Research on Computation Method of Bubble Volume Based on Machine Learning

Abstract In the case of low-speed jet flow and high undercooling degree, the condensation process in the reactor suppression system is often accompanied by low-frequency oscillations, namely the chugging phenomenon. The mechanism of oscillation is related to the formation, condensation, and collapse of bubbles. The accuracy of extracting bubble geometry parameters directly impacts the analysis of condensation oscillation mechanisms. The calculation of geometric parameters of bubbles using traditional image processing techniques often involves complex pre-processing, and the feature parameters need to be defined manually. There are some issues, such as significant user effects and long computation time. Therefore, based on traditional methods, this paper proposes a more efficient algorithm for extracting image feature parameters using convolutional neural networks (CNN). The results show that the performance of the newly proposed method under a single condition is satisfactory, and its generalization performance across different backgrounds is also good. The average additional error of extracted feature parameters is less than 2%, but the extraction cost can be significantly reduced.