Development of artificial neural networks for the prediction of the pressure field along a horizontal pipe conveying high-viscosity two-phase flow

Abstract

Pressure fluctuations are one of the primary concerns regarding the safety of two-phase flows in pipelines. Therefore, it is crucial to closely monitor pressure variations and predict their spatiotemporal behavior to the greatest extent possible. In this article, we present the development of a set of models based on artificial neural networks (ANN) to predict the time-dependent behavior of the pressures produced by high-viscosity gas-liquid flows at predetermined locations along a horizontal pipeline. Different volume fractions of glycerin and air constitute the mixtures of interest. The models use the measured values of the mass-flow rates of both fluids at the pipe’s inlet, together with the pressures measured at other locations further downstream. In order to determine the optimal architecture for the artificial neural network, we tested the following four transfer functions for the hidden layer: logsig, tansig, softplus, and swish. Additionally, we utilized the linear function purelin for the output layer. We employed the Levenberg–Marquardt algorithm to train the ANN models and the experimental data set to test their performance.