Utilizing Artificial Neural Networks and Combined Capacitance-Based Sensors to Predict Void Fraction in Two-Phase Annular Fluids Regardless of Liquid Phase Type

Assessing the void fraction in diverse multiphase flows across industries, including petrochemical, oil, and chemical sectors, is crucial. There are multiple techniques available for this objective. The capacitive sensor has gained significant popularity among these methods and has been extensively utilized. Fluid properties have a substantial impact on the performance of capacitance sensors. Factors such as density, pressure, and temperature can introduce significant errors in void fraction measurements. One approach to address this issue is a meticulous and laborious routine calibration process. In the current study, an artificial
neural network (ANN) was developed to accurately Assess the proportion of gas in a biphasic fluid motion, irrespective of variations in the fluid phase form or variations, eliminating the need for frequent recalibration. To achieve this objective, novel combined capacitance-based sensors were specifically designed. The sensors were simulated by employing the COMSOL Multiphysics application. The simulation encompassed five distinct liquids: oil, diesel fuel, gasoline, crude oil, and water. The input for training a multilayer perceptron network (MLP) came from data gathered through COMSOL Multiphysics, simulations for estimating the
Percentage of gas content of an annular two-phase fluid with a specific liquid form. The MATLAB software was utilized to construct and model the proposed neural network. The utilization of the novel and precise apparatus for measuring the intended MLP model demonstrated the ability to prognosticate the volume percentage with a mean absolute error (MAE) of 0.004.