Investigation of Time-Domain Feature Selection and GMDH Neural Network Application for Determination of Volume Percentages in X-Ray-Based Two-Phase Flow Meters

Abstract

Nowadays, there is a high demand to use radiation-based equipment to increase the precision of oil products monitoring systems. In this study, the application of X-ray tubes combined with feature extraction and artificial intelligence techniques were investigated for determining the volumetric percentages in two-phase flows. Firstly, a detection system consists of an X-ray tube, a horizontal Pyrex-glass pipe, and a NaI(TI) detector was simulated by MCNP code. Two flow regimes of annular and stratified were simulated in different volume percentages in the pipe. After acquiring the needed dataset using MCNP code, ten time-domain features were extracted from the dataset and then applied as the input of the neural network model of group method of data handling (GMDH). The GMDH neural network has the ability to diagnose the efficient input for achieving the best network configuration. Among extracted time characteristics of RMS, WL, ASS, MSR, ASM, average, STD, median, skewness, and kurtosis, GMDH neural network introduced the characteristics of RMS, ASS, MSR, ASM, and average as effective characteristics. Indeed, this GMDH neural network feature helped to determine effective characteristics for predicting volume percentages. The obtained predictions with a maximum root-mean-square error of 0.62 indicate that the use of the feature extraction methods is very applicable in determining volume percentages.