Applicability of time-domain feature extraction methods and artificial intelligence in two-phase flow meters based on gamma-ray absorption technique

Abstract Determining the type of flow pattern and gas volumetric percentage with high precision is one of the vital topics for researchers in this field. For this, in this paper, three different types of liquid–gas two-phase flow regimes, namely annular, stratified, and homogenous were simulated in various gas volumetric percentages ranging from 5% to 90%. Simulations were performed by Monte Carlo N Particle (MCNP) code. Metering system includes one 137Cs sources, one Pyrex glass, and two NaI detectors to register the transmitted photons. Because the signals which are received from the MCNP simulations contain high-frequency noises, the Savitzky-Golay filter has been applied to solve this problem. Then, thirteen characteristics in time domain were extracted from the recorded data of both detectors. Since none of features were capable of completely separating the flow regimes, two methods as “extracting two different features from the recorded data of both detectors” and “extracting three features from the recorded data of both detectors” were proposed. Using these methods, many different separator cases were found and the best separator cases were distinguished via S parameter. Finally, two artificial neural network (ANN) models of multilayer perceptron (MLP) were implemented for each method to identify the flow regimes and approximate the gas volumetric percentages. The proposed methodology and networks could diagnose all flow patterns properly, and also predict the volumetric percentage with a root mean square error (RMSE) of less than 0.60. Increasing the precision of two-phase flow meter by extracting time-domain features and signal processing techniques is the most important advantage of this study.