CNN-Based Ultrasonic Measurements in Vertical Gas-Liquid Two-Phase Flows

This study presents a method that integrates ultrasonic sensing and convolutional neural networks (CNN). The target flows are mainly vertical gas-liquid two-phase flows. This method can be applied to measure various pipe flow systems, such as those used in subsea resource extraction and other industrial applications. The aim of this study is to distinguish different flow patterns and obtain superficial velocities accurately by using pulsed ultrasound.



Some features are extracted from measured ultrasonic pulses reflected from within a pipe. These features are converted to create a visual representation. These visualized ultrasound images are then input into a CNN model. With CNN, classification to distinguish the flow patterns and regression to obtain superficial velocities were conducted. Additionally, the areas where the CNN focuses its attention are identified for each measurement.