Utilizing Features Extracted from Registered 60Co Gamma-Ray Spectrum in One Detector as Inputs of Artificial Neural Network for Independent Flow Regime Void Fraction Prediction

Abstract

In this paper, we demonstrate that void fraction could be predicted independent of type of flow regime in two-phase flows using 60Co source and one scintillator NaI detector. For this purpose, firstly three features (Feature No. 1: counts under Compton continuum; Feature No. 2: counts under full energy peak of 1173 keV; Feature No. 3: counts under full energy peak of 1333 keV) were extracted from registered gamma-ray spectrum in detector. Secondly, these three features were utilized as the inputs of artificial neural network model of multilayer perceptron (MLP) in order to achieve the best structure for predicting the void fraction. In each structure, void fraction was considered constantly as the output of MLP network. Using the optimum MLP network structure, void fraction was predicted independent of type of flow regime in gas–liquid two-phase flow with MRE of less than 2.5%. Although obtained error using one detector for predicting the void fraction is more than when two or more detectors are utilized, using fewer detectors has advantages such as making the detection system simpler and reducing economical expenses.