Proof of concept of acoustic detection of boiling inception and state transition using deep neural network

We develop a deep neural network model capable of detecting the boiling inception and state transitions from boiling acoustic measurements. We characterize acoustic spectrums of water boiling with different heating surface geometries, heat flux, and degree of subcooling. Our measurement result shows that the feature extraction of the boiling inception and state transition is possible from the boiling-sound frequency dataset in each specific target system. Notably, the deep neural network can distinguish the boiling inception and the state transition more accurately even at the high-level white-noise intensity where human beings and traditional data analysis methods cannot distinguish. This result suggests that the acoustic diagnosis with the deep neural network algorithm has great potential to detect the boiling inception and to monitor the boiling states in quasi-real-time during an early stage of the boiling phenomena, in the nuclear power plants.