flow images through physically conditioned inputs, namely superficial gas ( j g ) and liquid ( j f ) velocities. Firstly, 105 sets of two-phase flow experiments under varying conditions are conducted to collect 278,000 bubbly flow images with physical labels of j g and j f as training data. A multi-scale loss function of GAN is then developed, incorporating mismatch loss and feature loss to further enhance the generative performance of BF-GAN. The generative performance of the BF-GAN demonstrates a 39.6% improvement over conventional GAN in all generative AI indicators, establishing for the first time a quantitative benchmark in the domain of bubbly flow. In terms of image correspondence, the luminance, contrast, magnitude, homogeneity, and correlation of BF-GAN generated images are compared with experimental images, showing errors between 2.22% and 24.74%. For two-phase flow parameters, including void fraction, aspect ratio, Sauter mean diameter, and interfacial area concentration, the BF-GAN generated images are compared with experimental data, with errors ranging from 2.3% to 16.6%. The comparative analysis demonstrates that the BF-GAN is capable of generating realistic and high-quality bubbly flow images for any given j g and j f within the research scope, and these images align with physical properties.