Visualization of 3D behavior of dispersed bubbles using deep learning-based bubble detection technique

Uesawa, Shinichiro  ; Ono, Ayako ; Yoshida, Hiroyuki  

In order to obtain 3D behavior of bubbles, visualization using high-speed video-cameras has been used to identify 3D positions of bubbles. However, it was difficult to apply the technique to bubbly flow with the high void fraction because overlapping bubbles for the sight direction of the camera increased with the increase in the void fraction. JAEA has developed the deep learning-based bubble detector with Shifted window Transformer (Swin Transformer) to overcome the issue for the overlapping bubbles. In this study, we applied the bubble detection technique to images of bubble swarms visualized from two directions other than the direction of main flow and visualized 3D behavior of dispersed bubbles. The result showed that individual bubbles in bubble swarms were detected, and bubble diameters and aspect ratios were measured. Additionally, we obtained 3D positions of bubbles and 3D bubble velocities by linking the bubble positions for the direction of main flow in both images.