Detecting non-uniform structures in oil-in-water bubbly flow experiments.

In this work, we first design a series of oil bubbly flow experiments in a vertical testing pipe, and collected the fluid fluctuations as experimental observations. Then we establish a Variational Autoencoder-Generative Adversarial Network (VAE-GAN) [1] from 4272 fluid time-frequency features, which is used to detect anomalous fluctuations in the experimental oil bubbly flow signals. These anomalous fluctuations are then utilized to experimentally identify the non-uniform flow structures in oil-in-water bubbly flow. In addition, by analyzing the identified non-uniform flow structures in a vertical 20 mm inner diameter pipe, we investigate the oil bubbles coalescence phenomenon under various flow conditions. The impacts of mixed fluid velocity and phase volume fraction on the bubble coalescence behaviors are also investigated. The proposed method not only serves as an efficient tool for flow anomaly detection, but also provides a novel way to characterize the evolutionary behaviors of diverse bubbly flow systems. [ABSTRACT FROM AUTHOR]