[3J01-05] Anomaly Detection Technique and Robot

[3J02] Investigation on anomaly detection technique for cooling system device of sodium fast reactor by acoustic method

(2) Acoustic recognition of gas release in liquid based on its acoustic characteristics and deep learning

*Nao Mikami¹, Yoshitaka Ueki¹, Masahiko Shibahara¹, Kosuke Aizawa² (1. Osaka Univ., 2. JAEA)

Keywords:Gas-liquid two-phase flow, deep learning, acoustic recognition

To increase the safety of sodium-cooled fast reactors (SFRs), it is important to perform the early-stage detection and state sensing systems of two-phase flow with sodium-water reaction caused by heat transfer tubes damaging. This study focuses on the acoustic method with a short-time response and aims to develop a novel method to sense the states of two-phase flow with high accuracy based on convolutional neural networks (CNNs). We firstly perform experiments on the acquisition of pipe flow sound and two-phase flow sound, each of which simulates normal and anomaly sounds. Second, we extract acoustic features using time-frequency analysis and evaluate the feasibility of our method based on the classification accuracy of CNNs.