Artificial Neural Network to Determine Dynamic Effect in Capillary Pressure Relationship for Two-Phase Flow in Porous Media with Micro-Heterogeneities

Abstract

An artificial neural network (ANN) is presented for computing a parameter of dynamic two-phase flow in porous media with water as wetting phase, namely, dynamic coefficient (τ), by considering micro-heterogeneity in porous media as a key parameter. τ quantifies the dependence of time derivative of water saturation on the capillary pressures and indicates the rates at which a two-phase flow system may reach flow equilibrium. Therefore, τ is of importance in the study of dynamic two-phase flow in porous media. An attempt has been made in this work to reduce computational and experimental effort by developing and applying an ANN which can predict the dynamic coefficient through the "learning" from available data. The data employed for testing and training the ANN have been obtained from computational flow physics-based studies. Six input parameters have been used for the training, performance testing and validation of the ANN which include water saturation, intensity of heterogeneity, average permeability depending on this intensity, fluid density ratio, fluid viscosity ratio and temperature. It is found that a 15 neuron, single hidden layer ANN can characterize the relationship between media heterogeneity and dynamic coefficient and it ensures a reliable prediction of the dynamic coefficient as a function of water saturation.

Review of Data Science Trends and Issues in Porous Media Research With a Focus on Image‐Based Techniques

“…The experiments was extended further by the analysis of the dynamic coefficient of heterogeneous porous media by Das et al. (2015), where the correlation of six independent variables was studied in contrast to five variables in Hanspal et al. (2012).…”

“…Hanspal et al (2012) used water saturation, viscosity ratio, density ratio, absolute permeability, and temperature of the experiment as the inputs for different ANN architectures and reached a reasonable agreement with the experimental data. The experiments was extended further by the analysis of the dynamic coefficient of heterogeneous porous media by Das et al (2015), where the correlation of six independent variables was studied in contrast to five variables in Hanspal et al (2012). A single-layered ANN was employed with a varied number of neurons, where 15 neurons yielded the best acceptable results in accordance with the target values.…”

Time-resolved synchrotron X-ray micro-tomography datasets of drainage and imbibition in carbonate rocks

“…These datasets were collected with the aim of investigating pore-scale processes during immiscible fluid displacement under a capillary-dominated flow regime 22 . The time-dependent information can be used to validate models of pore-scale displacement, such as direct simulations 25–27 , pore-network 27 , 28 and neural network models 29–31 . Furthermore, the data can be used to quantify how the balance of viscous and capillary forces control the exact nature of trapping, and to further analyze the complex pore-scale processes during immiscible fluid flow in permeable media.…”

Artificial Neural Network (ANN) For Evaluating Permeability Decline in Permeable Reactive Barrier (PRB)

“…architectures were developed to determine dynamic capillary pressure effects in two-phase flow in heterogeneous porous domains which used data from computational flow physics-based studies (Das et al 2015). the input data can continuously be updated creating a larger database for better training and validation of the ANN structure, hence, enabling better predictions.…”