Using statistical features and a neural network to predict gas volume fractions independent of flow regime changes

Abstract

This work presents a methodology based on nuclear techniques and Artificial Neural Networks (ANNs) for Gas Volume Fraction (GVF) predictions in two-phase flow independent of flow regime changes. Using gamma-ray principles attenuation with an appropriate geometry, consisting of a single detector and a single-energy gamma-ray source, it is possible to obtain data which could be adequately correlated to the GVF using neural network. The novelty of this study, in contrast to previous research, lies in the utilization of novel statistical features extracted from the discrete wavelet transform applied to the gamma-ray spectrum. Instead of analyzing the entire gamma-ray spectrum, this approach focuses on specific features, thereby mitigating unwanted noise and enhancing the efficiency of predictive capabilities in real conditions. The two-phase flow loop is used to provide the training and testing data for the network. The GVF was predicted using extracted feature from fourth stage of discrete wavelet transform precisely with mean absolute error less than 0.009.