Presentation Information
[3C01]Development of Validation Technology for Detailed Two-Phase Flow Simulation Codes to Optimize Innovative Reactor Design(7) Development of Bubble Detection Technique for Rod Bundle Flow Channel Using Deep Learning
*Shinichiro Uesawa1, Natsuki Hiramatsu2, Koji Ono2, Naoki Horiguchi1, Hiroyuki Yoshida1 (1. JAEA, 2. HST)
Keywords:
Visualization,Bubble detection,Deep learning,Rod bundle channel
To realize the early practical application of innovative reactors, it is essential to utilize numerical simulations as alternatives to large-scale mock-up tests. To clarify the validity, we are developing measurement techniques capable of capturing instantaneous and local gas-liquid interface information. This presentation introduces a deep-learning-based bubble detection technique capable of measuring bubble diameters and other parameters from visualization results in rod bundle flow. To evaluate systematic errors, we used bubble images generated from numerical simulation results by using ray tracing. Comparing the data obtained from these images by using the bubble detection technique with the data directly extracted from the numerical simulation results, we confirmed that the technique is capable of accurately measuring the data for spherical and ellipsoidal bubbles.