Two-Phase Flow Regime Identification Using Fluctuating Force Signals Under Machine Learning Techniques

Abstract

For rational design of industrial machineries such as nuclear power plants and heat exchanging devices, understanding of the two-phase flow regime is crucial. In this study, a new method of flow regime identification using the two-phase fluctuating force signals is proposed. Unlike the existing methodologies to measure two-phase flow parameters, the advantageous feature of utilizing the fluctuating force signal is that the measurement can be conducted under completely intrusive environment. Experiments were conducted using the tri-axial force transducers installed at the 90 degrees pipe bend of the vertical upward flow. For signal classification, machine learning techniques were utilized to identify flow regime, and four types flow regimes, namely, bubbly, slug, churn-turbulent, and annular flows were considered. From the obtained fluctuating force database, the features that characterize the signal were selected in the time and the frequency domain. In the current study, three types of machine learning algorithms such as the artificial neural network (ANN), support vector machine (SVM), and decision tree were examined and results obtained by each learning technique was compared.