The application of ultrasonic measurement and machine learning technique to identify flow regime in a bubble column reactor

Abstract

This paper presents a novel technique to classify the flow regimes in bubble columns. The ultrasonic velocity profiler is employed to detect the velocity deviation and echo characteristic of bubbles rising in the column. This information is set as attribute data for the machine learning algorithm. Classification-based machine learning is utilized to classify the flow regimes: bubbly, transition, and churn turbulent, which are defined as categories of the algorithm. Several classifiers were applied in this work, such as K-nearest neighbors, Decision tree, Support vector machines, Naive bayes, and Logical regression. The experimental demonstration was conducted to verify the performance of the proposed technique. Three kinds of two-phase flow with stagnant liquid that had various viscosities were used for the experiment. The air within the superficial velocity range was injected to alter the flow regime. The flow regime classification model was set. The proposed method was applicable to identify the flow regimes. The classifiers were tested, and their accuracy was evaluated.