[2G14-21] 人工知能技術

[2G18] Integration of AI Technology and Thermal Hydraulics for the Development of a Data-Driven Methodology for Plant Safety Assessment

(3) Robust condensation bubble feature extraction in subcooled flow boiling using object detection and tracking technique

*Wen Zhou¹, Shuichiro Miwa¹, Tomio Okawa² (1. UTokyo, 2. UEC)

キーワード：condensation bubble, subcooled flow boiling, feature extraction, object detection, object tracking, deep learning

Purpose: Conventional research methods (including invasive or non-invasive devices, and conventional object detection methods) are very inefficient for detecting condensation bubbles in subcooled flow boiling. Therefore, the adaptation of a state-of-the-art AI for the robust subcooled flow boiling analysis is developed.
Method: To efficiently detect condensation bubbles, the state-of-the-art AI model is developed based on object detection and object tracking algorithm, and it is coupled with attention mechanisms to further improve the detection and tracking performance of the AI model.
Conclusion: Thermal hydraulic parameters (including void fraction, interfacial area concentration, Sauter bubble diameter, nucleation site density, departure diameter, and departure frequency) were efficiently extracted with the merit of AI and compared with empirical correlations, the comparison results show good agreement.