Thermal field predictions and boiling curve reconstruction from bubble dynamics using conditional generative adversarial networks

Abstract

Listen

Translate article

Although the accurate prediction of boiling phenomena is crucial for preventing equipment damage and ensuring energy system safety, it remains challenging owing to the intertwined thermal–hydraulic interactions. We predicted the temperature field of boiling surfaces from the visualization results of bubble dynamics using conditional Generative Adversarial Networks (cGANs). The network was trained to generate infrared images from side-view images in a flow boiling experiment, and its rationale is supported by intermediate activation maps. Additionally, the model was applied to a data-deficient scenario based on pool boiling to derive heat flux components and reconstruct the boiling curve. This approach enhances the insight into both the visible and invisible aspects of boiling phenomena, which is particularly advantageous when direct and elaborate data acquisition faces facility limitations.