Estimation of Void Fraction by Artificial Neural Network for behavior observation of gas-liquid two-phase flow

Abstract

In many industrial processes, gas-liquid two-phase flow occurs, and it is necessary to grasp the behavior in the pipe for safety design. The void fraction is one of the main parameters for gas-liquid flow behavior observation. In this research, the impedance difference between gas and liquid is measured by the Voltage-Current (VC) system. The void fraction is estimated by a neural network (ANN) with VC system as input data acquisition method. By extracting the feature quantities of the measurement data set using ANN, we aimed at the higher accuracy measurement that is not influenced by the flow regime. In the experiments, the ANN model specialized for each flow regime were created and evaluated. We also compared the accuracy of the model learned without considering the different flow regimes. As a result, it was found that performing predetermination of flow regime is valid to improve the estimation accuracy.